Young Researchers Seminar 2021 Seminar proceedings 15 – 17 September 202 1 Portoroz, Slovenia European Conference of Transport Research Institutes (ECTRI) Forum of European National Highway Research Laboratories (FEHRL) Forum of European Road Safety Research Institutes (FERSI) And University of Ljubljana, Faculty of maritime studies and transport 10th YOUNG RESEARCHERS SEMINAR YRS 2021 SEMINAR PROCEEDINGS 15. – 17. September 2021 Portoroz, Slovenia STEERING COMMITTEE Evangelos Bekiaris Claude Marin-Lamellet Jindrich Fric Ciaran McNally Felipe Jiménez Anne Silla REVIEWERS Jiří Ambros Roman Ličbinský Afroditi Anagnostopoulou Blaž Luin Fabienne Anfosso Lédée Heike Martensen Nicolas Bueche Irina Mazilu-Eyaz Laurent Carnis Thomas O'Brien Ricardo De Castro Enrique Onieva Evangelia Gaitanidou Conrad Piasecki Marie-Axelle Granié Agnieszka Stelling Libor Krejčí Sandra Vieira Gomes ORGANIZING COMMITTEE Ana Pererira Patricija Bajec Elen Twrdy Marina Zanne Kataložni zapis o publikaciji (CIP) pripravili v Narodni in univerzitetni knjižnici v Ljubljani COBISS.SI-ID 74498819 ISBN 978-961-7041-09-5 (PDF) The papers are peer-reviewed by international experts Publisher: Faculty of Maritime Studies and Transport, Portoroz, 2021 Editors: Marina Zanne, Patricija Bajec, Elen Twrdy Electronic publication: https:// © 2021 by FPP Portorož INDEX (in alphabetical order by the family name of the first author) Alessandra Boggio-Marzet, Andres Monzon 1 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE STUDY IN THE MADRID REGION Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos 12 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE SYSTEM TESTING Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 18 CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis 25 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND ACCESSIBILITY POINT OF VIEW Rins de Zwart 39 IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Capucine-Marin Dubroca-Voisin 46 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN STATIONS Olga Maria Giannakari 56 ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Blanka Hablovičová, Vítězslav Křivánek, Petra Marková 66 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH REPUBLIC Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg, Nils Fearnley, Rikke Ingebrigtsen 75 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Doron Hekič, Andrej Anžlin, Aleš Žnidarič 85 A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Jens Hellekes 93 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Mehdi Kalantari 100 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Kas Kamphuis 112 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S DISEASE Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 119 DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 126 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Barbara Likar, Darko Kokot, Stanislav Lenart 135 TESTING OF DYNAMIC CHARACTERISTICS OF POROELASTIC ROAD SURFACE MATERIAL i Anne-Farina Lohrengel, Lennart Meine 141 THE CLIMATIC IMPACT ON GERMANYŚ TRANSPORT SYSTEM - INCREASING ROAD INFRASTRUCTURE RESILIENCE TO NATURAL HAZARDS Christian Mathea 149 DYNAMIC MEASUREMENT SYSTEMS – USE OF FAST DRIVING MEASURING VEHICLES FOR THE NETWORK-WIDE RECORDING OF NIGHT VISIBILITY OF ROAD MARKINGS IN DRY CONDITIONS Mahdi Rahimi Nahoujy 156 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND THE DESIGN OF PAVEMENTS (MESAS) DATA Tibor Petrov 163 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR NETWORKS Eva Pitlová, Lucie Vyskočilová 169 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY OF BŘECLAV Leon Straßgütl 178 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Tamás Strommer 188 EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Daniel Szabó 200 A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Teun Uijtdewilligen 216 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez 222 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN UNSTRUCTURED ENVIRONMENTS Kartik Varma, François Combes, Pierre Eykerman 232 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD OBSERVATIONS Cyril Veve, Nicolas Chiabaut 245 DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Laura Vovčko, Stanislav Lenart 254 CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Panagiotis Zeibekis, Melina Karagkouni, Dimitrios Dimitriou, Maria Sartzetaki 262 VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS SOCIOECONOMIC FOOTPRINT ii Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE STUDY IN THE MADRID REGION Alessandra Boggio-Marzet, Researcher, PhD Student Andres Monzon, Full Professor, PhD Supervisor TRANSyT-Transport Research Centre, Universidad Politécnica de Madrid C/Profesor Aranguren nº3, 28040 Madrid, Spain alessandra.boggiomarzet@upm.es ABSTRACT Cities are undergoing a process of urbanisation and urban freight transport had to deal with the boom of e-commerce, which is accelerated by COVID-19 pandemic and produces a sharp increase in freight activities, bringing additional challenges to city logistics. Furthermore, the lack of real data concerning logistics companies makes the research of efficient solutions for urban logistics more complicated. Based on results from a real case study, this paper identifies the main (in)efficiencies of last-mile postal deliveries(LMD) under real traffic conditions, in terms of both energy consumption and operative features. During one month, thirteen drivers of the same logistic company drove light duty vehicles for their normal deliveries. Meanwhile, GPS position, speed and other parameters were instantaneously recorded:242 delivery routes were alnalyzed in the city center and the peri-urban area of Madrid, corresponding to 7,262km travelled. After each delivery route drivers fulfilled a short questionnaire and leave a feedback to report any incidents encountered while delivering. According to our results, clear inefficiencies dominate the current delivery performance: a large proportion of deliveries should be shifted to non-motorized transport modes especially in city center, where 40% of LMD is characterized by average speed lower than 10km/h. This aspect, coupled with the long waiting times necessary for delivering, the lack of parking space and traffic congestion, and the low percentage of first right time delivery reaching only 75%, emphasizes the urgency of implementing sustainable strategies of LMD (parcel lockers, UCC, cargo-bikes among others) to improve efficiency and reduce their negative environmental impact. Keywords: Last-mile delivery, real data, operative features, energy inefficiency 1. ENERGY EFFICIENCY IN impact on climate change, fragmentation and so on TRANSPORT: AN UNFILLED GOAL (Ecotale, 2014). The great transformation we are experiencing within According to the European Environment Agency, the current process of globalisation causes an transport sector causes a quarter of global increase in the mobility of people and goods at an Greenhouse Gas (GHG) emissions in the European astonishing rate. Indeed, in recent decades, different Union (EU) and contributes significantly to the lifestyles and the development of new services have climate change (EEA, 2019). While other economic changed our habits and our mobility, and, as a sectors, such as households and industry, have consequence, have influenced the liveability of our drastically reduced their emissions compared to the cities, town centres and suburbs (Ministero delle 1990 level (Kyoto Protocol), transport emissions Infrastrutture e dei Trasporti, 2020). have increased, being 29% above 1990 levels in 2018. Moreover, within the sector, road transport is As known, the transport sector enables the the main culprit, with higher growth than all other movement of people, animals and goods from one modes, both for passengers and freights (Eurostat place to another. Therefore, if on the one hand it is 2020). of foundamental importance for the prosperity of a country and its citizens, on the other hand it gives In Spain, transport sector is responsible for 24% of rise to side effects that can become costs when they greenhouse gas emissions (in particular CO2), and relate to congestion, accidents, air pollution, noise, the Country is facing a major challenge to ensure its 1 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia compliance with its international commitments ignition, acceleration and braking (Garcia-Castro et (IDAE, 2018), aware that there is still a long way to al., 2018; Lois et al., 2019). go. The dominant mode is still road, and the effect 1.1. Transportation in changing society of all the improvements - vehicles and fuels - have been offset by growth in demand and journeys: more The World Social Report 2020 affirms that for the trips and over longer distances have more than offset first time in history, more people live now in urban the limited improvements. than in rural areas (United Nations, 2020). More and In this context, the EU has launched in 2019 the so- more, indeed, there is a displacement trend of called "Green Deal". It proposes a Climate Pact, population from the countryside to the city. Today, combining a growth strategy with the objective of 55% of the world’s population lives in urban areas, preserving the planet for future generations. One of a proportion that is expected to increase to 68% by the key action elements is "accelerating the shift 2050: the urbanization, combined with the overall growth of the world’s population could add another towards smart and sustainable mobility", which aims to reduce current transport emissions by 90% by 2.5 billion people to urban areas by 2050 (United 2050 (EC, 2019). Nations, 2020). The objective of energy efficiency in road transport In this context, a paradigm shift will have occurred can be approached from different perspectives, both over the course of a century: the minor living in terms of passengers and freight: shifting to other environment has become the dominant one. vehicle or fuel typologies, improving vehicles and Consequently, the transport system must keep pace fuels efficiencies, increasing occupancy, and with this urbanization trend; it must adapt and try to making driving more efficient, both in terms of eco- move towards the application of sustainable driving or eco-routing. Over the years, it has been mobility strategies. In fact, although there are many seen that strategies applied to urban logistics can be environmental problems concentrated in cities, they more or less efficient, but it is necessary to are still the engine of the economy and the centre of differentiate between different geographical, social business. Thus, efficient and well-planned freight transport may improve a city’s functioning and and economic contexts, as well as by type of transport (goods or passengers) at the time of enhance its competitiveness. applying one strategy or another (Marcucci et al., According to the United Nation Development 2017). Programme, cities account for 60 to 80 percent of Figure 1 shows that a large proportion of road energy consumption and at least 70 percent of transport emissions are concentrated in urban areas: carbon emissions, even though they occupy just 3% of the Earth’s land (UNDP 2019). 25% of total emissions are directly urban, and a large The evolution of proportion of interurban movements originates in or urban mobility turns out to be a fundamental key to are destined for cities: movement of people and change the current reality. goods. 1.2. Current trends in freight transport Within urban mobility, Last-mile delivery (LMD) is defined as the movement of goods from a transport hub to the final destination. It is the last leg of the logistic chain directly to consumer, and it is usually one of the more expensive, least efficient and most polluting logistic chapters: it can reach 60% of the total logistics costs of a firm (Lafkihi et al., 2019). Indeed, in addition to the externalities proper of urban mobility, i.e. traffic congestion, occupancy of urban land, GHG emissions and noise, LMD problems are also related to the collection of goods and mails, and to those parcels that are subject to Figure 1: Share of transport emissions in the complaints or delivered to a wrong end customer. European Union (EEA, 2019) Previous studies estimate urban freight transport According to Figure 1, 42% of transport emissions counting up to 30% of total traffic emissions is caused by freight transport. Considering both depending on the local context (Marcucci et al., urban and interurban freight transport, the share is 2017). In larger European urban areas, freight equal to 32%, a figure that is noteworthy and vehicles are responsible for half of particulate matter requires proper reflection. Indeed, the inner nature (PM) and a third of transport-related NO of urban freight transport is characterized by X emissions (Macharis and Melo, 2011). Nonetheless, the urban frequent starts and stops; several studies have freight transport system has to content with the already demonstrated that vehicles pollute more in current boom in e-commerce, which has 2 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia dramatically increased the number of parcels hand, new and emerging technologies could drive delivered each day, and raised customer innovation in city logistics and potentially decrease expectations, which now include not just fast but these effects. 5G technologies, drones and also free deliveries. In Spain alone, e-commerce autonomous cargo vehicles are only few examples exceeded 48.8 billion of euros in 2019, up almost of insight into the future that have good potential for 25% since the previous year, with a steady increase improving the efficiency and sustainability of urban year-on-year (Comisión Nacional de los Mercados y freight system (Taniguchi et al., 2020). la Competencia, 2020). Certainly, this increment However, those initiatives encounter various further strengthened by the spread of the COVID-19 difficulties to implement in practice, such as real- pandemic. time information is missing to provide dynamic The current pandemic, in addition to its known vehicle route service; the investment costs are very effects on people's health, has had important social high for carriers; and freight competitors are not and cultural effects over society indeed, such as an willing to share valuable information to cooperate, increase in work from home, the redefinition of since it is part of their business. It requires more priority spaces, and the boom of digital channels feasible and cost-effective strategies or solutions to among others. Specifically, the lockdown following aid the implementation in urban freight deliveries. the spread of the COVID-19 pandemic results in Ranieri et al. (2018) grouped different last-mile more and more products that, instead of being solutions for freight transport according to four main purchased by the consumer from the shop shelves, aspects: innovative vehicles, proximity stations or were placed in virtual shopping trolleys and will points, collaborative and cooperative solutions and therefore have to be available in logistics centres and innovation in public policies and infrastructures. In delivered directly to our homes. the same line, Lafkihi et al. (2019) and Mangiaracina 1.3. Last-mile deliveries: challenges and et al. (2019) conducted a comprehensive and features extensive literature review of LMD challenges and opportunities and identified several effective last- According to Ragás Prat (2018), the “Boom” of e- mile solutions including night deliveries, which can commerce leads to two main consequences: first, the shift traffic to off-peak hours, thereby reducing number of deliveries and the mobility of goods congestion by day and hence the related GHG vehicles circulating in the city is multiplied. Second, emissions. According to Olsson et al. (2019), in any address becomes a potential delivery point. In order to optimize LMD operations, companies can other words, the urban distribution of goods is act on several levels: (1) strategic level, such as carried out in residential areas not designed to host location of distribution centres, (2) tactical level, this type of operation on a massive scale. In this such as fleet size and night deliveries, and (3) context, the urban distribution of goods is being operational level, such as vehicle routing. rapidly transformed and requires urgent responses from operators and administrations. Literature pointed out the importance of realizing that geographic, economic, social and cultural Several strategies and initiatives are promoted to circumstances affect city logistics and people's increase energy and operative efficiency and thus perception of critical issues related to the city mitigate negative impacts of LMD: logistics (Rai et al., 2021). Depending on the area of  vehicle-routing improvements (Musolino implementation and intensity, same freight Et al., 2019; Croce et al., 2020), initiatives may lead different results on reducing  green and clean logistic fleets (Schliwa et social or environmental impacts. Furthermore, most al., 2015; Albergaria et al., 2018; Giordano of the scientific publications about the theme they et al., 2018), address transport policy with a theoretical approach  without presenting real data; others simulate the technological vehicle innovations (Alho et solutions in a specific area, without taking into al., 2018; Taniguchi et al., 2020) and  account different surroundings to apply. Real data competitors’ cooperation (Savelsberg et al., concerning the operative features of logistic 2015). companies are missing, and the lack of real data Additionally, several authors have identified trends makes the research of sustainable solutions for urban and advances in technology leading to a new logistic more complicated. paradigm of urban logistics. According to In order to discover effective measure for urban Sevelsbergh and Woansel (2016), trends in freight transport (especially LMD), it is essential to population growth and urbanization, coupled with properly characterize urban freight trips in both city the desire for speed, the sharing economy and the centre and peri-urban delivery area, where have rise of e-commerce, all increase the complexity of higher demand on goods movements, considering city logistics, and exacerbate its negative impacts on that it still lacks a full understanding of the issue. It congestion, safety and environment. On the other is crucial to analyse and identify the main energy and 3 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia operational inefficiencies of LMD to figure out Finally, Once data processing has been completed, which strategy shall be implemented, taking into all data were analysed to deepening knowledge account geographical, social and environmental about the main inefficiencies of LMD. We contexts. developed a data analysis method, meaning the correct manipulation of items, to generate and To our knowledge, there is a real scarcity of studies address meaningful results regarding the research presenting an analysis of the inefficiencies of LMD objectives settled (Section 3). based on a real field-test data. Thus, we performed a real data collection campaign to identify the main 2.1. The case study (in)efficiencies of last-mile postal deliveries (LMD) under real traffic conditions, in terms of both energy The case study takes place within the Metropolitan consumption and operative features. Area of Madrid. Madrid is the Main city of Spain, capital of the Country and of the same Region of After introducing the research context and the Madrid. The associated metropolitan area counts motivation of the paper, Section 2 presents the with 6.6 millions of inhabitants, making it one of the methodology used to carry out the investigation, most populous in the EU; of these, half of the including the data collection campaign and the data population is concentrated in the city of Madrid processing to compile the dataset. The analysis of (Monzón et al., 2017). the results and discussion are presented in Section 3. Finally, the main findings and policy The Metropolitan Area of Madrid covers 8,030 km2, recommendations are described in Section 4. with a ratio of urbanised surface of 11%. Three orbital motorways encircle the city (the M-30, M-40 2. METHODOLOGY and M-50) which are accessed from seven radial motorways. The terrain around the town is relatively This research analyses data collected in a field test flat, except for the mountain range, which is relative of last-mile deliveries carried out by the remoteness and borderline, located in the NW countrywide Spanish logistic company, with the aim province. of characterizing the main patterns of current distribution operations, identifying the main The study is based on data derived from a real data operational inefficiencies, and specifying to what collection campaign carried out in 2018 in extent these operations vary in two different delivery cooperation with Correos, which is the Spanish areas (city centre and peri-urban area). national postal service (state-owned). It has 51,000 employees and distributes more than 3,600 million The methodology developed to carried out the shipments per year. Within the Madrid Region, research is composed by three main working Correos owns 27 logistic depots that carry out packages, briefly described below. distribution by Light Duty Vehicles (LDV): 15 depots are located within the city of Madrid and 12 First, we defined the case study, which consisted of depots are located outside the city. Two of these, one a data collection campaign in real traffic conditions carried out in the Madrid Region, with the located in the city centre, and the other in the peri- collaboration of the main Spanish Logistic urban area of Madrid, were involved in the field Company. The field test involved 13 drivers from test,providing us with enough data for in depth two distribution depots of the same company. They analysis of the LMD in this region. were responsible for completing their normal delivery route. Meanwhile, all vehicles taking part in the field test were monitored on a second-by- second basis to obtain instantaneous GPS position, speed, trip time and other engine parameters. After each delivery route, drivers were required to fullfil a short questionnaire in which they could report any incidences encountered. A systematic data processing approach was then applied to integrate, filter and process all the recorded data, considering that a delivery route contains a certain number of parcel deliveries (micro-trips), whose characteristics vary according to the operating conditions. During the data processing, we thus defined two different databases: the first composed by delivery routes, and the second composed by each single parcel delivery micro-trips. 4 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia based on real data, under real conditions, without simulating or hypothesizing. 2.2. Dataset definition To perform the data analysis, it was necessary to define the dataset in two different dimensions, namely delivery routes and single parcel delivery micro-trips. Each delivery route contains a certain number of parcel deliveries (micro-trips) whose characteristics vary according to different operating conditions. Thus, we created two different databases: the first takes the form of delivery route, and describes the main aspects of last-mile postal delivery in terms of general operating features; the second allows the characterization of the driving patterns proper of single parcel delivery micro-trips and the understanding of energy inefficiency while delivering. Figure 3 shows the concept of delivery route ad single parcel delivery micro-trip just presented. Figure 2: The Metropolitan Area of Madrid and the field-test delivery areas. Figure 2 shows the main difference between the two delivery areas involved in the experiment, which are a key point to understand the operative features of LMD. One of the two depots provided by Correos is located in the municipality of Las Rozas, located in the NW fringe of the Metropolitan Area of Madrid, and it covers a peri-urban delivery area of over 130 Figure 3: The concept of delivery route and single km2. The other depot is located in the Madrid city parcel delivery micro-trip. centre, and serves a delivery area ten times smaller We used all recorded data to manually define the (approx. 12 km2), but with much higher population delivery micro-trips within each delivery route, density (27 times) than Las Rozas. based on the instantaneous recordings. We selected one-second time intervals with null speed recorded, The data collection campaign took place during four and if the consecutive time with null speed was weeks. For one month, 13 professional drivers (11 greater than 2 minutes (± 30 seconds depending on males and 2 females, aged between 33 and 62) were the GPS positioning) we considered it as a delivery involved in the field test. They drove electric and stop among subsequent micro-trips. Although this diesel LDV along their normal working routes, was not an absolute criterion, and we had to while GPS position, speed and other parameters manually check with GPS data. We thus created two were collected instantaneously each second. In total, datasets designed as the basis for the analysis by 242 valid delivery-working periods (each one programming in R, a free well-known software corresponding to a delivery route) of about four environment for statistical computing and graphics. hours were recorded, covering 7,262 km of route. The first database (Delivery route) comprises 242 The routes were not predefined; rather, they were delivery routes, each lasting about four hours of defined each day differently by the driver to driving. The second database (Micro-trips) consists optimize all deliveries during the working day. of 4,479 single parcel delivery micro-trips. This In addition, after each delivery route, we collected research considers 1,965 km of route covered in feedback from each driver through a short Madrid city centre and 5,297 km in the peri-urban questionnaire in which, among others items area of Madrid. required, they could report any incidences The aim of the research is to identify the main encountered, such as traffic congestion or lack of inefficiencies of LMD in terms of both energy parking. This information was very useful to consumption and operative features. In the case of understand real problems related to the distribution the vehicle’s energy inefficiency analysis, we have of parcels according to different delivery areas, considered the database composed by single parcel delivery micro-trips, considering several driving 5 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia parameters that have been shown to have a Besides that, the operating speed of the vehicle is a significant influence on the energy consumption of decisive factor when it comes to define the energy the vehicle. For the case of operational efficiency of LMD. Figure 4 shows the inefficiencies, we rely on statistical analysis based instantaneous speed proper of a vehicle during two on the two databases created, and on the feedbacks typical delivery routes in city centre and peri-urban from drivers. delivery area. 3. RESULTS AND DISCUSSION We adopt statistical methods to characterize the operating performance of LMD, and to present the extent to which distribution operations vary for the same company in two delivery areas studied (city centre and peri-urban area), as well as their inefficiencies. Thus, we first defined the main operative features of LMD to improve knowledge on the delivery activity within the two areas considered. Then, we accordingly identified their main inefficiencies: both in terms of energy and operative features, both in city centre and in peri-urban Figure 4: Instantaneous speed during two typical delivery routes delivery areas. In city centre, drivers drove almost three time slower 3.1. Energy inefficiencies of LMD than in peri-urban area, and, on average, 20% of trip In order to have a concrete idea of the main aspects time is at stop state owing to road congestion, traffic of delivery activities, Table 1 is presented. light and out-door delivery. Nonetheless, the very Distinguishing between peri-urban delivery area or low operating speed can be caused by several city centre, Table 1 shows the average values, as factors. According to our data, in city centre, well as minimum, maximum, and standard between 16% and 27% of the trip time is spent at deviation, of the main operative features according operating speeds of less than 3 km/h. In particular, to delivery area. 35% of delivery micro-trips in city centre have median speed of 7 km/h, and speeds of over 19 km/h Table 1: Operative features according to delivery is attained only 5% of the trip time. These deliveries area could therefore be performed by other means of transport than duty vehicles, since the median distance for single deliveries is less than 480 m, a longitude that can easily be covered on foot or by bicycle, for example. In addition, the covariance of speed (meaning the speed variability while driving) during these deliveries is the highest recorded, up to Despite similar delivery trip time of both depots 0.8 in most cases, highlighting a driving style implicated in the field test (average 4 hours), in city characterized by high accelerations and centre the average distance covered during a decelerations which, partly due to the road network delivery route is 14.3 km, which is 3.5 times lower and traffic flows, increase fuel consumption and the than in peri-urban delivery area. In city centre, 21% resulting emissions (Lois et al., 2019). There, even of the trip time is spent at speed below 3 km/h. if the reduced speed is maintained, the high level of Moreover, 50% of single parcel delivery micro-trips covariance causes the consumption to increase in the area is shorter than 477 meters and is markedly, and consequently the emissions. In order performed with a median speed below 13 km/h. to reduce km travelled by LDV a sensible approach might be to implement an urban consolidation centre The peri-urban case is quite different in terms of (UCC) from which different cargo bikes can operating features. The average distance covered by delivery nearby without polluting. Indeed, it has each driver during the delivery route is 50.9 km, and been demonstred that the implementation of a UCC the average distance to deliver a single parcel is 2.25 reduce up to 36% of the total emissions released to km (SD: 3.24), implying a rather disperse area. the environment (Firdausiya et al., 2019). There, 50% of micro-trips are more than 1.1 km long. The median speed is average 25 km/h and the To make the concept clearer, Figure 5 shows the percentage of time at operating speeds below 3 km/h instantaneous positioning and operating speed of a is 9%, less than half the percentage proper of the city vehicle during a typical city centre delivery route. centre. The colours are consistent with the operating speed of the vehicle: red for speed lower than 10 km/h, 6 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia yellow for speed between 10 km/h and 30 km/h and green for speed greater than 30 km/h. Figure 5: instantaneous speed and positioning of a delivery route in city centre. As shown in Figure 5, in a delivery route there are some micro-zones within which the driver doesn’t reach operating speeds higher than 10 km/h. In general, in 40% of cases, in city centre, the average speed of a single parcel delivery micro-trip is less than 10 km/h. This aspect should make us reflect on the real need to transfer LMD operations from vehicles to non-motorised mean, especially in city centre zones. Moreover, as shown in Figure 5, there are some streets covered several times during the same delivery route, due to the one-way streets in Figure 6: Delivery points covered by a single driver the city centre, as well as the lack of parking spaces. according to the day of delivery This aspect can in part justify the very low operating From Figure 6, it emerges that the deliveries are speed proper of the sample. Indeed the driver is almost the same from day to day. Indeed, the obliged to go along the same street several times delivery route of a specific driver is quite during different micro-trips of the same delivery homogeneous each day, since each driver is route, because of the network layout of the area: one- responsible of a certain zone of the corresponding way streets characterize the area, and in city centre delivery area. Minor variations are due to a single it is quite typical the difficult of parking because of end customer located in a nearby road, in a particular the surrounding traffic. building or in a nearby building, but in the same On the subject of the energy efficiency of vehicles, “micro-zone” for which each driver is responsible. it has already been proven by literature that heavy In this context, it would be useful to perform a deep accelerations and decelerations cause a considerable analysis of the geolocation of delivery points and increase in vehicle’s energy consumption (Garcia- determine the optimal location to install a set of Castro et al., 2018; Wang & Boggio-Marzet, 2018). parcel lockers, since 18% of deliveries in the peri- Thus, we want to analyse the proximity of the stops urban area were made in less than a minute and half, that the driver makes during his journey. Apart from a few tens of meters apart. These deliveries refer to those stops caused by traffic, traffic lights and those made along broad roads flanked by large pedestrian crossings, the driver must adapt his houses, one next to the other, with low traffic, where driving to a sort of intermittency of situations the driver can stop the vehicle right in front of the characterised by frequent start and stop and door, and deliver the parcel to the concierge, then consequently producing high GHG and pollutants repeat the process along the whole avenue. This emissions. Figure 6 shows the delivery points process produces deliveries with an extremely high covered by two drivers, one responsible to deliver in environmental impact. Given that vehicles pollute the the peri-urban area and the other in the city more in ignition, acceleration and braking, it would centre, differentiating colours according to the day be advisable in these road contexts to implement of delivery. another delivery system that is not based on LDVs. 3.2. Operative inefficiency of LMD For this analysis, we also took into account the questionnaires that all drivers fulfilled after each 7 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia delivery route. Thus, the main aspects with regard to geolocation system, even if, in this case, we would operational inefficiency of LMD have been drown certainly be confronted with privacy issues. both from statistical analysis and, for the most part, Finally there are two main concerns from the directly from drivers. feedbacks reported in the questionnaires by drivers: The first inefficiency identified in both delivery the lack of parking space and the congestion related areas is the long waiting times needed to carry out problems. The lack of parking space is one of the certain deliveries. Although the average delivery main reasons resulting in longer trip distance and time is 5 minutes, there are some deliveries that lower speed circulating (searching for free parking require a waiting times of more than an hour. On plot), consequently increases the use of energy and average, the delivery is carried out in 04:17 minutes produces higher emissions and air pollutants. In this in the city centre, with standard deviation equal to case, real-time parking information is required and 06:14, which means that around 70% of deliveries other modes such as cargo motorcycles or cargo are made in less than 10 minutes. In the peri-urban bycicles, which take up less space than cars should area, the delivery is made in average 07:25 minutes be considered to replace LDVs in future. It is worth with standard deviation equal to 09:41, which means pointing out that lack of parking space represents that around 70% of deliveries are made in less than 29% of the problems detected while delivering in 15 minutes. Although, the maximum delivery time city centre, but, according to the feedback of drivers, recorded during the data collection campaign is it is not relevant in the peri-urban case: deliveries by respectively 52 minutes in city centre and 96 in peri- foot in city centre should be promoted, as they would urban setting, regardless of the driver. This is due to totally avoid the problem of lack of parking space. the fact that, the final customer may be an industrial The other key problem during LMD is the road estate or a large company with extensive premises, congestion. Around 40% of delivery routes carried so the driver must be identified and undergoes out report traffic jams, regardless of the delivery several security checks before making the delivery, area considered. This aspect should be taken into and then deliver to several buildings on the estate. account by policymakers, since the deep nature of Unfortunately, there is not a single driver LMD is characterized by short length, so that traffic responsible for all these kind of deliveries, and jams can have a much greater influence on the consequently there is no fixed route covering these operative efficiency of LMD than in long distance delivery points; thus, operative efficiency is lost, deliveries. Another time it seems an optimal solution increasing the percentage of long waiting times for the hypothesis of introducing a fleet of cargo all the drivers involved in LDV deliveries. motorcycles, that occupy less space than the LDV, Another interesting aspect concerns those deliveries contributing to reduce the loss of time caused by that fail, for example because the customer is not at search of parking and traffic congestion. home when the driver comes or because the address of the delivery point is wrong. If the driver does not 4. CONCLUSIONS AND POLICY find the customer at home, he/she must return a RECOMMENDATIONS second time, at the end of his/her delivery route. If Nowadays, cities are undergoing a considerable the customer continues to be absent, the driver process of urbanisation, and urban freights transport transfers the package to the driver responsible of the has to deal with the current boom in e-commerce, next delivery route, who will have to change his which produces a sharp increase in freight transport route on purpose. This obviously implies a loss of activity and brings additional challenges to city time and reduce the operational efficiency of LMD, logistics. Through this research, we characterized but it also generates a waste of energy consumption the main patterns of last-mile postal deliveries in for the vehicle, as well as more kilometers are terms of energy consumption and operative features, travelled, implying that more traffic is created, more both in city centre and peri-urban delivery area. pollutant are emitted etc. With the purpose of changing the current emissions The percentage of failed deliveries at the first intent trends produced by last-mile delivery over cities and is around 25% in both delivery areas, and it is citizens, we thus aimed to study in quantitative reduced to 11% in the case of peri-urban deliveries terms, which are the most evident inefficiencies and to 8% in the city centre by means of the further while delivering within different delivery areas of deliveries. It is interesting that the same poor result the same Madrid Region. is reportated in the work of van Duin et al. (2015). To this aim, 242 delivery routes were analysed, In fact, in their study they affirm that First right time corresponding to 4,479 single parcel delivery micro- delivery is just 75%. This phenomenon has an trips and covering a total of 7,262 km of route: 1,965 extraordinary effect on the environment in our cities, km in Madrid city centre and 5,297 km in the peri- whether in terms of congestion, pollution, noise, etc. urban area of Madrid. The main findings emerged Surely, it would be interesting to create an instant from our results are the following: 8 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia  There are clear inefficiencies in the current Through this research, we pointed out that most of delivery performance, in terms of both the operating inefficiencies of last-mile deliveries energy consumption and operative features: a are similar regardless of the delivery area, both in large proportion of deliveries currently made terms of operative features and energy consumption. by LDV should be shifted to non-motorised Nonetheless, it emerges that within the same transport modes, especially in city centre. delivery area, the optimal solution could vary Concretely, 40% of deliveries in city centre according to the micro-zone. are characterized by average speed lower Due to the large amount of data analysed, this than 10 km/h, and, being shorter than 400 m, research provides a broader view for transport these deliveries could be easily covered by planners and policymakers in terms of LMD foot, cargobike or cargo motorcycle.  management, operation and usage to design The road network and the population density emissions reduction strategies for LMD. Also, the are quite etherogenous within the same results and conclusions emerged in this study shall delivery area, mainly in peri-urban area: the be implemented into the SUMPs – sustainable urban distance between subsequent deliveries can mobility plans - into the parts addressing city be of some tens of meters along certain roads, logistcs. or it can be of more kilometers in other micro-zones. Sustainable LMD strategies Further research should expand the study area and according to micro-zones should be focus on measuring and quantifying the impact of implemented. different last-mile solutions according to socio-  The lack of parking space, as well as the demographic circumstances. Furthermore, it wouold failed deliveries (25% at the first time) and be interesting to investigate the final consumer the frequent traffic jams imply a loss of time perspective, both in terms of usage of the urban for the driver and reduce the operational freight delivery service, and in terms of acceptability efficiency of LMD; they also generates a towards certain solutions (for example, it is known waste of energy consumption for the vehicle, that in order to reduce traffic impacts, freight as well as more kilometers are travelled. delivery could be carried out at night, but this would Thus, an interesting solution is the imply an increase in the noise produced by vehicles). implementation of a urban consolidation Acknowledgements centre (UCC) from which deliveries could be done by cargo motorcycle or cargo bike The authors would like to warmly thank the which occupy less space than cars and do not willingness shown by Dr. Ing. Libor Krejcì in imply a loss of time searching for parking, providing valuable suggestions and ideas that nor do they contribute to increase traffic contributed to the improvement of the paper. congestion and pollutants.  From an energetic point of view, it would be REFERENCES useful to perform a deep analysis of the [1] Alho, A. R., e Silva, J. D. A., de Sousa, J. P., & geolocation of delivery points and determine Blanco, E. (2018). Improving mobility by the optimal location to install a set of parcel optimizing the number, location and usage of lockers in those kind of roads characterized loading/unloading bays for urban freight by high population density, since most of the vehicles. Transportation Research Part D: time the delivery points reached by drivers Transport and Environment, 61, 3-18. are almost the same, day by day. Moreover, [2] Comisión Nacional de los Mercados y la the proximity of the delivery points one to Competencia (2020). each other results in additional fuel wastage http://data.cnmc.es/datagraph/ Accessed 10 due to the frequent start end stop of the February 2021 vehicle added to those produced by the road [3] Croce, A. I., Musolino, G., Rindone, C., & context, i.e. due to traffic jams, traffic lights Vitetta, A. (2020). Route and Path Choices of or pedestrian crossings. Freight Vehicles: A Case Study with Floating  In terms of operational efficiency, there is a Car Data. Sustainability, 12(20), 8557. clear lack of homogeneity with regard to the time needed for delivery. Although on [4] de Mello Bandeira, R. A., Goes, G. V., Gonçalves, D. N. S., Márcio de Almeida, D. A., average it takes 5 minutes to make a delivery, & de Oliveira, C. M. (2019). Electric vehicles in it can take up to one hour. It could be useful the last mile of urban freight transportation: A to define a fixed route covering deliveries to sustainability assessment of postal deliveries in large companies, avoiding that all drivers Rio de Janeiro-Brazil. Transportation Research have to experience the same long waiting Part D: Transport and Environment, 67, 491- times. 502. 9 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia [5] Ecotale, 2014. External costs of transport and [16] Macharis, C., & Melo, S. (2011). Introduction- land equalization. city distribution: challenges for cities and http://www.ecotale.eu/files/it/ecotale_it.pdf researchers. City distribution and urban freight transport: multiples perspectives, Edward Elgar [6] EEA, European Environment Agency, “Annual Publishing, Northampton, 1-9. European Union greenhouse gas inventory 1990-2017 and inventory report 2019” [17] Mangiaracina, R., Perego, A., Seghezzi, A., & EEA/PUBL/2019/051. Disponible en: Tumino, A. (2019). Innovative solutions to https://www.eea.europa.es/publications/europe increase last-mile delivery efficiency in B2C e- an-union-greenhouse-gas-inventory-2019 commerce: a literature review. International Journal of Physical Distribution & Logistics [7] EEA, European Environment Agency, Management. “National action across all sectors needed to reach greenhouse gas Effort Sharing targets”, [18] Marcucci, E., Le Pira, M., Gatta, V., Inturri, G., 2019. Disponible en: Ignaccolo, M., & Pluchino, A. (2017). https://www.eea.europa.eu/publications/nationa Simulating participatory urban freight transport l-action-across-all-sectors policy-making: Accounting for heterogeneous stakeholders’ preferences and interaction [8] European Commission (2019). The European effects. Transportation Research Part E: Green Deal. Disponible online en: Logistics and Transportation Review, 103, 69- https://ec.europa.eu/info/sites/info/files/europea 86. n-green-deal-communication_en.pdf [19] Ministero delle Infrastrutture e dei Trasporti [9] EUROSTAT, 2020. Disponible online en: (2020). La logistica urbana in una visione https://ec.europa.eu/eurostat/statistics- integrata. explained/pdfscache/1180.pdf https://mit.gov.it/sites/default/files/media/pubbl [10] Firdausiyah, N., Taniguchi, E., & Qureshi, A. G. icazioni/2021- (2019). Impacts of Urban Consolidation Centres 01/La%20logistica%20urbana%20- for Sustainable City Logistics Using Adaptive %20n.1%20dicembre%202020.pdf Dynamic Programming Based Multi-Agent [20] Monzón, A., Cascajo, R., Pieren, G., Romero, Simulation. IOP Conference Series: Earth and C., Delso, J.: Informe del Observatorio de la Environmental Science (Vol. 328, No. 1, p. Movilidad Metropolitana 2015. (2017). 012071). IOP Publishing. Ministerio de Agricultura y Pesca, Alimentación [11] Garcia-Castro, A., Monzon, A., Valdes, C., & y Medio Ambiente. Romana, M. (2017). Modeling different [21] Musolino, G., Rindone, C., Polimeni, A., & penetration rates of eco-driving in urban areas: Vitetta, A. (2019). Planning urban distribution Impacts on traffic flow and center location with variable restocking demand emissions. International Journal of Sustainable scenarios: General methodology and testing in a Transportation, 11(4), 282-294. medium-size town. Transport policy, 80, 157- [12] Giordano, A., Fischbeck, P., & Matthews, H. S. 166. (2018). Environmental and economic [22] Olsson, J., Hellström, D., & Pålsson, H. (2019). comparison of diesel and battery electric Framework of last mile logistics research: A delivery vans to inform city logistics fleet systematic review of the replacement strategies. Transportation Research literature. Sustainability, 11(24), 7131. Part D: Transport and Environment, 64, 216- 229. [23] Ragàs Prat, I.: Logística urbana Manual para operadores logísticos y administraciones [13] IDAE. Second Action Plan of Spanish Strategy públicas. Marge books (2018). of Energy Saving and Efficiency 2011–2020. Madrid, 2011. Available online: [24] Rai, Heleen Buldeo, Sara Verlinde, and Cathy https://www.idae.es/uploads/documentos/docu Macharis. "Who is interested in a crowdsourced mentos_11905_PAEE_2011_2020._A2011_A_ last mile? A segmentation of attitudinal a1e6383b.pdf profiles." Travel Behaviour and Society 22 (2021): 22-31. [14] Lafkihi, M., Pan, S., & Ballot, E. (2019). Freight transportation service procurement: A literature [25] Ranieri, L., Digiesi, S., Silvestri, B., & review and future research opportunities in Roccotelli, M. (2018). A review of last mile omnichannel E-commerce. Transportation logistics innovations in an externalities cost Research Part E: Logistics and Transportation reduction vision. Sustainability, 10(3), 782. Review, 125, 348-365. [26] Savelsbergh, M., & Van Woensel, T. (2016). [15] Lois, D., Wang, Y., Boggio-Marzet, A., & 50th anniversary invited article—city logistics: Monzon, A. (2019). Multivariate analysis of fuel Challenges and opportunities. Transportation consumption related to eco-driving: Interaction Science, 50(2), 579-590. of driving patterns and external factors. Transportation Research Part D: Transport and Environment, 72, 232-242. 10 Alessandra Boggio-Marzet, Andres Monzon YRS 2021 DEFINING ENERGY (IN)EFFICIENCY OF LAST-MILE POSTAL DELIVERY: A REAL CASE 15. -17. September STUDY IN THE MADRID REGION Portoroz, Slovenia [27] Schliwa, G., Armitage, R., Aziz, S., Evans, J., & [31] United Nations, Department of Economic and Rhoades, J. (2015). Sustainable city logistics— Social Affairs: World Social Report 2020 – Making cargo cycles viable for urban freight Inequality in a rapidly changing world. transport. Research in Transportation Business https://www.un.org/development/desa/publicati & Management, 15, 50-57. ons/world-social-report-2020.html (2020). Accesse 10 February 2021. [28] Simoni, M. D., Marcucci, E., Gatta, V., & Claudel, C. G. (2019). Potential last-mile [32] Van Duin, J. H. R., et al. "Improving home impacts of crowdshipping services: a delivery efficiency by using principles of simulation-based evaluation. Transportation, 1- address intelligence for B2C deliveries." 22. Transportation Research Procedia 12 (2016): 14-25. [29] Taniguchi, E., Thompson, R., & Qureshi, A. (2020). Modelling city logistics using recent [33] Wang, Y., & Boggio-Marzet, A. (2018). innovative technologies. Transportation Evaluation of eco-driving training for fuel Research Procedia 46, 3-12. efficiency and emissions reduction according to road type. Sustainability, 10(11), 3891. [30] United Nations Development Program, Human Development Report 2019: Beyond income, beyond averages, beyond today: inequalities in human development in the 21st century. http://hdr.undp.org/en/2019-report (2019). Accessed 10 February 2021. 11 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE SYSTEM TESTING Victor Bolbot, Dr Gerasimos Theotokatos, Prof Evangelos Boulougouris, Dr Dracos Vassalos, Prof Maritime Safety Research Centre Department of Naval Architecture, Ocean and Marine Engineering University of Strathclyde 100 Montrose street, G4 0LZ, Glasgow, United Kingdom ABSTRACT The introduction of autonomous ships is becoming a more tangible reality as a number of projects are dealing with developing next-generation ships operating in the European waters. One of such projects is AUTOSHIP aiming at developing two demonstrator ships with autonomous capabilities, one for short sea operations and another for inland waterways operations, funded by the European Commission. One of the key enabling technologies to be developed for this purpose is the collision avoidance system, whose safe performance is critical for autonomous ships. This can be ensured by proper testing of the collision scenarios in a simulated environment and during sea trials. Still, the quality of testing results will be depending on the number of considered scenarios. The aim of this study is therefore to suggest a systematic way to identify potential collision conditions which can be used as input for testing of the collision avoidance system. In our systematic approach, the potential situations that collision avoidance will have to handle are generated in a semi-automatic way by defining potential values for environmental features, objects' parameters, and intelligent agents' behavior. The generated hazardous scenarios are used to demonstrate the applicability of the developed approach. Keywords: Autonomous ships; Collision avoidance; Safety; Testing scenarios 1. INTRODUCTION and verification techniques which would allow earning sufficient confidence in the collision We live in an era when novel systems are being avoidance functionalities. In other words, only once introduced, including marine autonomous surface an adequate number of scenarios has been tested in a ships (MASS) [1]. Major industrial initiatives, such virtual or real environment can we be assured that as Yara Birkeland [2] and ASKO [3], have already such a system will not jeopardise safety, once put into launched commercial autonomous ships projects operation. along by projects such as MUNIN [4], AAWA [5], SISU, SVAN [6], AUTOSHIP [1] and many others. The navigation of ships is primarily regulated by COLREGs [11]. However, the COLREG The collision avoidance system can be considered as requirements have been designed having crew in a critical system on the autonomous ships as it will be mind and not the automatic systems. They do not making decisions [7] affecting the safety of the own provide numerical criteria for crew actions and their and surrounding ships. A number of accidents have implementation rely on the crew judgement, so it can occurred in the other industries involving similar not be used to develop a comprehensive set of testing functions as referred in the literature since the design scenarios. They do not include information on the of such systems is associated with a number of frequency of encountered situations and do not challenges[8, 9]. One of the challenges is ensuring contain a comprehensive list of objects with which adequate situational coverage of the potential the MASSs will be intracting. conditions that may be encountered by the ship associated with environmental complexity [10]. Automatic Identification System (AIS) data from ship traffic systems can be a valuable data source for This problem can be tackled through systematic identifying such situations. It has been widedly used identification and analysis of the potential collision for the analysis of traffic conditions and identification and interaction scenarios and by developing testing of the most probable collision situations (see for 12 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia instance [12-15]). However, the quality of AIS data First the input parameters are identified. This can be is under question, as the ships which have switched implemented with the assistance of the dedicated off their transponder and small recreational ships, questionnaires and with the support of the AIS and which do not require to have AIS, are not visible on meteorological data. this data [16]. Also objects other than ships and buoys Then, during the second step, the encountering are not included in this data. situations are generated using one of the sampling Several previous works have suggested the techniques: Sobol sequences (SB)[25], random development of collision avoidance system on ships Monte Carlo (RMC) sampling [26] or Latin e.g. [17-19], but very few focused on the testing and hypercube (LH) [27]. Uniform distribution is testing scenarios generation [20, 21]. A number of assumed for the sampled parameters. other studies focused on testing scenarios generation The SB are a quasi-random low-discrepancy such as in automotive [22, 23], aviation [24], etc. sequence that uses two as basis for successively finer The present study aims at developing testing uniform partition of the unit in multidimensional scenarios for ship collision avoidance system. A space [25]. Example of SB in one dimension is the number of sampling techniques is employed to following sequence: generate various encountering conditions, but only SB the hazardous situations are selected for testing. The 1 = [1,1/2, ¼, ¾, …] (1) hazardous situations are identified using some As it does not include stochastic calculations, every criteria. In the study one Own Ship (OS) and two call of the SB sequence will generate the same result, Target Ships (TS) are considered. The ships are in comparison to the RMC and LH sampling considered to operate in an open ocean. More details techniques. are provided in the next sections. The RMC generates samples based on repeated 2. METHODOLOGY random sampling according to a predefined probability distribution [26]. The RMC are widely The methodology is presented using pseudocode in used in a number of application areas and scientific Table 1. fields. The problem of random sampling can be likened to the picking up a ball with a number ranging Table 1: The pseudocode used for scenarios generation from 1 to 100 from a Bingo machine. Algorithm Testing scenarios generation 1: Procedure: Pseudocode for testing scenario generation Lastly, the LH is also method for generating a near-2: Input: speed, weather, currents, ships size, number of test random samples [27]. In terms of sampling, it splits points (n), number of ships, etc. the sample space into a square grid and takes only one 3: Generate potential situations using SB, LH, RMC sampling sample in each row and in each column. The For i=1:n % for all the sample points sampling can be likened to rook chess problem, 4: Estimate where the rooks must be set on such squares that they 𝐷1 %distance between MASS and high speed cannot be taken by the other rook, with the constrain craft 𝑇𝐶𝑃𝐴 that there must be a rook on each row and column. 1 %time of closest approach 𝐷𝐶𝑃𝐴1 %distance of closest approach For each of the encountering situation generated --------------------------------------------------------------------- --- using the sampling techniques (SB, RMC, LH), 𝐷2 %distance between MASS and sailboat geometric distance between the OS and TSi (𝐷𝑖), time 𝑇𝐶𝑃𝐴2 %time of closest approach to the closest point of approach between the OS and 𝐷𝐶𝑃𝐴2 %distance of closest approach --------------------------------------------------------------------- TSi (𝑇𝐶𝑃𝐴𝑖) and distance at the closest point of --- approach between the OS and TSi (𝐷𝐶𝑃𝐴𝑖) are 𝑎1 %estimation of ellipse primary axis for OS estimated. The equations for these metrics are 𝑏1 %estimation of ellipse secondary axis for OS provided below (eq. 2-6), where (𝑥, 𝑦) is location for 5: If 𝐷1<𝐷2 then 6: If 𝑇𝐶𝑃𝐴 each ship, 𝑢 is speed,𝜑 is ship speed direction, 0 is 1>0 & 𝐷𝐶𝑃𝐴1<𝑎1 & 𝐷1 <1nm then Situation should be considered used to denote OS and 𝑖 TSi based on information Elseif 𝑇𝐶𝑃𝐴2>0 & 𝐷𝐶𝑃𝐴2<𝑎1 & 𝐷2 <1nm then presented in [18, 20, 21]. The equations 3 and 4 Situation should be considered present the intermediate steps of the calculations. End if 7: Elseif 𝐷2<𝐷1 then 8: If 𝑇𝐶𝑃𝐴 𝐷 2>0 & 𝐷𝐶𝑃𝐴2<𝑎1 & 𝐷2 <1nm then 𝑖 = √(𝑥0 − 𝑥𝑖)2 + (𝑦0 − 𝑦𝑖)2 (2) Situation should be considered Elseif 𝑇𝐶𝑃𝐴1>0 & 𝐷𝐶𝑃𝐴1<𝑎1 & 𝐷1 <1nm then 𝑘 2 2 2,𝑖 = 𝑢0 − 2𝑢0𝑢𝑖 𝑐𝑜𝑠(𝜑0 − 𝜑𝑖) + 𝑢𝑖 (3) Situation should be considered End if 9: End if 𝑘1,𝑖 = 2 (𝑢0𝑦0 𝑐𝑜𝑠(𝜑0) − 𝑢0𝑦𝑖 𝑐𝑜𝑠(𝜑0) − 10: Calculate the MSRE, mean for sampling methods 𝑢𝑖𝑦0 𝑐𝑜𝑠(𝜑𝑖) + 𝑢𝑖𝑦𝑖 𝑐𝑜𝑠(𝜑𝑖) + 𝑢0𝑥0 𝑠𝑖𝑛(𝜑0) − 11: End procedure 𝑢0𝑥𝑖 𝑠𝑖𝑛(𝜑0) − 𝑢0𝑥𝑖 𝑠𝑖𝑛(𝜑0) − 𝑢𝑖𝑥0 𝑠𝑖𝑛(𝜑𝑖) + 𝑢𝑖𝑥𝑖 𝑠𝑖𝑛(𝜑𝑖)) (4) 13 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia 𝑘 𝑇𝐶𝑃𝐴 1,𝑖 3. INVESTIGATED CASES 𝑖 = − (5) 2⋅𝑘2,𝑖 For the analysis we have employed a small short-sea 𝐷𝐶𝑃𝐴 2 2 ship (OS) from the AUTOSHIP project [1] which is 𝑖 = √𝑘2,𝑖 ⋅ 𝑇𝐶𝑃𝐴𝑖 + 𝑘2,𝑖 ⋅ 𝑇𝐶𝑃𝐴𝑖 + 𝐷𝑖 (6) operating outside coasts of Norway and is interacting with a sailing boat (TS1) and a high speed craft Several criteria were considered for triggering the (TS2). The input parameters of the investigated critical situations identification, namely: situations are provided in Table 2. The sampled a) Firstly, 𝑇𝐶𝑃𝐴 parameters with their ranges are provided in Table 3. 𝑖 > 0, as we are not interested in the situation where the closest encounter These 18 parameters are varying from 0 to their happened in the past. maximum value by using the sampling techniques b) Secondarily, the current distance between ships which have been referred previously (SB, RMC, LH). For comparing sampling techniques performance, should be equal or less than 1 nm (𝐷𝑖 <1nm). 100 up to 100,000 samples are generated. Wave COLREGs do not specify any specific distance, height is used as an independent parameter to and the 1nm has been set in line with other estimate the mean and MSRE, since it is not used as publications [18, 28]. It is the distance at which input to eq. (2)-(8). The test area is set to [0 3nm] x it can be considered that OS should take action [0 3nm] in line with [18]. to avoid the collision. c) The third is criteria is related to the dangerous Table 2: The input parameters. zone. The dangerous zone is defined using an Fish feeding Sailboat High ellipse set at ship location with axis dependent ship speed on the OS length and speed according to craft equations provided in [18], were 𝑎 Length 74.7m 6m 12m 1 denotes the Beam 13.6m 2m 2.5m primariy axis, 𝑏1 the secondary axis of ellipse Max speed 15kn 10kn 40kn and L the ship length (shown also in Figure 1). Max current 3m/s This is expressed mathematically as 𝐷𝐶𝑃𝐴2<𝑎1. Max waves 2m height (4 − 0.3(10 − 𝑉 Max wind 14 kn 𝑎 1))𝐿, 𝑢 ≤ 10 𝑘𝑛 1 = { (7) speed (4 + 0.3(𝑉1 − 10))𝐿, 𝑢 > 10 𝑘𝑛 (1.6 − 0.14(10 − 𝑉 Table 3: The sampled parameters. 𝑏 1))𝐿, 𝑢 ≤ 10 𝑘𝑛 1 = { (8) (1.6 + 0.14(𝑉 Random parameters Range 1 − 10))𝐿, 𝑢 > 10 𝑘𝑛 Fish feeding ship speed [0 max] Fish feeding ship speed direction [0 2pi] The criteria must concurrently be satisfied for a Fish feeding ship location [0 3nm] x [0 3nm] situation to be considered as hazardous. The analysis Sail boat speed [0 max] of criteria starts with the ship that is the closest to the Sail boat speed direction [0 2pi] OS. Sail boat location [0 3nm] x [0 3nm] High speed craft speed [0 max] The performance of sampling techniques is assessed High speed craft direction [0 2pi] using the following criteria a) the number of High speed craft location [0 3nm] x [0 3nm] identified collision scenarios; b) the difference Current speed [0 max] between the anticipated and actual mean for one of Current direction [0 2pi] Waves height [0 max] the independent parameters that are not used as input Waves direction [0 2pi] to the metrics and criteria in eq. (2) to (8) and c) the Wind speed [0 max] mean square root error (MSRE) of each sample for Wind direction [0 2pi] the independent parameter. 4. RESULTS AND DISCUSSION 4.1. Encountering situations generated using Sobol The generated encountering conditions for SB number = 100 are provided in Figure 2. As it can be observed, the algorithm is effective in generating and identifying hazardous situations that may occur between the OS and the other two TS. Still, it can be also observed that out of the 100 potential encountering situations only 4 were classified as critical and important for testing. It practically indicates that during automatic scenarios generation Figure 1: The demonstration of dangerous zone a lot of ‘non-hazardous’ encountering situations will around ship 14 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia be produced. Still, this is dependent on the selected and LH sampling technique and 5 times in Table 5, criteria and metrics. as RMC and LH include stochastic calculations. Only for N=105 it was repeated 3 times. No repetition was 4.2. Comparison between sampling required for SB samples, as it is a quasi-random techniques technique, so the calculations are deterministic. It can The coverage of the 3nmx3nm test area using the be observed that SB samples in this algorithm three sampling techniques in the present paper is identify similar number of scenarios with other provided in Figure 3. It can be visually observed that techniques with the effectiveness at 4.5%. High the coverage of the area using SB sequences is more discrepancy between the estimated mean waves uniform than the other sampling techniques. For height value and some samples mean was observed RMC sampling, clustering can be observed e.g. at as demonstrated in Table 4 for RMC and LH x=2000m and y=5000m. At the same coordinates at compared to SB. The stability of the results is the LH sampling a void can be observed. significant advantage of the SB sequences due to their quasi-random nature. As a consequence, there is Further conclusions can be drawn through the metrics no need to repeat the results saving computation time estimated for the different sampling techniques in when applying SB sequences. This is in line with the Table 4 and Table 5. The results in Table 4 are findings for SB in other engineering or mathematical generated repeating the sampling 10 times for RMC problems [29-31]. Figure 2: The hazardous encountering situations generated using Sobol sequences 15 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia Figure 3: The filling of the test area using different Sampling techniques Table 4: The comparison between different sampling  The Sobol sampling permitted more stable techniques with 100 samples results compared to the other techniques due Waves height to quasi-random nature. (expected mean 1) N=100 Mean sampling Our future work will focus on specifying more repetition complex encountering conditions such as ship in (total 10) 1 2 3 4 5 … proximity to shore. Another direction of future Collision research is to specify coverage criteria for Sobol scenarios n 6 11 7 4 3 … 5.80 Waves height sampling. LH mean 0.73 1.26 1.14 0.65 0.62 … 1.01 Acknowledgments MSRE 0.38 0.36 0.31 0.48 0.61 … 0.35 The study was carried out in the framework of the Collision scenarios n 4 2 5 7 7 … 4.50 AUTOSHIP project [1], which is funded by the MC Waves height European Union's Horizon 2020 research and R mean 0.84 0.91 0.45 0.98 0.96 … 0.98 innovation programme under agreement No 815012. MSRE 0.21 0.58 0.39 0.45 0.34 … 0.32 The authors also greatly acknowledge the funding Collision from DNV AS and RCCL for the MSRC scenarios n 4 4.00 establishment and operation. The opinions expressed SB Waves height mean 0.83 0.83 herein are those of the authors and should not be MSRE 0.27 0.27 construed to reflect the views of EU, DNV AS, RCCL or other involved partners in the AUTOSHIP Table 5: The comparison between different sampling project. techniques with 102, 103, 104, 105 samples REFERENCES N= 102 103 104 105 [1] AUTOSHIP. Autonomous Shipping Initiative for Collision scenarios n 5.80 42.60 446 4548 European Waters. 2019. LH Waves height mean 1.01 0.99 0.98 0.99 MSRE 0.35 0.33 0.33 0.34 [2] Yara. Yara Birkeland press kit. 2018. Collision scenarios n 4.50 42.00 443.40 4564 C [3] Smartmaritime. ASKO to build two autonomous MR Waves height mean 0.98 0.98 0.99 1.00 vessels for Oslo fjord operations. 2020. MSRE 0.32 0.36 0.34 0.33 [4] MUNIN. Maritime Unmanned Navigation Collision scenarios n 4.00 40.00 476.00 4496 through Intelligence in Networks. 2016. SB Waves height mean 0.83 0.91 1.04 1.00 MSRE 0.27 0.29 0.35 0.34 [5] AAWA. AAWA project introduces the project’s first commercial ship operators. 2016. 5. CONCLUSIONS AND FURTHER [6] Daffey K. Technology Progression of Maritime RESEARCH Autonomous Surface Ships. 2018. In this study an approach for automatic generation of [7] Bolbot V, Theotokatos G, Boulougouris E, hazardous ship traffic conditions has been presented. Vassalos D. A novel cyber-risk assessment It was found that: method for ship systems. Safety Science. 2020;Under review.  The approach is effective for generating hazardous ship traffic conditions.  The Sobol sampling provided accuracy comparable with other techniques. 16 Victor Bolbot, Gerasimos Theotokatos, Evangelos Boulougouris, Dracos Vassalos YRS 2021 METHODOLOGY FOR DETERMINING SCENARIOS FOR COLLISION AVOIDANCE 15. -17. September SYSTEM TESTING Portoroz, Slovenia [8] National Transportation Safety Board (NTSB). [21] Pedersen TA, Glomsrud JA, Ruud E-L, Collision Between a Car Operating With Simonsen A, Sandrib J, Eriksen B-OH. Towards Automated Vehicle Control Systems and a simulation-based verification of autonomous Tractor-Semitrailer Truck Near Williston, navigation systems. Safety Science. Florida May 7, 2016 Highway Accident Report. 2020;129:104799. United States, Washington D.C.: National [22] Li L, Huang W, Liu Y, Zheng N, Wang F. Transportation Safety Board; 2017. Intelligence Testing for Autonomous Vehicles: A [9] Guiochet J, Machin M, Waeselynck H. Safety- New Approach. IEEE Transactions on Intelligent critical advanced robots: A survey. Robotics and Vehicles. 2016;1:158-66. Autonomous Systems. 2017;94:43-52. [23] Fremont DJ, Kim E, Pant YV, Seshia SA, [10] Alexander R, Hawkins HR, Rae AJ. Situation Acharya A, Bruso X, et al. Formal Scenario- coverage–a coverage criterion for testing Based Testing of Autonomous Vehicles: From autonomous robots. 2015. Simulation to the Real World. 2020 IEEE 23rd International Conference on Intelligent [11] COLREGS. International Regulations for Transportation Systems (ITSC)2020. p. 1-8. Preventing Collisions at Sea - Articles of the Convention on the International Regulations for [24] Lindvall M, Porter A, Magnusson G, Schulze C. Preventing Collisions at Sea. 1972. Metamorphic model-based testing of autonomous systems. 2017 IEEE/ACM 2nd [12] Gao M, Shi G-Y. Ship collision avoidance International Workshop on Metamorphic Testing anthropomorphic decision-making for structured (MET): IEEE; 2017. p. 35-41. learning based on AIS with Seq-CGAN. Ocean Engineering. 2020;217:107922. [25] Sobol' IyM. On the distribution of points in a cube and the approximate evaluation of integrals. [13] Mou JM, Van der Tak C, Ligteringen H. Study Zhurnal Vychislitel'noi Matematiki i on collision avoidance in busy waterways by Matematicheskoi Fiziki. 1967;7:784-802. using AIS data. Ocean Engineering. 2010;37:483-90. [26] Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E. Equation of State [14] Goerlandt F, Goite H, Valdez Banda OA, Calculations by Fast Computing Machines. The Höglund A, Ahonen-Rainio P, Lensu M. An Journal of Chemical Physics. 1953;21:1087-92. analysis of wintertime navigational accidents in the Northern Baltic Sea. Safety Science. [27] McKay MD, Beckman RJ, Conover WJ. A 2017;92:66-84. Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of [15] Kulkarni K, Goerlandt F, Li J, Banda OV, Kujala Output from a Computer Code. Technometrics. P. Preventing shipping accidents: Past, present, 1979;21:239-45. and future of waterway risk management with Baltic Sea focus. Safety Science. [28] Rødseth ØJ, Faivre J, Hjørungnes SR, Andersen 2020;129:104798. P, Bolbot V, Pauwelyn A-S, et al. AUTOSHIP deliverable D3.1 Autonomous ship design [16] IMO. Revised guidelines for the onboard standards, Revision 2.0. 2020. operational use of shipborne Automatic Identification Systems (AIS) Resolution [29] Burhenne S, Jacob D, Henze GP. Sampling based A.1106(29). 2015. on Sobol’sequences for Monte Carlo techniques applied to building simulations. Proc Int Conf [17] Huang Y, van Gelder PHAJM. Collision risk Build Simulat2011. p. 1816-23. measure for triggering evasive actions of maritime autonomous surface ships. Safety [30] Kucherenko S, Albrecht D, Saltelli A. Exploring Science. 2020;127:104708. multi-dimensional spaces: A comparison of Latin hypercube and quasi Monte Carlo sampling [18] Namgung H, Kim J-S. Collision Risk Inference techniques. arXiv preprint arXiv:150502350. System for Maritime Autonomous Surface Ships 2015. Using COLREGs Rules Compliant Collision Avoidance. IEEE Access. 2021;9:7823-35. [31] Qian G, Mahdi A. Sensitivity analysis methods in the biomedical sciences. Mathematical [19] Brcko T, Androjna A, Srše J, Boć R. Vessel biosciences. 2020;323:108306. Multi-Parametric Collision Avoidance Decision Model: Fuzzy Approach. Journal of Marine Science and Engineering. 2021;9:49. [20] Woerner K. COLREGS-compliant autonomous collision avoidance using multi-objective optimization with interval programming. MASSACHUSETTS INST OF TECH CAMBRIDGE; 2014. 17 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Ellen Boudry, Researcher Ricardo Nieuwkamp, Senior Researcher Peter Silverans, Research Manager Vias institute Haachtsesteenweg 1405, Brussels 1130, Belgium Ellen.Boudry@vias.be Vanlaar Ward, Chief Operating Officer Traffic Injury Research Foundation 171 Nepean Street, Ottawa, Canada tirf@tirf.ca ABSTRACT Background. Driving under the influence of alcohol is a major threat to road safety. An important question is how recidivism can be prevented (i.e., reoffending). Sanctions can be distinguished in traditional and alternative sanctions. In this study the effectiveness of Driver Improvement (DI) courses – as alternative sanction – is determined based on recidivism data. Method. A quasi-experimental design (i.e., a static group comparison) is applied. Drivers in the comparison group (traditional sanction) are matched with the experimental group (DI course) on five aspects. 606 drivers convicted between 2010 and 2014 (reference conviction) were included. Recidivism was followed-up until January 2018. The effect is measured by comparing recidivism rate between groups and with a Cox regression analysis of time to recidivism. Results. After the follow-up period, recidivism rate of the experimental group was 41% lower than the comparison group. Based on the Cox regression analysis, the experimental group was 2.6 times less likely to reoffend. However, when the experimental group reoffends, their Blood Alcohol Concentration is higher than the comparison group. Conclusions. In line with international research, imposing a DI-course can reduce recidivism by 41%. Based on risk factors for recidivism, we aimed to identify driver profiles who benefit more from DI-courses. Profiles could not be identified due to ambiguous results. Replication with a larger sample can enhance the identification of driver profiles and contribute to an improved road safety in Belgium. Keywords: Effectiveness, Driver Improvement, Recidivism, Driving under the influence (DUI) institute (formely known as Belgian Road Safety 1. INTRODUCTION Institute) estimated that 1.9% of the total kilometers driven in Belgium were completed by a driver who 1.1. Driving under the influence of alcohol tested above the legal alcohol limit of 0.5 g/l (Brion, and road safety Meunier, Pelssers, Leblud, & Silverans, 2019). The consumption of alcohol makes a driver less able During weekend nights, the prevalence of DUI is to correctly assess a traffic situation (Caird, Lees, & estimated to be even higher. 12.6% of the Belgian Edwards, 2005). In addition, alcohol has a visible drivers tested above the legal alcohol limit during effect on driver performance; there is an increase in weekend nights (Brion et al., 2019). Next, it is lane waving and a higher variablility in speed (Irwin, estimated that the crash risk of these drivers is 40% Iudakhina, Desbrow, & Mccartney, 2017). In 2019, higher compared with sober drivers (Leskovšek & 9% of Belgian drivers involved in an accident tested Goldenbeld, 2018). In addition, repeat traffic positive on the use of alcohol (Slootmans, 2020). offenders are considered to be harmful for the Hence, driving under the influence (DUI) of alcohol society (Carnegie, Strawderman, & Li, 2009). More is a major treat to road safety worldwide. specifically, it is assumed that repeat traffic offenders have a higher chance to be involved in a Despite the higher risk of a traffic accident, DUI of traffic accident (Goldenbeld & Twisk, 2009). alcohol occurs relatively frequent. In 2018, Vias 18 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia Therefore, the question raises how to prevent that the recidivism rate by 40%, which means that drivers, who drove under the influence of alcohol, participants of a DI course reoffended 40% less will reoffend in the future. compared to traditional sanctioned drivers (Slootmans et al., 2017). 1.2. Belgian legal framework In Belgium, to the best of our knowledge, only two Belgium’s legal alcohol limit is, as in most European studies have been published in the (grey) literature countries, set at a blood alcohol concentration that investigated the effectiveness of DI on the base (BAC) of 0.5 g/l for all drivers. Except for of recidivism data. The first study evaluated the professional drivers the legal alcohol limit is set at a initial and general DI course developed by Vias BAC of 0.2 g/l. Next to this, the legal framework in institute in 1996 by examining drivers between 1997 Belgium distinguishes between a main sanction and 1999. The results show a decrease in the (e.g., a fine) and an additional sanction (e.g., driver recidivism rate of 7%. However, that decrease was license withdrawal) (Hoet, 2013). An argument for not significant in the first effectiveness study. an additional sanction is that a main sanction doesn’t Nevertheless, there was a trend that participants of take the context of the offences and the personal the DI course reoffended less often and slower than circumstances of the offender into account (Hoet, participants of the comparison group. (Vanlaar, 2014). Moreover, sanctioning drivers with a Kluppels, Wiseur, et al., 2003). The second, more traditional sanction (i.e., a fine or an imprisonment) recent study found a decrease in recidivism of 43% has been shown to be ineffective in preventing by following convicted drivers between 2012 and recidivism (Af Wåhlberg, 2011) and stricter 2013. This study has some limitations, such as the sanctions (i.e., higher fines) don’t imply a reduction small sample size, the limited observation period in recidivism (Elvik & Christensen, 2007). and the inclusion of only one juridical district In respons to the findings above, Vias institute (Waeyaert, 2017). Therefore, the results should be developed alternative sanctions that aim to change interpretated with caution. the behavior of offenders. Alternative sanctions 1.4. Research design function as a substitute for the main sanction (Kluppels, 2018). Within the Belgian legal An experimental design is considered to be the best framework1, the main and additional sanction can be scientific method to measure the effectiveness of an suspended. One of the conditions to suspend the intervention. In an experimental design, participants main sanction is to attend a sensibilization or driver are randomly assigned to the experimental group improvent course (DI).2 DI can be defined as an (i.e., drivers that participated in a DI course for DUI educational measure whereby the offender of alcohol) or to the comparison group (i.e., drivers expierences a learning process. The main objective that were sanctioned with a traditional sanction). In of this learning process is situated on the level of a juridical context, it is not possible to randomly insight, increasing risk perception and becoming impose a certain sanction to offenders. Therefore, a aware of one's own behavior and attitude towards quasi-experimental design (i.e., a static group traffic' (Kluppels, 2018, p. 13-14). comparison) is usually applied in previous research on the effectiveness of DI as well as in the current 1.3. The effectiveness of driver study. Using existing groups implies that differences improvement found between the experimental and comparison Usually, recidivism data are used to evaluate the group can not be unambiguously explained by effectivesness of a DI course. A systematic literature participation in a DI course. review, in which six studies (mainly from the USA) A matching procedure ensures that both groups are were included with recidivism for DUI of alcohol as comparable regarding important characteristics outcome measurement, found a decrease in related to redivism. Hence, the problem of non- recidivism in five out of the six studies. The decrease random allocation to research groups can be in recidivism ranged from 11% until 61% (Miller, partially eliminated. Curtis, Sønderlund, Day, & Droste, 2015). In these studies, drivers sanctioned with a traditional 1.5. This study sanction and drivers participated in a DI course were Since the first effectiveness study in which followed up by measuring reoffences of DUI of recidivism was used as an outcome measure in alcohol. Drivers that followed a DI course Belgium by Vanlaar and colleagues (2003), the reoffended from 11% until 61% less for DUI of number of drivers that followed a DI course are alcohol compared to traditional sanctioned drivers. substantially increased (Nieuwkamp & Slootmans, Recently, a meta-analysis on the effectiveness of DI 2019). Even more important is that the content of the for DUI of alcohol found evidence for a decrease in DI course is significantly evolved since the first 1https://www.ejustice.just.fgov.be/cgi_loi/change_lg_2.pl?langu 2https://justitie.belgium.be/sites/default/files/downloads/201803 age=nl&nm=1964062906&la=N 21_jp_15.pdf 19 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia courses in 1996. In the initial phase of the DI course, 4. Drivers who followed a DI course will have there was only a general DI course established for all a lower BAC in the case of recidivism for types of traffic offenders. Whereas, nowadays there DUI of alcohol compared to offenders is a specific module for DUI of alcohol traditionally sanctioned. implemented. The current study examines the effectiveness of the module for DUI of alcohol. As 2. METHOD in the previously mentioned studies, recidivism is The five matching criteria related to recidivism in the main outcome measure for effectiveness of the the current study are the following: DI course. To determine the effectiveness, two groups of drivers who committed an offence for DUI 1. Gender; of alcohol are compared. One group was sanctioned 2. Age; with a traditional sanction (e.g., a fine or an 3. Blood alcohol concentration (BAC); imprisonment), the other one with an alternative 4. Period of conviction; santion (i.e., DI course). After following the DI 5. Type of prior convictions course or after the conviction, we observed whether 2.1. Participants the drivers reoffended. It is expected that drivers who followed the DI course will reoffend less (in The experimental group was composed using time and frequence) compared to the drivers participant lists of the DI department of Vias sanctioned with a traditional sanction. institute. 2241 offenders followed a DI course in Belgium between 2010 and 2014. 303 offenders that 1.6. Research questions and hypothesis completed a DI course for DUI of alcohol between The main research questions of the present study are 2010 and 2014 were selected from two different as follows: juridical districts. To compose the comparison group, 303 drivers in the same juridical districts with 1. Does following a DI course for DUI of a traditional sanction for DUI of alcohol were alcohol lead to less recidivism compared to selected using the five matching criteria described a traditional sanction? above. 2. Is there an effect of the type of sanction (DI course vs. traditional) on the number of The juridical information for the data collection offences committed after the reference process had to be encoded manually using a database conviction (recidivism frequency)? of the Belgium court (i.e., MaCH database). The 3. Is there an effect of the type of sanction (DI Board of Attorneys General gave their approval to course vs. traditional) on the time between use this database for the current study. Although, the the reference conviction and the next database has been used by police courts in Belgium conviction (time until reoffending)? for several years, it is not very suitable for research 4. Is there an effect of the type of sanction (DI purposes (Nieuwkamp, Slootmans, & Silverans, course vs. traditional) on the trend of the 2017). BAC of the offender? 2.2. Follow-up period 5. Is it possible to identify different profiles of drivers, based on the information available For all offenders, it was checked whether they had in the file of the offender, that can benefit committed any new offence after the date of the last more from a DI course on the one hand or session of the DI course or the date of the conviction a traditional sanction on the other hand? for those traditionally sanctioned. Recidivism was In congruence with the previous research on this followed up until January 2018 for all the drivers in topic, we propose the following research the study. By doing this, it is guaranteed that all hypothesis: offenders are followed for at least three years after their reference conviction. 1. Following a DI course for DUI of alcohol reduces the risk of recidivism compared to 2.3. Statistical analysis a traditional sanction. In order to answer the research questions, a number 2. Offenders who followed a DI course have of statistical tests are performed. Basic statictical a lower recidivism frequency compared to tests, such as a chi-squared test and ANOVA are those with a traditional sanction. executed using the statistical software IBM SPSS Statistics 25. More advanced statical tests, such as a 3. We expect that the time to reoffending is Cox regression analysis are performed in the longer for offenders who followed a DI statistical program R. This regression analysis, also course compared to offenders with a known as a survival analysis, examines for each traditional sanction. driver how much time there is between the reference conviction and a possible new conviction. For more 20 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia information about this statistical method, Bijleveld recidivism rate of offenders that followed a DI and Commandeur (2009) can be consulted. course is 41% lower compared to offenders traditionally sanctioned. Most of the drivers 2.4. Characteristics offenders and reoffended for DUI of alcohol ( n = 151; 49.5%), comparability between the groups followed by reoffending for other traffic offences ( n At the time of the reference conviction, the average = 137; 44.9%). 17 drivers (5.6%) reoffended for age of the offenders was 34 years (SD = 13.1) with non-traffic related offences. See Table 3 for an a range from 18 until 71 years. The BAC ranges overview of the type of recidivism disaggregated for from 0.50 g/l until 4.44 g/l with an average of 1.76 the experimental (i.e., DI course) and comparison g/l. The majority of the offenders were male ( n = group (i.e., traditional sanction). 534; 88.1%) and 37% of the offenders had a criminal Table 3: Overview of the type of recidivism record at the time of the reference conviction for Comparison DUI of alcohol. Table 1 and 2 show a breakdown of Experimental group Type of group (DI Total these results for the comparison group (traditionnaly (traditional recidivism course) sanctioned) and experimental group (DI course) to sanction) verify the matching procedure. N % N % N % No 188 62.0 109 36.5 297 49.3 Table 1: Mean, standard deviation (SD) and statistics recidivism for the variables age and BAC disaggregated by DUI of 65 21.5 86 28.8 151 25.1 alcohol comparison group (C) and experimental group (E) Other Variable Group Mean SD Statistics traffic 43 14.2 94 31.4 137 22.8 offences Age C 34.90 13.30 F(1, 6023) = 1.53, All types of p = ns, η2 <.01 7 2.3 10 3.3 17 3.0 E 33.50 12.90 offences Total 303 100 2995 100 602 100 BAC C 1.75 0.60 F(1, 5934) = 0.42, p = ns, η2 <.01 E 1.78 0.60 The median of the recidivism frequency is 1 (SD = 1.62) with a minimum of 1 and a maximum of 12. Table 2: Percentage and statistics for the variables The recidivism frequency is statistically different gender and criminal record disaggregated by the between both groups. Drivers that followed the DI comparison group (C) and experimental group (E) course ( Me = 1,74; SD = 2.21) have a higher Variable Label Group % Statistics recidivism frequency compared to drivers C 88.1% (χ2(1, n = traditionaly sanctioned ( Mc = 1,20; SD = 1,10), F(1, Men 606) = E 88.1% 306) = 8.17, p = .005, η2 = .03. This result is Gender 1.00, p = C 11.9% contradictory with hypothesis two. Women ns) E 11.9% C 37.0% χ2(1, n = The median of time until reoffending is 481 days Yes with a minimum of 0 days and a maximum of 2463 Criminal 606) = E 37.0% record 1.00, p = ns C 63.0% days. As shown in Figure 1, fewer drivers No E 63.0% reoffended after following a DI course compared to drivers traditionally sanctioned. More specifically, after 1000 days, approximately 25% of the drivers Based on this, it can be concluded that the who followed the DI course reoffended. Whearas, experimental group is similar to the comparison approximately 50% of the drivers traditionally group with regard to the variables age, BAC, gender sanctioned reoffended. and criminal record (dichotomized). 3. RESULTS 3.1. Recidivism as measurement for effectiveness At the end of the follow-up period of minimum three years, 309 of 606 drivers (51.0%) reoffended. From the drivers traditionally sanctionted, 194 of 303 (62.8%) reoffended. In contrast to 115 of 303 drivers (37.2%) that followed a DI course. This difference is statistically significant ( χ2 (1, n = 606) = 41.2, p <.001, Cramer’s V = .26). Consequently, the 3The variabele age was unknown for three drivers. 5The type of recidivism was unknown for four drivers. 4The variable BAC was unknown for 12 drivers. 21 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia Based on a one-way ANOVA, the mean BAC level at the reference conviction was not significantly different between the two groups ( Me = 1.77, SDe = 0.58 ; Mc = 1.74 , SDc = 0.55), F(1, 592) = 0.63, p = 0.43, η2 <.01. Subsequently, it is important to note that there is a different trend between the two groups in the level of BAC at the reference conviction and the next conviction. More specifically, the BAC level of the traditionally sanctioned offenders declines with 7% after being traditionally sanctioned. Whereas, the BAC level of the drivers that followed the DI course slightly increases with 3% after completing the DI course ( Me = 1.83, SDe = 0.58 ; Mc = 1.74 , SDc = 0.55). Thus, the BAC level Figure 1: Time until recidivism of the next conviction for DUI of alcohol is significantly higher for offenders following a DI A Cox regression analysis measures the time until course compared to traditionally sanctioned recidivism to calculate the effect of DI on offenders, F(1, 181) = 4.17, p = .04, η2 = .02. recidivism. In the current study the hazard ratio (HR) represents the risk to reoffend. A HR more 3.3. Identification of profiles of drivers then one means that the risk for reoffending Finally, we examined whether it is possible to increases after following a DI course. Whereas, a identify different profiles of drivers that can benefit HR less than one means that the risk to reoffend more from imposing an alternative or traditional decreases. The type of sanction was included in the sanction. Herefore, the type of sanction (DI course Cox regression analysis. The result show that vs. traditional) is combined with a level of four other following a DI course reduces the chance on variables available in the file of the offender using a reoffending with 2.6 (1/0.38; [0.28-0.52]; p < 0.001) Cox regression analysis. These four variables are compared to when the driver received a traditional respectively age, BAC, gender and criminal record. sanction. Thus, a DI course reduces the risk of The reference level corresponds to offenders reoffending. traditionally sanctioned in combination with another 3.2. Trend in blood alcohol concentration value of a variable mentioned in the file of the offender. By doing this, the added value of the DI To answer research question four, we investigated course can be investigated. See Table 4 for an whether there is a difference in recidivism severity overview of the results. (i.e., a different trend in BAC related to the type of sanction). Figure 2 shows the trend in BAC between For the variable age, it can be observed that older the reference conviction and the next conviction in drivers have less chance to reoffend after following relation to the type of sanction (traditional vs. DI a DI course compared to younger drivers. In other course). It should be noticed that only the words, the older the offender, the lower the chance reoffenders for DUI of alcohol are included in this to reoffend after completing a DI course. Drivers analysis. with a BAC level of more than 1.50 g/l at the reference conviction have 2.36 less chance to reoffend compared to drivers with the same BAC level traditionally sanctioned. Regarding to gender, both men and women can benefit from a DI course. However, women who followed the DI course have 4.44 less chance for recidivism. Whereas, men have a 2.67 lower chance to reoffend compared to men who were traditionally sanctioned. Drivers with no criminal record who followed the DI course have 3.44 less chance to reoffend compared to drivers with a criminal record that were imposed with a traditional sanction. This effect is also noticeable, but to a minor extent, for drivers without criminal record that were imposed with a traditional sanction. The latest group has 1.48 less chance to reoffend Figure 2: Trend in blood alcohol concentration of compared with drivers with a criminal record drivers alternative sanctioned (orange line) and traditionally sanctioned. driver traditionally sanctioned (blue line) 22 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia Table 4: Overview of the type of sanction combined between the type of sanction and other variables was with age, BAC, gender and criminal record using a investigated. It could be stated that the effect of the Cox regression analysis DI course on recidivism for all types of offences is Exp(coef) more pronounced for women (4.44 lower chance to [lower p reoffend) than for men (2.67 lower chance to chance to reoffend) in comparison with men with a traditional reoffend] sanction. Next to this, the criminal record also plays Age category * type of sanction an important role. Drivers without a criminal record, 36+ * DI 4,14 *** 36+ * traditional NA 0,08 who followed a DI course, have a 3.44 lower chance 26 - 35 * DI 4,11 *** to reoffend compared to drivers with a criminal 26 - 35 * traditional NA 0,16 record traditionally sanctioned. 18 - 25 * DI 2,37 *** 18 - 25 * traditional (Reference) NA NA Although the results are broadly in line with the BAC * type of sanction findings from international review and meta- 0,50 - 0,99 g/l * DI 3,52 * 0,50 - 0,99 g/l * traditional NA 0,57 analysis (Miller et al., 2015; Slootmans et al., 2017) 1,0 - 1,50 g/l * DI 2,66 *** and follow the trend of previous Belgian studies 1,0 - 1,50 g/l * traditional NA 0,18 (Vanlaar, Kluppels, Wisseur, et al., 2003; Waeyaert, 1,50+ g/l * DI 2,36 *** 2017), the results of this study also differentiates 1,50+ g/l * traditional (Reference) NA NA from previous research. According to the results of Gender * type of sanction Women * DI 4,44 *** Waeyaert (2017) for example, the BAC level of the Women * traditional NA 0,07 reoffenders for DUI of alcohol decreased after Men * DI 2,63 *** completing the DI course. In contrast with the Men * traditional (Reference) NA NA current study in which a longer follow-up period is Criminal record * type of sanction used, we find that when drivers have completed the No criminal record * DI 3,44 *** DI course and reoffend for DUI of alcohol, their No criminal record * traditional 1,48 * criminal record * DI level of BAC was higher compared to drivers 2,50 *** criminal record * traditional (Reference) traditionally sanctioned. It seems that for an NA NA Note: ‘Exp(coef)’ indicates the lower chance to reoffend for important group of drivers, the DI training has a every interaction effect compared to the reference level. positive effect, but that for a minority of offenders, Significant differences with respect to the reference category are DUI of alcohol becomes even more severe. indicated by an asterisk (*); ***:<0,001; **: <0,01; *: <0,5. 4.1. Limitations 4. DISCUSSION It should be noted that recidivism in this study is In 2003, a first study in Belgium was conducted to limited to registered behaviour. In other words, we measure the effect of a DI course on recidivism for do not know the dark number of drivers who have DUI of alcohol (Vanlaar, Kluppels, Wiseur, et al., reoffended again for traffic offences. Furthermore, 2003). This study served as a guideline for the our recidivism results are only based on the MaCH methodology of the current study. More than 15 database used by the Belgian court until January years later and after some important developments 2018. in the content of the Belgian DI courses, a new effectiveness study was required in order to measure Finally, in this study, a quasi experimental design the effect of a DI course for DUI of alcohol on was chosen since offenders with a alternative or recidivism. traditional sanction are pre-existing groups of drivers. The validity of quasi experimental research The most important result of this effectiveness study is not guaranteed because the drivers were not is that following a DI course for DUI of alcohol is randomly assigned to a group (Choi, Kho, Kim, & an effective measure to reduce recidivism for Park, 2019). different types of traffic offences. More specifically, the recidivism rates of the group of drivers who 4.2. Future research followed the DI course was 41% lower in When databases are becoming available for research comparison with traditionally sanctioned. purposes, future research should focus on measuring Furthermore, an analysis of the risk factors for the effect of DI courses on recidivism by using a recidivism shows that drivers who followed the DI bigger sample. Hereby, it can be possible to course have 2.6 less chance to reoffend compared determine the effect of a DI course on different types with traditionally sanctioned drivers. Moreover, in of traffic offences, broader than DUI of alcohol. the case of recidivism, the group of drivers who completed the DI course reoffended later compared to reoffenders with a traditional sanction. In order to identify which drivers would benefit more from following a DI course, the interaction 23 YRS 2021 Ellen Boudry, Ricardo Nieuwkamp, Peter Silverans, Ward Vanlaar 15. -17. September CAN DRIVER IMPROVEMENT COURSES REDUCE RECIDIVISM IN BELGIUM? Portoroz, Slovenia REFERENCES [10] Irwin, C., Iudakhina, E., Desbrow, B., & Mccartney, D. (2017). Effects of acute alcohol [1] Af Wåhlberg, A. E. (2011). Re-education of consumption on measures of simulated driving: young driving offenders: Effects on recorded A systematic review and meta-analysis. offences and self-reported collisions. Accident Analysis and Prevention, 102, 248– Transportation Research Part F: Psychology and 266. Behaviour, 14, 291–299. [11] Kluppels, L. (2018). Educatieve maatregelen [2] Bijleveld, C. C. J. H., & Commandeur, J. J. F. voor verkeersovertreders. Het nut van leren als (2009). Multivariate analyse. Een inleiding voor straf. Brussel, België: Politea. criminologen en andere sociale wetenschappers. Den Haag: Boom Juridische uitgevers. [12] Leskovšek, B., Goldenbeld, Ch. (2018). Influenced Driving – Alcohol, European Road [3] Brion, M., Meunier, J.-C., Pelssers, B., Leblud, Safety Decision Support System, developed by J., & Silverans, P. (2019). Alcohol achter het the H2020 project SafetyCube. Retrieved from stuur : de situatie in België – Nationale www.roadsafety-dss.eu gedragsmeting “Rijden onder invloed van alcohol” 2018. Brussel, België: Vias institute - [13] Miller, P. G., Curtis, A., Sønderlund, A., Day, Kenniscentrum Verkeersveiligheid. A., & Droste, N. (2015). Effectiveness of interventions for convicted DUI offenders in [4] Caird, J., Lees, M., & Edwards, C. (2005). The reducing recidivism: a systematic review of the Naturalistic Driver Model: A Review of peer-reviewed scientific literature. The Distraction, Impairment and Emergency American Journal of Drug and Alcohol Abuse, Factors. IDEAS Working Paper Series from 41(1), 16–29. Retrieved from RePEc. Retrieved from http://www.tandfonline.com/doi/abs/10.3109/0 http://search.proquest.com/docview/169817851 0952990.2014.966199 8/ [14] Slootmans, F. (2020). Statistisch rapport 2020. [5] Carnegie, J. A., Strawderman, W. E., & Li, W. Verkeersongevallen 2019, Brussel, België: Vias (2009). Study of Recidivism Rates among institute – Kenniscentrum Verkeersveiligheid Drivers Administratively Sanctioned by the New Jersey Motor Vehicle Commission. [15] Nieuwkamp, R., & Slootmans, F. (2019). FHWA NJ-2009-019. Retrieved from: Statistisch rapport 2019. Handhaving: controle https://www.nj.gov/transportation/refdata/resea en sancties. Brussel, België: Vias institute - rch/reports/FHWA-NJ-2009-019.pdf op Kenniscentrum Verkeersveiligheid. 28/01/2019. [16] Nieuwkamp, R., Slootmans, F., & Silverans, P. [6] Choi, Y.-Y., Kho, S.-Y., Kim, D.-K., & Park, (2017). Verkeersrecidive nader bekeken: B.-J. (2019). Analysis of the duration of Verkenning van een steekproef van compliance between recidivism of drunk gerechtelijke dossiers uit Leuven en Mechelen. driving and reinstatement of license after Brussel, België: Belgisch Instituut voor de suspension or revocation. Accident Analysis & Verkeersveiligheid - Kenniscentrum Prevention, 124, 120–126. Verkeersveiligheid. [7] Goldenbeld, C., & Twisk, D. A. M. (2009). [17] Slootmans, F., Martensen, H., Kluppels, L., & Verkeersovertredingen, veelplegers en Meesmann, U. (2017). Rehabilitation courses as verkeersonveiligheid. (R-2009-7). Stichting alternative measure for drink-driving offenders. Wetenschappelijk Onderzoek European Road Safety Decision Support Verkeersveiligheid SWOV, Leidschendam. System, developed by the H2020 project SafetyCube. Retrieved from www.roadsafety- [8] Hoet, P. (2013). Werkbare probatie en dss.eu on 11/01/2019. herroeping. In P. Traest, J.-F. Leclercq, & E. Goethals (Eds.), Amicus Curiae. Liber [18] Vanlaar, W. (2018). Methode ter evaluatie van amicorum Marc De Swaef (pp. 197–219). de Vias Driver Improvement cursussen voor Antwerpen: Intersentia. Retrieved from bestuurders onder invloed van alcohol. Ottawa, https://www.jurisquare.be/en/book/978940000 Canada: Traffic Injury Research Foundation 4306/werkbare-probatie-en-herroeping/ (TIRF). Unpublished internal report. [9] Hoet, P. (2014). Vijftig jaar Probatiewet: [19] Vanlaar, W., Kluppels, L., Wiseur, A., & verleden en toekomst van de individualisatie Goossens, F. (2003). Leiden van de bestraffing. In CBR Jaarboek 2013-2014 sensibilisatiecursussen voor bestuurders onder (pp. 319–391). Mortsel: Intersentia. Retrieved invloed van alcohol tot een lager recidivegehalte from dan klassieke straffen? Panopticon, 4, 365–390. https://www.jurisquare.be/en/book/978940000 [20] Waeyaert, E. (2017). Stageopdracht Emiel 4498/vijftig-jaar-probatiewet-verleden-en- Waeyaert. Unpublished internship report. toekomst-van-de-individualisatie-van-de- bestraffing/ 24 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND ACCESSIBILITY POINT OF VIEW Hannes Delaere, PhD researcher Samyajit Basu, Senior researcher Imre Keserü, Professor Cathy Macharis, Professor Vrije Universiteit Brussel Pleinlaan 2, Brussel, Belgium Hannes.Delaere@vub.be, Samyajit.Basu@vub.be, Imre.Keseru@vub.be, Cathy.Macharis@vub.be ABSTRACT The impact of digital mobility services on the mobility landscape is growing and users are increasingly having more options to choose from every day. There is however another side to this growth, the side where vulnerable-to-exclusion groups do not have the possibility to equally benefit from these services. In order to bring changes that can extend the benefits of digital mobility services and applications equally to the vulnerable-to-exclusion groups, it is necessary to understand what the drivers and barriers are for the development and deployment of inclusive and accessible digital mobility services. In this study we have tried to explore those drivers and barriers by performing semi-structured interviews with stakeholders involved in the development and day to day operations of 10 different digital mobility services throughout Europe which served as case studies. As part of these case studies, a literature review and desktop research were performed, in combination with semi-structured interviews and a content analysis. Three main groups of stakeholders (developers, operators, policymakers and in some cases also user group representatives) were interviewed with a focus on the regulatory framework in effect and how inclusion and accessibility were taken into account. First, the semi-structured interviews were transcribed, and a qualitative thematic analysis was performed. After this, the collected experiences of the stakeholders were consolidated and discussed at an online workshop, involving the three main stakeholder groups. The main conclusions drawn from this research are that there is a multifaceted problem; clear communications, collaboration and the related co-creation are not present in most services, no mobility or development related information is shared, and although the concept and importance of inclusion were known to the interviewees, only very few services take action towards providing an inclusive digital service. Keywords: Inclusive digital mobility, vulnerable-to-exclusion groups, case studies, semi-structured interviews, qualitative thematic analysis 1. INTRODUCTION traffic jams, better routes, arrival times of public transport, location of shared mobility stations, etc. The rise of the internet and - shortly after - the emergence of the mobile phone with internet access, In his first two laws of technology Kranzberg says: have opened new possibilities for people to organise “Technology is neither good nor bad; nor is it their travel. Since these technologies have become neutral.” and “Every technical innovation seems to widely spread, personal mobility has changed require additional technical advances in order to dramatically: real time information has made it make it fully effective.”(Kranzberg, 1986). The first possible for people to adapt to the journey law simply says that the impact technological environment by having access to information about inventions have, depends on the way humans use the invention. The second law indicates that an invention 25 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW always needs additional technological development MaaS. The final part of this paper contains the so that it can be used effectively in our society. The discussion and conclusions. two laws can be applied to the current technological developments resulting in the creation of the digital 2. METHODOLOGY transport system. The digital transport system has a This research focused on the collection of data by significant impact on our society, but it has not been interviewing stakeholders involved in the used to its full potential. Kranzberg approaches the development or operation of digital transport problem from a strictly technological point of view, services, using semi-structured interviews. The but for the digital transport system there is need for a analysis is based on the case study method. For significant human impact factor. Technological socio-economic research, case studies are one of the adoption in combination with a change in behaviour principal means used to collect data (Bates et al., takes time. An example to clarify: according to 1998; Robinson et al., 2003). Eurostat data from 2019, on average 27% of the European inhabitants have no access to internet when ‘A Case study is an in-depth exploration from on the move, 40% have never ordered goods or multiple perspectives of the complexity and services using the internet and 14% have never used uniqueness of a particular project, policy, institution, the internet (Eurostat, 2020a, 2020b, 2020c). So, program or system in a “real life” context’ (Simons, developers of mobility services need to be aware that 2009). a lot of people do not have the skills, financial tools The case studies focus on the process of the or understanding of the digital system to use it. This deployment of the digital mobility services from the results in the current development and operation of a viewpoint of the different stakeholder groups. high impact service that cannot be used by a Concretely, with these cases we will try to understand significant number of people in Europe (Franckx & and create an overview of the drivers and barriers, Mayeres, 2016). requirements and needs developers, operators and In reference with Kranzbergs’ laws, there is need for policy-makers experience when attempting to understanding the process which led to the develop/deploy some form of digital mobility introduction of the new digital mobility services. To service. do that, the development process needs to be studied For this research three main stakeholder groups are with a focus on the stakeholders involved in the identified: developers, operators, policy makers. development process of the digital mobility services. Some user representatives were also interviewed to Three main groups are considered pivotal in this gain a better understanding of the different vulnerable process: developers, operators, policy makers. Each to exclusion users. of these groups can have a significant impact on the way digital mobility services are developed and by We have identified six steps in the methodology: the whom they can be used: what decisions are made, first step is the selection of the 10 deployment case how do these stakeholders see the involvement of studies. Secondly, a desktop study is conducted; currently excluded groups, what initiatives have thirdly, for each case study relevant stakeholders already been taken and what exactly are the barriers have to be identified. In step 4, semi-structured and drivers for the stakeholders to develop a more interviews (SSI) are developed and conducted with inclusive digital mobility service. the stakeholders. Next the interviews are transcribed and analysed using the qualitative thematic analysis In this paper we try to understand the process leading method. Finally, in step 6, the results of the case to the current digital transport services, how the studies are consolidated and discussed with the development and deployment took place and if that stakeholders. process has resulted in the exclusion of several vulnerable groups in our society. Therefore the main 2.1. Identification 10 deployment case research question in this paper is: “What are the main studies barriers and drivers during the development and deployment phase for an inclusive digital transport The case studies have been selected taking into service?” account the following criteria: This paper is structured in 4 sections: the  Inclusion of digital mobility services that are methodology, the results, the discussion and already being deployed or are promising in conclusion. In the methodology the selection of terms of their future development cases, related stakeholders and the data-analysis  Coverrage of a great variety of digital method are explained. In the results we discuss the mobility services so that we can highlight any four main digital transport service types: car-and similarities and differences in the barriers and ridesharing, micromobility and bikesharing, smart drivers logistics and the multimodal routeplanners and  Ccoverage of multiple cities or regions 26 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW  Involvement of a variety of stakeholders  How were the regulatory framework taken into effect? Therefore, we used the digital mobility services as  How was the service or application identified by Kedmi Shahar et al. (2020), as a starting developed/deployed? point and identified digitally driven e-scooter services, crowd logistics services, free-floating car- During the development of the questions we decided sharing, on-demand ridesharing and Mobility-as-a- that the use of semi-structured interviews (SSI) is Service (MaaS) applications as minimum categories preferable for this research. The language barrier, in that should be covered by the case studies. combination with unknown stakeholders and topics that are relatively new, which might require more A first selection of cases is based on input from elaborate explanations, make the SSI the more suited previous projects, personal knowledge and networks method. and by an online research to contact services that comply with the criteria described above. The first We have developed the interviews based on literature contact with a potential case took place in July 2020, (Goodwin, 2013; Pope, 2020; Schmeer, 1999; Wick, during the COVID-19 crisis, which possibly had a 2012). The questions were developed based on topics negative impact on the services willing to provide related to user involvement, the current regulatory help for this research. framework in effect, how a planning was realised and followed, how decisions were made, the barriers and 2.2. Identification of relevant stakeholders drivers that were experienced and other relevant In order to identify and contact the relevant topics. Depending on the type of stakeholder, the stakeholders, a definition for a stakeholder must be topics touched during the interview were slightly integrated within the method. A stakeholder in the different. singular meaning is: 2.4. Analysis of the interviews ” Any group or individual who can affect or is affected The analysis of the interviews consists of different by the achievement of the organization's objectives” steps. The overall method used is the thematic (Freeman, 2010, p.46) analysis method, as explained by Braun and Clarke For the identification of stakeholders, a stakeholder (2006, p. 79). The authors describe the thematic analysis was conducted. The stakeholder analysis is method as an independent qualitative method that can not a single tool used to identify stakeholders but be defined as “a method for identifying, analysing consists of multiple different methodologies (Crosby, and reporting patterns (themes) within data. It 1991). Two methods were used to identify the minimally organises and describes your data set in deployment case study stakeholders: the snowball- (rich) detail.” mapping method and the stakeholder-led For this research, NVivo analytic software is used, categorisation. For each of the stakeholder groups the which made it possible to analyse the interviews in a snowball mapping-method for identifying new structured and thorough manner. For a good stakeholders is initially used, while afterwards, the presentation of the results, verbatims are also used to stakeholders were asked to categorize themselves as strengthen certain claims made based on the input one of the stakeholder groups. Stakeholder mapping from the interviews that were validated during the provides an overview of the different stakeholders, workshop. but also gives an indication about the strength of the inter- and intra-stakeholder group relationships 2.5. Development of the thematic analysis (Brugha & Varvasovszky, 2000). codes For each of the cases multiple stakeholders are During the initial stage of the analysis, the desktop interviewed, their answers analysed using a thematic research was used to create a first summary of the analysis and the results discussed during an context in each of the cases. This provided some of interactive workshop. In total, 22 interviews were the initial information for the development of the conducted, 5 with developers, 8 with operators, 9 semi-structured interviews and for the codes used with policy makers and one with a user group within the thematic analysis. During the actual representative. thematic analysis some additional codes were added 2.3. Development of the interviews to complement the predefined ones. In Table 1, all codes are split up in two categories: “primary coding The data collected from the interviews includes topics” and “secondary coding topics” which are several aspects of the digital mobility services: respectively the initial topics selected for the thematic  analysis and the topics developed during the thematic How weas inclusion and accessibility aspects analysis itself, based on new information collected taken into account? during the analysis of the interview. 27 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW Table 1: Thematic analysis codes some general accessibility and inclusion-related Primary coding topics Secondary coding topics remarks are included e.g. a clear and accessible Regulatory framework Future ideas and concepts subscription system should be in place, the operator Inclusion aspects Financial aspects can also choose to make the cars available without Data-protection/-collection Business as usual (BAU) the need for a subscription (Belgisch staatsblad, and privacy development User involvement/ co- General drivers for 2016). creation development General barriers for 3.2.1. Interviews development Source: own data As mentioned above, car- and ridesharing services are the oldest types of shared mobility studied in this 2.6. Co-creation workshop project. This results in a regulatory framework that After the initial results from the interviews, an online is older and has been further developed and adapted workshop was organised to consolidate and discuss compared to other services. During the interviews it the first results from the stakeholder interviews with was noticeable that the developers/operators have external experts in mobility and logistics services been in the business for a longer time and that the consisting of transport and logistics service market is more stable. Because of that stability and operators, software developers, policy makers, the fact that carsharing is already quite popular, it has mobility researchers. In total, 36 experts participated been present in all major Belgian cities for quite some in the interactive workshop. The objective of the time. workshop was to present and discuss the insights Several aspects related to inclusion have already gathered in the case studies. The discussion was been introduced within carsharing-services, mostly mainly used to validate the information that was focusing on providing a service available in multiple collected from the interviews with the stakeholders. neighbourhoods within the city. Much fewer adaptations have been made based on digital 3. RESULTS inclusivity, financial inclusivity, etc. For Cambio, 3.1. Selection of cases and stakeholders this is different, their main ‘inclusivity’-aspect is embedded in the core of the service: the call service. The ten services that were selected to serve as case On their website they also explicitly mention they studies are presented in Table 2. The results are want to provide a service that is affordable for all presented in four different sections, based on the (Cambio Flanders, 2020). major categories in the digital transport system: first car- and ridesharing is discussed, followed by bike The interviewee from the Cambio operator also said sharing and micromobility, smart logistics systems that not only the digital exclusion of lower educated and finaly also multimodal routeplanners and MaaS. citizens is an issue to use their service, but general knowledge about carsharing is lacking. Especially Over the period of half a year, 22 stakeholder among the lower educated people, the knowledge organisations, linked to a mobility or logistics about shared or digital mobility was almost non- service, were contacted to participate in this research. existing. This shows that even though there is a lot of Finally, 10 suited cases were selected with 22 digital marketing for similar services, the vulnerable- interviewees willing to participate. to-exclusion groups are not reached. 3.2. Car-and ridesharing User involvement and the strongly related bottom-The topic of car-and ridesharing is the first of four up method for development of services could lead to topics and is covered by two case studies, Cambio a more inclusive service. One of the operators of the and Mobitwin, which respectively resulted in two and Mobitwin app, whose primary job was actually that of a social worker, said: “knowledge of vulnerable to three interviews. Cambio is a provider of station- exclusion groups is low among developers” based shared cars in Belgium. The service was first which introduced in Germany, but is also active in Belgium was confirmed by the developer of the application. since 2002. Mobitwin is a rideservice for elderly The issues of app development and co-creation are people and people who experience difficulties with not that easily solved according to the stakeholders their mobility. For Cambio, a developer/operator and because of several reasons: first, it is very hard for a policy maker from Brussels Mobility, the regional developers to reach vulnerable to exclusion groups, administration responsible for mobility were secondly, there is much difficulty to find enough interviewed. For the Mobitwin case, three interviews people to have a sufficient large group and thirdly, activating and persuading ‘vulnerable’ groups to took place with a developer, an operator and a user group representative. For the Brussels Capital region, participate in e.g. testing phases has proven to be a a framework was developed related to shared major issue. mobility. There are no specific demands in the framework related to inclusive digital mobility, but 28 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW For the data collection, protection & privacy of Next to bike sharing schemes, there are also plenty of different partners in a digital environment there are free-floating scooters present in the city such as Dott, several issues that can be addressed, according to the Lime and Bird. Bird was already present in the city stakeholders. First, it is important that the from 2018 untill the winter of 2019, after which they implications of data sharing are made available and retracted their scooters until November 2020. The understandable for all users. Elderly people do not company restarted their service in Brussels with 200 have the knowledge about the amount of private data e-scooters (Belga, 2020). Even though several other they share. Secondly, they are rather keen on personal providers have left the city, the scooters are still connection which led in some cases in sharing too widely used by more than 100 000 users, but there is much personal information. The operators from also a lot of criticism because of their impact on the Mobitwin complained in multiple cases about the public space and especially related to safety issues users having access to personal phone-numbers of the when used on sidewalks, as they often are voluntary drivers (which is sometimes allowed in (Wallemacq, 2019). case of emergency): 3.3.1. Interviews The Mobitwin app was tested among elderly people in 2019 and, after being introduced to the service, Although no actual bike provider was interviewed, many of them downloaded the app and used it a few the free-floating bikes have similar issues as the free- times, but afterwards actual use of the app was floating e-scooters. The experts in the workshop negligible. Many considered it too difficult with too considered their barriers and strategies to overcome many options. A completely different reaction came these barriers applicable to both e-scooters and from the social and healthcare workers in elderly shared bikes. homes, who considered the app a very useful tool The first aspect discussed is the way in which the which made the process for making reservations a lot regulatory framework is being developed in the two easier and faster. The operators from both services cases and how this might impact the digital said that there was need for better guidance and inclusivity of the service. Many European countries understanding of those specific groups. have developed some sort of regulatory framework 3.3. Bike sharing and micro mobility for shared mobility. The regulatory framework is often developed at a national level, but in many cases, For the micro-mobility related services, three also at a city-level (Eltis, 2020). Both for Brussels interviews were performed: two with stakeholders of and Lisbon, a regional policy has been developed. the HIVE e-scooters in Lisbon and one with a policy This has not yet been fine-tuned so there are still a lot maker for micro-mobility services in Brussels. HIVE of gaps, also related to the limited knowledge (about (part of Free Now) is a free-floating e-scooter service, use, impact, (dis)advantages) policy makers often started in Lisbon in 2018 and is part of the Daimler have about the types of services. Both have some group. They introduced 600 shared e-scooters in the form of regulations to increase the accessibility and city and are even expanding, combined with the inclusiveness of the services provided in their city, introduction of extra e-scooters, they also put a lot of but they are based on physical accessibility rather effort in close collaboration with other local than digital accessibility and inclusiveness. In organisations and especially with local policy makers Brussels and Lisbon, the regulatory framework was (Hinchliffe, 2018). The app can be used in 11 primarily developed for car- and ridesharing services; languages, with the Catalan language included. only later were micro mobility and bike sharing Lisbon is currently using a soft regulatory framework added. Although a lot of efforts have been made for in order to attract providers of shared mobility. the continuous development of the framework, the For the Brussels’ case, no provider of micro mobility Brussels policy maker stated that the framework is was willing to take part in the research. During the still not extensive enough and a bit outdated. interview with the policy maker, some questions Lisbon was one of the early adopters within the related to the operation of a micro mobility or shared shared and digital mobility market. The city chose a bike service were answered, since the Brussels soft regulation approach, which making it more Region also has its own public shared bike scheme attractive for operators, which resulted in a large ‘Villo’. Villo is a station-based bike sharing service number of them wanting to operate in Lisbon. This provided by the Brussels municipality. The service was also confirmed by the interviewed operator: “I currently offers about 5000 bikes stalled in 360 think the approach of Lisbon it just gives the benefit stations all over the Brussels Region. Blue bike is of the doubt for the companies to come” These types also present at all train stations in Brussels providing of regulations were combined with hard regulations station-based shared bikes. Villo and Blue bike are as the city’s parking regulations prohibiting vehicles the two best known providers of shared bikes, but to park on the sidewalks. When the local government several other providers have been present as well received complaints from elderly people and people over the years. 29 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW who are visually impaired regarding all the scooters and having a positive effect on the match of what is on the sidewalk, the city used parking regulations to being developed and what is needed, but also prohibit scooters to park on the sidewalk. admitted that this type of cooperation is still lacking in many cases. Due to other hard regulations, such as the parking regulations, which have effect on the way in which There are two main issues concerning the co-creation e.g. e-scooters are used in public space, the city can or info-events: first there is the composition of the still impose their will on the e-scooter providers. group knowing about and attending these events and secondly, many of these events, as described by the The Lisbon policy maker suggested a second reason Brussels policy maker, are rather info-events, where why soft regulations were introduced. There was a information that has already been decided is shared lack of knowledge about micro-mobility among the with social organisations (e.g. unions, local social policy makers. Hard regulations simply require more groups, neighbourhood groups, etc.). Certain groups knowledge about the sector and its effects on society in society (e.g. older people, migrants, people with and public space in order for these regulations to be limited education) are not as easily reached and as a effective and useful. In these cases the policy maker result, their opinion and issues are not heard. So, said it was better to use soft regulations. In Brussels these workshops will have a positive effect on the a stricter, license-based approach was adopted, but the Brussels’ policy maker stated that, many cities service, but many vulnerable groups are still overlooked. In Brussels some of these issues are (incl. Brussels) do not have the funding or personnel being tackled by involving social organizations in the to monitor if all the digital transport services info-events (e.g. private organizations, workers operating in the city comply to those regulations. unions etc.), policy makers hope that by introducing Strongly related to the regulatory framework is the these organizations to the development of the introduction of measures for a more inclusive services they will get wider support, and the service service. The main issue the interviewees mentioned developed will be better tailored to the needs of was the instability of the market and the fact that, different groups within society. although it evolves quickly, it is still a niche-product, According to the Brussels policy maker, the concept which leads to the providers/developers wanting to of data collection, protection & privacy will also develop the service in its totality, with only a (very) become an important matter for the creation of better limited focus on the digital inclusivity of the service. digital mobility services. The stakeholder stated: For operators, developers and policy makers the “Well, it's (the exchange of data and protection) development of a good ‘general’ service, within a definitely a work in progress, this trust issue between reasonable regulatory framework that is financially private sector and public authority, and public stable is currently the main goal. Policy makers did transport operator as well, actually, it's definitely acknowledge that research into more inclusive something that we have to address.” One of the major services is needed, but as stated before, it is currently arguments against intensive data collection given by not possible due to lack of knowledge and financial the Lisbon operator was: “… we have a less pressure resources. It was considered as a failure of policy by to collect massive data. And, I think it's good because both policy makers. Although in general no massive it just takes a very big risk in terms of data attempts or regulations are made to change this on a management.” The same issue comes up in each of large scale, some financial-inclusivity aspects are the interviews, data-collection is hard, the advantages considered, such as 30 min free e-scooter time for are not always very clear, lack of trust, lack of inhabitants of Lisbon with limited financial means, knowledge, not enough resources for specific data hoping this would show the opportunity micro- collection. When asked for a vision on future data mobility can provide for these people. Compared to collection and especially management, the operator the shared car market, it shows that financial and did not think intensive collection of data (e.g. data market stability leaves room for more inclusive about vulnerable groups, specific people, areas etc.) development. It was mentioned by stakeholders for will become the standard. This vision was not shared both topics that if financial or market stability is not by the policy makers, they see a lot of potential in present, not many developers and/or operators are data collection and management. willing to spend a lot of money on inclusiveness. 3.4. Smart logistics services User involvement and communication will result in better relations between the developers, operators and Two types of logistics are studied. First there are stakeholders, which has a positive impact on the smart lockers, in Valencia and the ‘Mobile Lockers’ service. In both cities, information events are in Flanders, and the delivery of goods using organized by the local authority for people to learn (cargo)bikes in Madrid. Logistics services struggle about the new services and their digital aspects, and with several aspects related to digital inclusion (e.g. when asked, all stakeholder groups interviewed how do blind people access lockers, how does consider the input from end-users as being important someone without a smartphone order food and how 30 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW can last mile package delivery become lucrative, but 3.4.1. Interviews also available for all?). For logistics services, the development of a legal For the Citypack lockers in Valencia a city framework is less clear compared to the mobility representative, who was involved in the deployment cases. There have been problems with the status of of the lockers was interviewed. For Mobile Locker, riders, their payment and the social and job protection their CEO responded the questions from both a they enjoyed while working (Gómez, 2020). developer and an operator point of view. For the Although some form of legal framework is present in Coopcycle case in Madrid, three people were some of the cities where these services are active, interviewed, a developer, a policy maker and an none of the interviewees could give a clear answer on operator. The Coopcycle case is a cooperative the content of that framework. The Madrid policy service, focusing on food delivery using maker claimed that there was a plan for a decent (cargo)bikes. Some of the issues related to the regulatory framework to protect the riders but could delivery of food in cities, which has grown not go into detail about the content or when this exponentially over the years, were dominating the framework might be ready. Except for a potential news for quite some time, especially related to the regulatory framework to protect the drivers, there is Coopcycle case in Madrid. The Coopcycle service no such framework to protect the customers, no rules was developed in 2017 as a response to the initial were mentioned in order to be able to access problems that surfaced related to the ‘self- vulnerable to exclusion groups. employment’ status of the drivers for other good- delivery services. La Pajara started working on their Ini the madrid case, the applications and service itself were developed by ‘activists’, within Coopcycle with first collaboration projects and helped customers to fight against the working conditions of a major focus on developing fair services for which mainstreaming platform models (security, abuse, they are active on two levels: they are lobbying on the holidays, sick pay) inside national and international European level, but also with local government. networks and associations. For this case a developer, Because of its foundation within the political activist- operator and local policy maker were interviewed. scene, Coopcycle have developed their own framework to protect their riders and customers. The second case study is also based in Spain and was developed as part of the European Horizon2020 The Belgian policy makers did not mention the project SPROUT in the city of Valencia. The existence of a legal framework related to good development and operations are performed with delivery by lockers, they provide a rather supportive ‘Ferrocarrils de la Generalitat Valencian’ (local role without too much interference in the way the public transport agency) and ‘Fundación market is run. The developer of the Mobile Locker Valenciaport’ (the local seaport organisation). The system claimed that, with the current status of the market, “The government should be inv use of lockers, provided by Citypack, for the last-mile olved as less as possible.” but also stated that the lack of profits in delivery give operators more flexibility as it decreases the operational costs and reduce the failed the last mile is caused by the prices that are too low, home deliveries. To the city level, it reduces the which could be adapted with a regulatory framework urban freight vehicles occupying the public space and developed on a nation-level. Inclusion wise there are traffic jumps. In this project, public infrastructures some general rules, related to access of screens in such as metro stations opened up the adoption of this public places for people in a wheelchair, but no service to any commuter using the metro. recommendations of rules have been drafted from a digital exclusion point of view. The third and last logistics service is a Flemish private locker service named Mobile Locker, Generally speaking, no clear framework has been providing very large variety of smart lockers. The developed, although several of the interviewees idea was developed in 2012 and has grown to become stated there were plans (e.g. policy makers) for a a high-end provider of lockers for all types of events regulatory framework in which only very few aspects and locations. Contrary to many other providers, they related to digital inclusion were mentioned. That lack can provide tailored lockers developed in can be explained by the urge to create a decent collaboration with the customer, which can be a service, which is in the first place accessible to the (local) government, an organization or a company in broader public. Secondly, and especially related to need of solving issues with their package flow. These lockers, the profit in last and first mile delivery is lockers are permanent and provide, next to the negligible, so before a profit can be made on these logistic aspect, also some extra services, such as deliveries, not many investments are going to be WiFi, cellphone charger on solar energy, etc. For the made for a more (digital) inclusive system. use of a locker, not even an application is needed. Just Often closely related to the lack of a regulatory by scanning the QR (quick response) code, someone framework is the limited factors promoting (digital) can use the service. inclusiveness. Efforts in order to develop a digital 31 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW inclusive service are very limited from a policy point During the interviews, co-creation was only of view, in both the Spanish and Belgian cases (and mentioned a few times even though it is embedded in this is applicable in Europe) there were some efforts the development of the Coopcycle service, which is towards inclusion, mostly related to the accessibility co-owned by everyone participating within the of buildings and services as regulated by the business and therefore is submitted to a lot of government (e.g. screen displays should be different viewpoints. The user involvement part is accessible for people in a wheelchair). mainly based on the input from the riders, who were underrepresented before this service was developed. Initiatives for a more digital inclusive system are not Before and during the development of the lockers in mandatory and not really promoted as well. the city of Valencia, potential users were asked to Coopcycle, being a quite inclusive organisation, has provide input, but no further comments were made some aspects they are working of in order to create a about this during the interview. more inclusive service. Reservations are possible using the phone, cash can always be used for payment La Pajara have a very close relationship with the and when neighbourhoods are too far away to make developers of their service (Coopcycle) which results delivery profitable, they try and work out a solution. in good communications and faster response to This was corroborated by the Madrid policy maker, errors. “It is a good thing that the more users you who acknowledged the problem, but except for have, the more they validate it (the app), the more it acknowledging the issue, nothing much is happening is tested and the more errors are seen.” As was said to solve it. Based on the inputs from the intrviewees above, collaboration with other organisations is it can be concluded that there is a structural problem, pivotal to the business idea of Coopcycle. For the rather than only the lack of funds and initiatives. development of the service, they worked together with an organisation specialised in promoting The lockers in Valencia are developed by the public cooperatives. Their vision on this form of railway organisation and are therefore located near cooperation was very positive and gave them the time train stations. The link between these stations and the to learn, create a network and discover alternative lockers creates a more inclusive story. People with ways of thinking and working. Related to the lockers, more limited mobility due to not having a car have co-creation is very limited or even absent. The more difficulty to combine trips and to carry goods lockers from Mobile lockers are developed with input and food home after work for example. The during the prototyping, if the project is big enough. combination with lockers, where both food and goods could be delivered, not only provides the users with The fourth main aspect is data collection, protection benefits, but also reduces stress on the public & privacy, which is very different for the bike transport. If fewer (connected) trips related to delivery and the lockers. In the case of the lockers, shopping need to be made, there are less people using the collected personal data in both cases is rather the public transport. Also, public transport does not limited and mostly focused on the use of the lockers. favour one group over other, so the lockers are The main goal is to advance the way in which the available to a broad spectrum of society. Similar to service works, which is done in two ways. First, the one in Madrid, the Valencia policy makers the general data is collected from the locker (time, use, acknowledged that issues about excluded groups are which types of lockers, duration etc.), which is used known, but not much action is undertaken at this by both providers, and is especially useful when the time. developer is in charge of the actual locker (e.g. when it is not bought by a government for example, but is For the smart logistics services, there is still a lack of installed on their own initiative). Secondly, and this rules, efforts and methods to make the services more was mainly the case for the lockers in Valencia, a lot (digitally) inclusive. Most of the efforts made of studies were performed to find out what the reach towards this goal are own initiatives from the of the lockers was, meaning the area that was companies, rather than regulated by policy makers. serviced by these lockers, to evaluate how profitable User involvement such as co-creations could be a they were. The policy maker for Valencia said: potential solution for some of the earlier mentioned “Since it was a data intensive project, but it was only issues, keeping in mind that for a more elaborate and me who used and had access to that data and it was better regulated framework, user involvement solely for their benefit, it was not going to be published or with potential users will not suffice. A bottom-up anything.” Which led to the users providing a lot of approach, combined with close collaboration with information about the acceptance level of the lockers. developers, operators and especially the policy Coopcycle was not that busy with data collection, makers (both at local level and beyond) are necessary their focus was primarily on the creation of a fair to achieve an inclusive regulated service. The issue system for the drivers and customers. here, is that during the development of the services, certainly the lockers, a very limited amount of co- creation is part of the development process. 32 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW 3.5. Multimodal routeplanners and MaaS environmentally friendly) from the user, presents a multimodal route for commuters. While they are For this part, focusing on multimodal routeplanners currently only working with companies or and Mobility as a Service or MaaS systems four cases organisations, there is already a beta-testing version are studied: the BKK FUTAR app Budapest, the for users available on the website. An application is HVV Switch app, used in Hamburg, the HSL MaaS in development but is not yet available. One of the app from Helsinki and Jeasy, a new MaaS developer strong suits of the Jeasy service is the integration of active in Belgium. MaaS in the Belgian mobility budget. The first case is a multimodal application for the city of Hamburg. It is a mobile app, named Switch, 3.5.1. Interviews operated by Hamburger Hochbahn AG and the There are more barriers/drivers compared to the other ‘Hamburger Verkehrsverbund’ (HVV), the local topics because of the recent and fast development of Transport Association that makes moving around MaaS-like systems, the large amount of data and Hamburg easier and faster. With this app it is possible information that is needed and because of strong to buy HVV tickets and book your ride with MOIA debates about the development of new tools. Another (local car sharing service) with just one registration. point of discussion is the relationship between public HVV is a company that coordinates the public and private initiatives. The cases also show that both transport system in and around Hamburg, Germany. private and public organisations have introduced The full application was only available from 2020 multimodal routeplanners or MaaS or are at least on onwards, in German and English. The app does not their way to do so. have multiple digital payment options, the only one is PayPal. The regulatory framework for routeplanners or MaaS varied strongly in different cities and between The second case study took place in Budapest the organisations. None of the operators mentioned a (Hungary) where the BKK FUTÁR service was strict regulatory framework which they had to take studied. The application is a multimodal route- into account. For the three (semi-) public planning service mainly focusing on tram- and bus- organisations (BKK FUAR, Switch and HSL) there use, but it can be used also for subway. For each of were however some general rules, mostly focusing on the modal options, real-time information has been providing a service for ‘everyone’, ‘the general or integrated in the application and webservice, a main broad public’, although this was not always difference with the ‘Switch’-app is the lack of an supported by other claims about their target groups: integrated payment option. The app is available in 2 “…with 6 type of tickets you get 95% of all people languages, English and Hungarian, so for tourists this satisfied. Their approach is in line with the comments might cause some issues. A web-based version of the about the lacking regulations about the expectations service is available as well, leading to more people from the city, regions they are operating in, the lack having access to the service. The application provides of (inter)national regulations about data sharing and real-time information about the mobility services and the lack of standards. HSL tries to solve part of this their stops, also information about ticket vending issue by releasing licenses for limited time with some machines and kiosks is available. aspects of control (similar to KPI’s). The third service is HSL (Helsingin Seudun Liikenne Inclusion seems to be quite the challenge for or Helsinki regional transport) Public transport multimodal routeplanners and MaaS-systems. None application used in the city of Helsinki and the of the interviewees could indicate that they were surrounding areas. Helsinki is a leading example for considering specific groups of people vulnerable to multimodal and sustainable urban mobility and the exclusion. However, the developer from Jeasy HSL app is a good example of how apps can be approached inclusive design in another way, as part developed and how other services are integrated: the of a stepwise evolution of their service, which is, app contains fully integrated real time information from a financial point of view a logical choice: “…it and payment systems (debit- or credit card or by is difficult to work with those groups (vulnerable to using your phone bill). The application is not a full- exclusion), as you said, we should include them. But, option MaaS app because of its focus on public to make a service successful, you start with the transport (e.g. in comparison with Whim from MaaS biggest group.”. Also, the stakeholder considered it Global). The application is available in English, very important to collect information about excluded Finnish and Swedish. groups and to broaden their knowledge about the The last service is the multimodal routeplanner Jeasy excluded groups and how they can be approached. which is currently only available in the Belgian B2B It’s also clear that different stakeholders were market, focusing on multimodal commuting. Jeasy focusing on different aspects of inclusion, Jeasy, has developed an application that, based on working in the B2B market, wants all employees to preferences (e.g. shortest route, cheapest, benefit from the advantages of the mobility budget, 33 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW in combination with MaaS, which could replace All of the stakeholders mentioned similar aspects company-cars, so they were interested to see how about user involvement, if it was present, it was generational differences would impact the mostly in a very limited capacity, and could rather be preferences people had. defined as collaboration between companies, mostly in the form of tenders or assignments. Multiple User involvement is present in a very limited manner organisations also mentioned their interest to provide for the development of routeplanners and MaaS- access to those people who were currently excluded. systems. According to the Helsinki policy maker it is As confirmed by Jeasy: because of the newness, a very competitive market where both private and competition and volatility of the market the main public organisations are active. The first difference focus in on the ‘masses’ in order to create a viable between both is the development. For Jeasy it is an ‘in service, which can later on, stepwise, become more -house’ development, for the other services the digitally inclusive. development and coding itself is carried out by specialised ‘third-parties’, which in some cases Data collection, protection & privacy are very results in the developers having almost no input relevant in the case of multimodal routeplanners and contentwise, next to only developing an app as MaaS-systems, similar to data collection, protection ordered by the operator, and in other cases results in & privacy when providing door to door logistics the lack of collaboration in the app development. It is services. In order for these services to work at their clarified by the HSL policy maker why their best, a lot of personal information is needed: involvement is kept to a minimum: “Because they preferences, skills, needs, expectations etc. This was can be different people this year, different people also acknowledged by the stakeholders as described next year.”. Although this might change in time. In above. There is quite some data collection from the all the cases it was also clear that co-creation with users in order to have a better tailored service, and, potential end users was not really considered, rather compared to the other services seems to be analysed they did interviews and surveys to see how many more in-depth. people were using a service, how often and how satisfied people were using the service: “We Another thing pointed out by the HSL policy maker is the need for shared data and information about concentrated essentially on a few co-workers and routeplanners and MaaS-systems. This was also then on the enterprises which were involved anyway strongly confirmed by the Jeasy developer. When in the development, thus not necessarily were regular asked about European guidelines, almost all users involved directly in the app development.”. interviewees confirmed that if the EU wants decent Similar answers and remarks were made by the other MaaS systems, there is a need for standardised data stakeholders, Jeasy however, did claim they could provide API’s sharing. Currently, data sharing is already present on (Application Program Interface) to a minimal level and is performed according to the other organisations (e.g. representing vulnerable user EU-regulations, but in many cases this framework groups) in order for them to use the MaaS application needs to be elaborated. In the Hamburg case the in combination with the app for the vulnerable users. importance of data protection was also confirmed. It also becomes clear that the speed of development However, at the moment this is still too limited and of services is very different for the public service services are being developed in all countries and even compared to the private sector. In the first group there cities: “What I do see is that countries are working are more stakeholders, more boards and development alone, so they don't work with other countries and moves at lower pace compared to the private that within those countries, cities are working alone. initiatives which was considered a barrier by the Antwerp does not have the same expectations from developer from Jeasy. Nevertheless, the developer the MaaS players than in Brussels.”. considered the conversations with policy makers and city representatives as very useful. by the developer: “… every time we knocked on a door, we reached a discussion that was constructive, not always providing what we expected, but at least it was constructive, positive and so forth.” This gave the impression that the slower pace at which policies change related to these services had as a result that close co-creation was not a real possibility at the moment. For the public organisations in Helsinki, Hamburg and Budapest, there are many more conversations, discussions at hand before changes can be made. 34 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW Table 2: Synthesis results driver and barrers for an inclusive mobility serice Data collection/ Type of Regulatory Other relevant Service name Co-creation Inclusivity protection and service framework aspects privacy A regulatory Services are Knowledge about Only mobility Internal or third framework is unknown among vulnerable groups related data, party app Cambio present in most vulnerable to too limited with limited use, development Brussels cases, usually exclusion groups Depending on especially about influences on a national and not enough service different vulnerable to flexibility. level. accessible attempts: call exclusion users Users are The framework communication. center, cheap, Elderly people suspicious is present due to Rather info-events personal approach, have trouble towards new Car- & stability market and introductions. test-events. understanding technologies. ridesharing and is starting to Homogeneity of Digitally privacy related Lack of push towards ‘test-groups’: contacting elderly issues and don’t resources for Mobitwin more inclusive highly educated is hard understand research services, but this and white. Post- importance of towards is not development co- GDPR. inclusion mandatory in creation events are vulnerable case of digital important. people. inclusivity. Bike sharing HIVE Lisbon Regulatory Lack of funding Small steps are Data collection Impact on and micro framework is and instability of taken towards a is very limited public space is mobility regional and market make more inclusive and is not really major, affecting based on providers focus on service, but no used for elderly people, carsharing ‘easy to reach’ regulations are analysis. blind people framework, so users, in order to present to push Contradicting etc. not always be financially towards more vision on use of Young and Brussels suited to micro feasible. inclusivity. data: in detail or volatile market Mobility mobility. More frequent and Need for rather a general was severely Knowledge intensive smartphone and approach. impacted during about service communication credit card in Data is needed COVID-19 among policy between almost all cases. to address crisis. makers is too stakeholders and Link with public impact on public Main focus on limited. with users. transport and use space financial Both hard or In most cases of one payment revenue. soft regulations there is a lack of method are adopted to communication control micro with local mobility. authority. Smart La Pajara No national Co-creation Acknowledgement Only Locker Logistics Madrid regulatory during of need for more information providers services framework for development with inclusive services, relevant to showed no the protection of riders created but lack of funding delivery is intention for the couriers, more fair value an knowledge. stored, but not development of initiative by sharing. Similar to the used for any inclusive organisations All regulations, there other purpose. services, rather themselves. neighbourhoods are no rules to focus on the are provided for make the service financial without extra cost more inclusive. aspects, this Mobile No consensus Except for the If inclusivity is not Information was not the case Locker on need and location, there are introduced by the about for the public Citypack purpose of not that many developer/operator, efficiency/use of provider. Valencia regulatory measures towards it is not present in the lockers is Lockers framework an inclusive the service. collected and among different service. No used to find services collaboration with most profitable users or other location. organisations present. Multimodal HSL No relevant Isolated No measures were Fear that sharing Public service, routeplanners HVV Switch regulatory development mentioned by data will lead to open data & MaaS BKK FUTAR framework was simultaneously in operators or advantage for sharing with Jeasy present different cities developers. competitors other cities is Different Public Lack of trust in becoming more development organisations public services important. speed in private develop services handling data. Unfair vs. public for all inhabitants, competition organisations. but this goal is not between private Closer often reached. and public cooperation with Service still development users necessary unknown Source: Interviews and results from co-creation workshop 35 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW 4. DISCUSSION operators and developers. This aspect was observed among all developers and operators of the different Several topics were discussed, based on which a services studied and it was also confirmed by the number of common requirements, needs and user group representatives. If knowledge about user challenges of policy makers, operators and groups with disabilities is available, it is about the developers that are relevant for all types of services needs of people with physical disablilities. These are were investigated. These barriers and drivers are also often the groups that are not reached for related to the market position of the services, the information events. Another issue linked with this is regulatory framework in place, the integration of the lack of diversity among groups participating the service into the public transport service in info- or co-creation events. There is currently no network, the diversity of the vulnerable-to-or limited interest from the operators to explore why exclusion groups, knowledge about the certain groups of society do not use their services. vulnerable groups, the level of user involvement They are mainly interested in market expansion, (co-creation), the fast evolution of digital mobility focusing on target groups and areas that can provide services and communication and collaboration sufficient profit margins. between stakeholders. Combining data from multiple operators and other The market position of a service is defined by how sources (e.g. statistical databases of local established an operator is in a local market, i.e. how authorities) could expand the possibilities for the long it has been operating, how many users it analysis of users’ needs and for the identification of reaches, if it makes profits and what its position is non-users. Sharing of mobility data is a very compared to the competitors.In a more stable and sensitive topic. The fear for sharing data is twofold. predictable market, there is a long-term growth Firstly, many private providers of digital mobility do potential for service operators. This has been not trust private and public organisations in terms of observed by the interviewed car-sharing operators. sharing their data. Secondly, many providers are In contrast, if the market is volatile with many afraid they will lose their market share to players appearing and disappearing in a short time competitors when sharing information, so they all period, and with uncertain profitability, the focus of rather keep their data. This showed in the lack of the market players is short-term profit communication and especially collaboration maximalization that may exclude any accessibility between stakeholders. and inclusivity features. The level of user involvement (co-creation) in the The regulatory framework can set minimum development of new services remains a key issue. service requirements in terms of accessibility and The importance of involving (potential future) users inclusiveness. It can also contain requirements for in the development of applications and services has sharing service data, define key performance been mentioned as a key requirement by most indicators and require collaboration between stakeholders. However, actual co-creation, i.e. the complementary services. Current frameworks in the involvement of the users in the design of multiple case studies focus primarily on the regulations aspects of a service or application is limited. User related to operators keeping to certain rules e.g. involvement can be very diverse. From the case parking regulations, regulations related to use of studies about Coopcycle, HSL, Cambio and HIVE public space, speed regulations etc., the topics of we identified several approaches which can act as accessibility and inclusivity are not usually covered. drivers for creating more inclusive and accessible The regulatory frameworks are often adapted to just services:There are, however, also many barriers. A one type of digital mobility service, usually the one barrier to co-creation is that many potential users that was introduced as the first one (usually car have very limited knowledge about the service, sharing). The frameworks are then extended to other showing there is still a lack of decent introduction, types of services, but often not addressing the explanation and guidance for groups who have no specificities of these new services (e.g. e-scooters). access to the digital transport system. It is possible that for some groups it is very hard to make the Related to the regulatory framework, the transition from physical to digital services, as was integration of the digital service into the public proven in the Mobitwin case. transport service network may act as a major driver for accessibility and inclusion. Once a service is considered a ‘public service’, it needs to be 5. CONCLUSION available for all citizens, it needs to comply with We can conclude that there are a multitude of digital minimum accessibility standards and eventually also mobility services available in the current transport with a regulated pricing system. system. Although all four services are very different they struggle mostly with the same issues. The Knowledge about the needs and requirements of development of an inclusive service seems to be a vulnerable to exclusion groups is limited among 36 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW difficult challenge depending on and influenced by [9] Kedmi-Shahar, E., Delaere, H., Vanobberghen, many different factors. The carsharing services were W., & Ciommo, F. D. (2020). D1.1 – Analysis the most inclusive, both from a digital and a physical Framework of User Needs, Capabilities, point of view. This was mostly related to their Limitations & Constraints of Digital Mobility Services. 105. stability and the approach they take in providing a fair service. Both the micro mobility and logistics [10] Eurostat. (n.d.). Internet access and use services struggle with the concept of inclusivity, statistics—Households and individuals. which they claimed had to do with their financial Eurostat Statistics Explained. Retrieved 15 instability and with a lacking regulatory framework January 2021, from https://ec.europa.eu/eurostat/statistics- related to these topics. Similar to the carsharing explained/index.php?title=Archive:Internet_acc services, most inclusivity measures were developed ess_and_use_statistics_- as part of their business plan. The routeplanners and _households_and_individuals&oldid=379591 MaaS were not inclusive as they were still rather at the start of their deployment process and were still [11] Eurostat. (2020a). Digital economy and society statistics—Households and individuals. looking for their market share. The main aspect that Eurostat Statistics Explained. needs attention regarding these services is a https://ec.europa.eu/eurostat/ regulatory framework with specific focus on shared statisticsexplained/index.php/Digital_economy data and the management of that data. The digital _and_society_statistics_households_and_indivi transport system has a lot of potential for the future, duals but as was stated in the introduction, futher [12] Eurostat. (2020b). Individuals using mobile development of the services will be pivotal to devices to access the internet on the move provide its service to all groups in society. (Dataset No. tin00083).https://ec.europa.eu/eurostat/tgm/tabl REFERENCES e.do?tab=table&plugin=1&language=en&pcod e=tin0 [1] Belga. (2020). Deelsteps van Bird strijken opnieuw neer in Brussel. De Tijd. [13] Eltis. (2020). Overview of policy relating to e- https://www.tijd.be/ondernemen/transport/deels scooters in European countries. teps-van-bird-strijken-opnieuw-neer-in- https://www.eltis.org/resources/casestudies/ove brussel/10263011.html rview-policy-relating-e-scooters-european- countries [2] Brugha, R., & Varvasovszky, Z. (2000). Stakeholder analysis: A review. Health Policy [14] Franckx, L., & Mayeres, I. (2016). Future trends and Planning, 15(3), 239–246. in mobility: The rise of the sharing economy and https://doi.org/10.1093/heapol/15.3.239 automated transport-Annex A (Deliverable D3.3; Mind-Sets, p. 113). European [3] Bates, R. H., Greif, A., Levi, M., Rosenthal, J.- Commision. L., & Weingast, B. R. (Red.). (1998). Analytic narratives. Princeton University Press. [15] Freeman, R. E. (2010). Strategic Management: A Stakeholder Approach. Cambridge University [4] Belgisch staatsblad. (2016). Wet van 28 april Press.https://doi.org/10.1017/CBO9781139192 2016 betreffende de houdende voorwaarden 675 voor het gebruik van voorbehouden parkeerplaats aan operatoren van gedeelde [16] Gómez, M. V. (2020, September 24). Spanish voertuigen. Supreme Court rules food-delivery riders are http://www.ejustice.just.fgov.be/cgi_loi/change employees, not self-employed. EL PAÍS. _lg.pl?language=nl&la=N&cn=2016042817&t https://english.elpais.com/economy_and_busin able_name=wet ess/2020-09-24/spanish-supreme-court-rules- food-delivery-riders-are-employees.html [5] Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in [17] Goodwin, K. (2013, January). A Stakeholder Psychology, 3(2), 77–101. Interview Checklist. Boxes and Arrows. https://doi.org/10.1191 https://boxesandarrows.com/a-stakeholder- /1478088706qp063oa interview-checklist/ [6] Brussel Mobiliteit. (2017, December 14). [18] Hinchliffe, T. (2018, November 28). mytaxi Alternatieve mobiliteit. https://www.brussel.be Launches ‘hive’ E-scooter Pilot Program in /alternatieve-mobiliteit Lisbon—Portugal Startups. https://portugalstartups.com/2018/11/mytaxi- [7] Cambio Flanders. (2020). Voordelen | Cambio hive-e-scooters-lisbon/ autodelen | Vlaanderen. https://www.cambio.be/nl-vla/voordelen [19] Kedmi-Shahar, E., Delaere, H., Vanobberghen, W., & Ciommo, F. D. (2020). D1.1 – Analysis [8] Crosby, B. L. (1991). Stakeholder Analysis: A Framework of User Needs, Capabilities, Vital Tool for Strategic Managers. USAID Limitations & Constraints of Digital Mobility Implementing Policy Change Project, 2, 6. Services. 105. 37 Hannes Delaere, Samyajit Basu, Imre Keserü, Cathy Macharis YRS 2021 UNDERSTANDING THE DRIVERS AND BARRIERS IN THE PROCESS OF DEPLOYMENT 15. -17. September AND OPERATION OF DIGITAL MOBILITY SERVICES FROM AN INCLUSION AND Portoroz, Slovenia ACCESSIBILITY POINT OF VIEW [20] Kedmi-Shahar, E., Delaere, H., Vanobberghen, [25] Schmeer, K. (1999). Stakeholder Analysis W., & Di Ciommo, F. (2020). D1.1 – Guidelines. 48. AnalysisFramework of User Needs, [26] STARS. (2018). The influence of Capabilities, Limitations & Constraints of socioeconomic factors in the diffusion of car Digital Mobility Services (Deliverable D1.1; sharing Deliverable D4.1 (Deliverable 4.1; INDIMO, p. 103). Shared Mobility OppoTunities And Challenges [21] Kranzberg, M. (1986). Technology and History: For European CitieS, p. 243). European ‘Kranzberg’s Laws. Technology and Culture, Commision. 27(3), 16. [27] Wallemacq, T. (2019, May 7). De verborgen [22] Pope, L. (2020). Stakeholder interviews: What kant van elektrische deelsteps | Metro. De to ask and how to ask it. Verborgen Kant van Elektrische Deelsteps. https://gathercontent.com/ https://nl.metrotime.be/2019/05/07/must- blog/stakeholderinterviews-what-to-ask-and- read/de-verborgen-kant-van-elektrische- how-to-ask-it deelsteps/ [23] Robinson, J., Acemoglu, D., & Johnson, S. [28] Wick, A. (2012). Six Effective Elicitation (2003). In Search of Prosperity: Analytic Questions to Ask Your Stakeholders. BA Times. Narratives on Economic Growth. Princeton https://www.batimes.com/articles/six-effective- University Press. elicitation-questions-to-ask-your- stakeholders.html [24] Simons, H. (2009). Case Study Research in Practice. SAGE Publications, Ltd. https://doi.org/10.4135/9781446268322 38 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Rins de Zwart SWOV – Institute for Road Safety Research Bezuidenhoutseweg 61, The Hague, The Netherlands rins.de.zwart@swov.nl ABSTRACT The introduction of connected and automated vehicles (CAVs) has far-reaching impacts on the traffic system and society. The Levitate project (EU, Horizon2020, levitate-project.eu) aims to forecast those impacts. Road safety is one of the relevant areas of impact that is analysed in the Levitate project. CAVs influence road safety directly by replacing the human driver, and indirectly due to secondary impacts such as modal split and rate of travel. In order to forecast the road safety impacts of an increasing penetration rate of automatic vehicles several methods are combined. Microsimulation is used to model traffic behaviour of automated and human driven vehicles in different road networks. Vulnerable road users (pedestrians and cyclists) are not included in the microsimulation model but are affected by the introduction of connected and automated vehicles. Additional safety effects due to a change in modal split are determined by estimating the changes in travel mode and it’s effects on total accident numbers. By combining the different approaches a more complete overview of the impacts can be made. The goal of this paper is to describe the road safety impacts of increasing CAV penetration in the traffic system as they are identified in the Levitate project. Keywords: Automated transport, Automated vehicles, Vulnerable road users, Safety 1. INTRODUCTION In order to forecast the impact of AV introduction it is important to take into account the moment and The development and introduction of connected and speed of the introduction. As it is exceptionally automated transport systems (CATS) is expected to difficult to predict when AVs will become have far-reaching impacts on the traffic system and commonplace the project uses the market penetration society as a whole. This poses policy makers with rate of automated vehicles instead of different years. new opportunities and challenges for the integration This way it is possible to fit the results to new of these technologies in a controlled and safe manner. predictions on when certain penetration rates are met. The Horizon2020 project LEVITATE (levitate- project.eu) aims to aid policy makers by forecasting The introduction of automated transport is expected the societal, environmental and economic impacts to influence road safety in a number of different caused by the rise and eventual widespread use of ways. Within the LEVITATE project a distinction is CATS. One of the main impacts taken into account made between two types of impact: primary and within the LEVITATE project is road safety. This secondary impacts. Primary impacts is an impact that paper describes the approach taken within the project directly results from the automation of a vehicle. to determine the different road safety impacts and the Secondary impacts, also called rebound impacts, are results of this method. generated by the primary impacts and feed back to the source of that primary impact. Within the project the impacts on road safety are looked at from three different use cases: public Automated vehicles (AVs) affect road safety directly transport, personal transport, and freight transport. (primary impact) as they will have a lower risk of These use cases offer different looks into the impact being involved in a crash compared to human driven of automated transport on road safety. Further vehicles. This effect becomes even greater when specification within the use cases is done by way of vehicles are able to communicate with each other, i.e. different sub-use cases that explore the impacts of they become interconnected. In addition to these specific implementations or interventions. primary impacts some secondary rebound effects are to be expected. For example, it is likely that the total 39 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia distance traveled and the modal split of travel change separate method is used. Additional safety effects as a response to the availability of automated from modal split changes are not captured in the transport. Changes in modal split and distance above mentioned methods and are thus determined traveled have the potential to impact the number of separately by estimating changes in travel mode and crashes. the effects on total accident numbers. These three methods will be shortly discussed below. The current paper aims to give a concise overview of the different methods used within the levitate project In order to forecast the effects of increasing and describes the identified road safety impacts of an penetration rates of AVs a plausible implementation increasing penetration of CAVs. scenario is chosen. This scenario consists of different penetration steps that are chosen to represent the most 2. METHODOLOGY likely steps of implementation over time. A distinction is made between two different types of Within the project the safety impacts are looked at automated vehicle, cautious vehicles and ambitious from three different use cases: public transport, vehicles. These are chosen in order to represent personal transport, and freight transport. This technical developments over time resulting in more distinction between use cases is made due to the advanced vehicles at a later stage. Cautious vehicles different expected effects on road safety of the three represent automated vehicles that do have full self different types of transport. A further specification driving capabilities but do not drive closer together within the use cases into so-called sub-use cases is than human vehicles, instead choosing a similar time-done afterwards by looking at specific headway. Effects of a shorter reaction time and never implementations or interventions. Safety impacts are waning attention are present in these vehicles. first determined for the overall introduction of Ambitious vehicles represent the later stages of automated transport, with specification of impacts on implementation where penetration rates are high. a sub-use case level taking place in a later stage. These vehicles follow much closer together than The project combines multiple different human driven vehicles, choosing time-headways that methodologies in order to quantify the impacts of AV would be considered unsafe for human drivers. Table introduction. For the road safety impacts a major part 1 shows the different deployment scenarios used is played by traffic microsimulation. Traffic within the project. microsimulation entails the simulation of traffic in Scenario A represents the current composition. There different situations, allowing for the manipulation of are no fully self-driving vehicles available so vehicle and environment characteristics. This enables penetration rates are 0%. Scenarios B and C represent the modelling of traffic behavior of automated and the first introduction of cautious AVs, with a quicker human driven vehicles in different settings. Results introduction within freight transport. From scenario of the microsimulation allow for the determination of D onwards the share of cautious AVs does not further effects relating to surrogate safety measures such as increase. Instead the share of ambitious AVs time-headway and speed differences. As the increases until all human driven vehicles are replaced microsimulation is limited to vehicles such as cars, with automated vehicles. Scenarios G and H trucks and busses, the vulnerable road users are not represent the final stages of implementation where included. However, these are of course also part of older more cautious AVs are replaced with later more the traffic system and are impacted by the ambitious versions. introduction of automated vehicles. In order to include the safety impacts on vulnerable road users a Table 1: AV deployment scenarios Type of Vehicle A B C D E F G H Human-Driven Vehicle - 100% 80% 60% 40% 20% 0% 0% 0% passenger vehicle 1st Generation (Cautious) CAV - 0% 20% 40% 40% 40% 40% 20% 0% passenger vehicle 2nd Generation (Aggressive) CAV 0% 0% 0% 20% 40% 60% 80% 100% -- passenger vehicle Human- driven - Freight vehicle 100% 80% 40% 0% 0% 0% 0% 0% Freight CAV 0% 20% 60% 100% 100% 100% 100% 100% 2.1. Microscopic simulation well calibrated models for human behavior and utilizing model inputs that are known to represent As human behavior modelling has improved it real world safety performance at given locations a became possible to also use microscopic traffic prediction on impacts can be made. This allows for simulation to determine impacts related to road safety the estimation of different impacts such as speed, (Guido, Vitale, Astarita, & Giofrè, 2019). By using 40 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia traffic volume and congestion from an increasing For the accidents that can not be prevented by AVs penetration rate of automatic vehicles. By comparing the assumption is made that there is the potential for different sets of situations with each other the impact a reduction in accident severity based on the lower of different parameters can be determined. It should reaction times of automated vehicles compared to be noted that microsimulation is not a perfect human driver reaction times. Using this reduction in representation of the real word situation, assumptions reaction time it follows that the braking action of an have to be made. However, by comparing different automated vehicle starts earlier than that of a human situations with each other it is possible to work driven vehicle, resulting in a lower speed at the time around the disadvantage of assumptions as they will of impact. By comparing fatality rates at the different influence all situations the same. impact speeds an estimate can be made about the safety effects of improved reaction times for those In order to evaluate the microsimulation results with accidents that can not be fully prevented by replacing regards to road safety, a surrogate safety analysis is the human driver with an automated one. performed using SSAM. This produces a number of conflicts based on vehicle trajectories. The number of 2.3. Modal split changes conflicts is then converted to a number of crashes using a crash-conflict ratio based approach. The end The above mentioned methods are used for result allows for a comparison of the number of determining safety effects for replacement of human conflicts and crashes in a before situation with driven vehicles with automated ones and the effects numbers in alternate situations. Information about on vulnerable road users. The effects relating to distance travelled and where the traffic is located are changes in mobility and driving behavior are also output from the microscopic simulation. This estimated using a third approach concerning the enables the comparison of different implementation modal split changes. scenarios and interventions by comparing crash data Determining the impact of modal split change on per kilometer travel. It is important to note that while road safety is done by combining information about driving behavior of AVs is different, travel behaviors the exposure of different travel modes with such as amount of travel of all vehicles or time of information about the risks of different travel modes. travel do not change just by the introduction of The introduction of automated vehicles is expected to automated vehicles in the microsimulation model. change the balance between private, public and active Effect on travel behavior related to behavioral transport. In order to determine the changes in changes are excluded from the microsimulation amount of travel for the different travel modes a model. In layman terms the microscopic simulation combination of a mesoscopic model and a system replaces human driven vehicles one-to-one with dynamics approach is used. This model utilizes automated ones, reducing factors such as reaction agent-based modelling to simulate people completing time, following distance, speed variation. It does normal daily tasks while different options for however not change the number of vehicles that want transport are available. By modifying the settings and to travel or the distance those vehicles would travel. parameters it is possible to determine what changes These changes are collected in the modal split occur in the travel mode distribution. An increase in paragraph. travel with a certain mode translates to an increase in Within the Levitate project the microsimulation exposure for that mode. This change in exposure can environment AIMSUN (www.aimsun.com) is used then be combined with the risk of that type of with a model based on the city of Manchester. transport. The introduction of automated transport also impacts 2.2. Vulnerable road users the risks associated with different kinds of travel. The interactions of vulnerable road users with Reduction of human driver error will also reduce the automated vehicles are complex and require a risks for all road users that interact with human carefully chosen approach in order to ensure safety drivers. In order to capture this change in risk a for these users for whom a collision would be even comparison is made between the current risks of more dangerous than a car to car collision (Constant different transport modes and information about the & Lagarde, 2010). Due to the inability of capturing new risks coming from the microscopic simulation safety impacts on vulnerable road users with the and vulnerable road user methods. microsimulation environment the project relies on a Multiplying the change in risk rate of a certain mode different two pronged approach to estimate the of transport by its change in share of the modal split relevant safety impacts. Using accident statistics gives an estimate of the change in road safety impact. based on data available from Austria an analysis is When this is done for all modes of transport a new performed to determine current accident causes. This total road safety impact can be determined. A allows for the separation of accident causes between distinction is made between four different modes of VRUs and cars that could be prevented by AVs and transport within this approach: human driven those accidents that can not. vehicles, cautious CAVs, ambitious CAVs, 41 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia vulnerable road users, and other transport which unexpected situations. When the automation consists of busses and trucks. A two-step approach is technology is not yet fully capable of handling every taken which consists of a separate VRU situation it might be necessary to transfer control determination and motorized transport back to the human driver in certain situations. This determination, which are then combined for a total brings an increase in risk due to the transition of impact on safety. The end result allows for a control that has to happen, often suddenly and with comparison of the different modal splits that occur at minor warning (Zhang, de Winter, Varotto, Happee, varying market penetration rates of CAVs. & Martens, 2019). The most important direct impacts are discussed 3. RESULTS below in more detail. Within the Levitate project a distinction is made between two different kinds of impacts with relation 3.1.1. Human driver error to road safety, direct and indirect impacts. A direct Even the most experienced drivers make mistakes. impact is a direct result for the automation Human drivers sometimes disobey the traffic rules technology being implemented, while an indirect and make unintentional mistakes. These mistakes and impact occurs as a rebound effect or behavioral violations increase the risks of a crash occurring. adaptation. More information about all impacts When fully automated vehicles take over the driving considered within the Levitate project can be found task from the human occupant the risk of a driver in Elvik et al. (2019). error is eliminated. Automated vehicles have a direct effect on road Automated vehicles are assumed to always follow the safety due to a lower crash risk than the traditional rules of the road and will therefore reduce or human driven vehicles (Fagnant & Kockelman, eliminate the crashes that are a result of disobeying 2015; Logan, Young, Allen, & Horberry, 2017). This the traffic rules. Additionally, human issues such as is especially the case when vehicle to vehicle tiredness, distraction and lack of situational communication is added and the vehicles become awareness will likely be things of the past when connected (Malone et al., 2014). In addition to these vehicles can drive themselves. As a final point it is direct effects there are indirect effects expected. It is expected that automated vehicles will be able to likely that the modal split of travel and the amount of respond much quicker than human drivers and will be travel are effected by an increasing penetration rate better equipped for detecting other road users due to of automated vehicles, and it is known that modal a multitude of cameras and sensors. These split and the distance traveled have an impact on the improvements are even bigger when vehicles are able number of crashes. to communicate between each other and become 3.1. Direct impacts connected. At the core of automated transport is the goal of 3.1.2. Transitional effects replacing the human driver with a automated system that allows the vehicle to drive itself. Most crashes While fully automated vehicles are able to perform involve the human driver making an error at some the driving task in all situations there will likely be a point, even if this does not always mean that the period of transition where less advanced vehicles are driver is at fault or to blame. By moving the driving being used. These less advanced vehicles will be able task away from the human it is expected that to perform the task most of the time but require the automated vehicles will reduce the effect of these human driver to either pay attention or take over driver errors in crashes. While it is still unclear control when the system can no longer keep up. This exactly what percentage of human driver errors can transitional period has the potential for decreased be prevented by introducing CAVs, it is clear that road safety as unfamiliarity with a imperfect system sufficiently advanced vehicles should be able to can result in an increase in human error. These errors prevent most (Fagnant & Kockelman, 2015; Logan et are most notable when the human driver assumes the al., 2017). vehicle can drive itself when it can not (mode confusion) or when the vehicle indicates the driver However, CAVs also have the potential to introduce needs to take control back (take-over requests). new risks. Equipment or software might fail due to the many different parts involved, software When a take-over request occurs the human driver malfunctioning or cyber security problems. Other has to stop what they are doing and resume manual issues might occur with the AV not being able to control of the vehicle. This is a demanding task. detect or recognize other road users and infrastructure Literature reports different amounts of time needed to such as road signs and markings due to poor visibility perform a take-over safely, varying between less than in the case of bad weather or temporary lane a second and more than 20 seconds (Zhang et al., markings confusing the system. In more general 2019). This time is influenced by what the driver was terms, human drivers are likely better at dealing with doing at the time of the request, prior experience, type 42 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia of take-over request and capabilities of the vehicle. large scale study performed in the Netherlands shows Studies show that reaction times increase when that 2 out of 3 test drivers with a mandatory speed automation can no longer perform the driving task assist experienced negative interactions with other (Strand, Nilsson, Karlsson, & Nilsson, 2014), and an drivers (AVV, 2001). This is likely due to the increase in collisions when requests for take-over are inability of the test drivers to exceed the speed limit, presented (Radlmayr, Gold, Lorenz, Farid, & frustrating the drivers behind. Bengler, 2014; Rudin-Brown & Parker, 2004). These In addition to human drivers, cyclists and pedestrians results indicate that take-over requests pose a risk for might also change their behavior when confronted road safety. However, due to the complexity of take- with an AV. Research shows that pedestrians are over requests and the incomplete knowledge about currently not confident in the capabilities of self- real-life situations the actual impact is as of yet driving vehicles to detect and stop when approaching unclear. a pedestrian (Rodríguez Palmeiro et al., 2018). 3.1.3. System security and failure However, no difference in pedestrian crossing behavior was found between vehicles that were As with any vehicular system, failure of the system explicitly identified as self-driving compared to those can occur. In addition to the normal systems present that were not (Rodríguez Palmeiro et al., 2018). A in vehicles, AVs introduce a multitude of sensors and similar experiment designed for cyclists shows no software for object recognition. This has the potential differences in crossing behavior either (Vlakveld, van to increase the risk of a system failure resulting in the der Kint, & Hagenzieker, 2020). An overview of the vehicle being unable to detect an object or other road current knowledge gaps related to pedestrian and user. cyclist interactions with automated vehicles can be found in (Schagen, Kint, & Hagenzieker, 2017). Even when the vehicle systems perform well there is the potential for outside interference. As AVs are 3.2.2. Travel changes computer controlled they introduce so-called cyber risks. These risks become even more prevalent when The introduction of automated transport is likely to vehicles have the ability to communicate with each influence the decisions people make concerning other and have connection capabilities. Some of the travel. These changes might be due to higher current vehicles on the road already have the ability availability of motorized transport, the ability to to communicate with outside devices and are travel longer distances without the need for the driver therefore at risk of a hacking attempt. The industry to rest, or a lower valuation of travel time because the has not yet been willing to significantly invest in driver is able to engage with other tasks than driving. cyber security in order to prevent outside attacks on There is also the potential of vehicles making empty vehicle software. kilometers, in order to return home and be available for others or in order to avoid parking fees. 3.2. Indirect impacts Interventions have the potential to influence the Next to the direct impacts of AV introduction, amount of travel generated directly. For example by indirect impacts are also present. These impacts limiting access for AVs or implementing pricing on occur due to changes in factors that in their turn empty travel. impact road safety. People could for example change Generally speaking an increase in travel results in an their travel behavior due to the availability of increase of exposure. This increased exposure results automated vehicles. in an increase in the number of crashes. It is also relevant to know where the travel takes place, as the 3.2.1. Driver behavior adaptation risks of different road types can differ greatly. As the share of automated vehicles on the road Changes in route choice might occur due to the increases other road users such as cyclists, connected nature of automated transport enabling the pedestrians and human driver might adapt their sharing of information about congestion. There is driving behavior to the automated vehicles. Human evidence from current navigation systems drivers might adopt the smaller time-headways that influencing the route choices and impacting road the AVs are using when surrounded by these safety due to increased travel on roads not meant for vehicles. Research shows that this can already occur through traffic (Cabannes, Fighiera, Ugirumurera, when driving next to a platoon of freight vehicles Sundt, & Bayen, 2018). driving close together (Gouy, Wiedemann, Stevens, Brunett, & Reed, 2014). As automated vehicles 3.2.3. Modal split changes strictly adhere to the traffic rules a discrepancy might As automated transport becomes more readily occur between AVs and human driven drivers. This available it has the potential to greatly effect the split could result in unwanted behavior when a human between different modes of transport. The ability to driver has to adjust their speed or overtake the vehicle use an AV without a driver license, do work while due to the AV’s strict adherence to the speed limit. A traveling and automated options for urban transport 43 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia and goods delivery all influence the modal split. As occur due to the possibly of cyber attacks on the cars become a more attractive choice the usage of automated vehicles. public transport or active modes of transport such as Indirect impacts relate to effects that occur in walking and cycling declines. As crash rates differ response to the direct impacts. Changes in driving between the modes of transport it is important to and travel behavior due to automated transport know what the changes to the modal split are and how availability are the most important indirect impacts. these impact the overall road safety. While human drivers share the road with an 4. DISCUSSION increasing number of automated vehicles it is expected that human driver behavior will adapt some As the introduction of CAVs into the traffic system is of the automated driving behavior. Automated still years away from reality there are a number of vehicles are likely to drive closer together in order to assumptions that have to be made within the project. maximize road capacity, something that is possible These assumptions are made based on currently only due to improved reaction times. When this available literature and expert opinion but are still behavior is copied by human drivers it has a negative assumptions that might prove to be incorrect at a later effect on road safety. Next to expected changes in date. It is assumed that CAVs will be able to adapt driving behavior there are also changes expected in their behavior to their specific conditions in order to the choice around travel. The amount of travel is minimize risks. CAVs are expected to always respect likely to change with the introduction of automated traffic rules and have a positive impact on road vehicles. This could for example be due to the easier safety. The scope of this paper is limited to effects of access to travel when a driving license is no longer CAVs on road safety, additional effects of CAV needed. In addition to changes in the amount of travel introduction related to societal and environmental there are also likely changes in the split between ways issues is not discussed but does have the potential to of travel. The modal split is likely to move away from effect the implementation scenarios greatly. active modes of transports and move towards a bigger proportion of single vehicle travel such as cars. The project uses different methods to create a complete overview of the road safety impacts of The general approach to road safety effects as CAVs. However, the different methods utilize described in this paper enables policy makers and sources of information from different locations. For safety experts to get a clear overview of expected future projects is might be valuable to focus on a safety improvements and areas that could be of single location in order to allow for easier concern, allowing for adjustments to be made on time comparison. The approach used in the current paper in order to guarantee safe implementation of does allow for wider generalization. automated vehicles. 5. CONCLUSIONS REFERENCES The Levitate project aims to determine the impacts of [1] AVV. (2001). Evaluatie Intelligente SnelheidsAanpassing (ISA): Het effect op het automated vehicle introduction. Road safety is one of rijgedrag in Tilburg. the important aspects to take into account when looking at automated vehicles. The impacts on road [2] Cabannes, T., Fighiera, V., Ugirumurera, J., safety are determined using multiple different Sundt, A., & Bayen, A. (2018). The impact of methods that allow for the inclusion of city wide GPS-enabled shortest path routing on mobility: a game theoretic approach. traffic simulation, vulnerable road user impacts and travel choice analysis. This paper describes the [3] Constant, A., & Lagarde, E. (2010). Protecting methods used within the Levitate project to determine Vulnerable Road Users from Injury. PLOS the different road safety impacts and provides an Medicine, 7(3), overview of identified impacts. https://doi.org/10.1371/journal.pmed.1000228 [4] Elvik, R., Quddus, M., Papadoulis, A., Cleij, D., Impacts related to road safety are split into direct Weijermars, W., Millonig, A., Vorwagner, A., impacts and indirect impacts. Direct impacts Hu, B., & Nitsche, P. (2019). A taxonomy of represent the impacts that are direct effects of the potential impacts of connected and automated usage of automated vehicles as opposed to human vehicles at different levels of implementation driven vehicles. Reductions in reaction time, Deliverable D3.1 of the H2020 project deviations from traffic rules, distraction and mental LEVITATE. impairment are expected when humans are no longer [5] Fagnant, D. J., & Kockelman, K. (2015). performing the driving task. However, in the Preparing a nation for autonomous vehicles: transition toward fully automated vehicles there opportunities, barriers and policy might be a period where the human driver serves as a recommendations. Transportation Research Part fallback for the automation system. Sudden need to A: Policy and Practice, 77, 167-181. take over the driving task from the automated system https://doi.org/https://doi.org/10.1016/j.tra.2015. poses a risk to road safety. Additional risks could 04.003 44 YRS 2021 Rins de Zwart 15. -17. September IMPACT OF AUTOMATED VEHICLES ON ROAD SAFETY Portoroz, Slovenia [6] Gouy, M., Wiedemann, K., Stevens, A., Brunett, [11] Rodríguez Palmeiro, A., Kint, S. v. d., Vissers, G., & Reed, N. (2014). Driving next to automated L., Farah, H., Winter, J. C. F. d., & Hagenzieker, vehicle platoons: How do short time headways M. (2018). Interaction between pedestrians and influence non-platoon drivers' longitudinal automated vehicles: A Wizard of Oz experiment. control? Transportation Research Part F: Traffic Transportation Research Part F, 58, 1005-1020. Psychology and Behaviour, 27, 264-273. https://doi.org/https://doi.org/10.1016/j.trf.2018. https://doi.org/10.1016/j.trf.2014.03.003 07.020 [7] Guido, G., Vitale, A., Astarita, V., & Giofrè, V. [12] Rudin-Brown, C. M., & Parker, H. A. (2004). P. (2019). Comparison Analysis between Real Behavioural adaptation to adaptive cruise control Accident Locations and Simulated Risk Areas in (ACC): Implications for preventive strategies. An Urban Road Network. Safety, 5(3). Transportation Research Part F: Traffic https://doi.org/10.3390/safety5030060 Psychology and Behaviour, 7, 59-76. https://doi.org/10.1016/j.trf.2004.02.001 [8] Logan, D., Young, K., Allen, T., & Horberry, T. (2017). Safety Benefits of Cooperative ITS and [13] Schagen, I. v., Kint, S. v. d., & Hagenzieker, M. Automated Driving in Australia and New (2017). Zelfrijdende voertuigen: wat betekent dat Zealand. voor fietsers en voetgangers? [9] Malone, K., Rech, J., Hogema, J., Innamaa, S., [14] Strand, N., Nilsson, J., Karlsson, M., & Nilsson, Stefan, H., Van Noort, M., & Gusrafsson, D. L. (2014). Semi-automated versus highly (2014). DRIVE C2X Deliverable 11.4 Impact automated driving in critical situations caused by assessment and user perception of cooperative automation failures. Transportation Research systems. Part F Traffic Psychology and Behaviour, 27, http://personal.ee.surrey.ac.uk/Personal/K.Katsa 218-228. ros/papers/DRIVE_C2X_D11.4_Impact_Assess https://doi.org/10.1016/j.trf.2014.04.005 ment_v1.0_full_version.pdf [15] Vlakveld, W., van der Kint, S., & Hagenzieker, [10] Radlmayr, J., Gold, C., Lorenz, L., Farid, M., & M. P. (2020). Cyclists’ intentions to yield for Bengler, K. (2014). How Traffic Situations and automated cars at intersections when they have Non-Driving Related Tasks Affect the Take- right of way: Results of an experiment using Over Quality in Highly Automated Driving. high-quality video animations. Transportation Proceedings of the Human Factors and Research Part F: Traffic Psychology and Ergonomics Society Annual Meeting, 58, 2063- Behaviour, 71, 288-307. 2067. https://doi.org/https://doi.org/10.1016/j.trf.2020. https://doi.org/10.1177/1541931214581434 04.012 [16] Zhang, B., de Winter, J., Varotto, S., Happee, R., & Martens, M. (2019). Determinants of take-over time from automated driving: A meta-analysis of 129 studies. Transportation Research Part F: Traffic Psychology and Behaviour, 64, 285-307. https://doi.org/https://doi.org/10.1016/j.trf.2019. 04.020 45 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN STATIONS Capucine-Marin Dubroca-Voisin, PhD student LVMT, Ecole des Ponts ParisTech Boulevard Newton, Champs-sur-Marne, France capucine-marin.dubroca-voisin@enpc.fr ABSTRACT This paper addresses the problem of managing increasing pedestrian flows in metropolitan rail stations, in the context of decreasing financial resources. Based on research carried out both in academic and corporate contexts, the paper proposes a novel architecture including data collection, management levers activation, collaborative policy-making, efficiency evaluation and agent formation. The functional architecture is adapted into two potential systems, a simple one designed to test and validate the principles, and a more mature one. The application of these systems to an actual station is then explained. In urban rail mass transit systems such as Paris’ RER, passenger demand has been heavily growing in the past years, while the transportation offer struggled to develop. This resulted in an increased concentration of passengers in the major stations. Reached flows and densities became difficult to manage for the train and stations operators, who lack theoretical and organisational tools. This paper aims to contribute to improving the organisational side, by proposing a novel digital architecture. In such a solution, data could be collected, fused and analysed in order to activate pedestrian management levers. Once they have been activated, collected data would be used once more to assess the efficiency of the used strategy. Finally, a core part of the architecture would be to use this process as a pedagogic framework, enabling an efficient formation for involved operator agents. Once functionally described, the system is adapted to the case of Bibliothèque-François-Mitterrand station in Paris. Keywords: pedestrian management, stations, architecture, pedestrian flows, railway traffic 1. INTRODUCTION to develop data-collection and real-time management, but they lack theoretical and In a pre-covid world, travellers flows in the organisational tools. metropolitan cities were mostly increasing. These flows concentrate in high-frequented rail lines and This paper starts by a brief presentation of our our stations, leading in pedestrian accumulation, with use case station, Bibliothèque François Mitterrand, the potential risk of crowd movements. located in Paris. We then present the difference In urban rail mass transit systems, it can be difficult elements which constraint both the flow to manage the pedestrian flows for the railway management itself and the development of a flow operators, especially if they are much more familiar management system. with the historical railway procedures than with In a third part, we provide a brief overview of crowd control. systems already proposed by research teams. The fourth part focuses on the data collection In the meantime, it is complicated to develop the techniques, those available in most stations and in transportation offer or to extend the stations, as these our use case station. developments are extremely expensive while the Inspired by these examples, we propose our own public finances are largely constrained. functional architecture. We apply it to the case of This paper proposes an architecture aiming to help Bibliothèque François Mitterrand station, and railway and station operators to deal with these propose a technical architecture fitted to the local pedestrians flows, in a context where operators try needs. 46 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia That paper was conceived during a PhD contract most people staying at the end of the both in the LVMT research laboratory and in the platform to minimise their walking at French national railway operator, SNCF. It then destination. benefits from an inner perspective, but may be  There is also an unequal distribution biased due to the same reason. between stairs and the different escalators, with a saturation of the middle escalator, 2. THE BIBLIOTHEQUE FRANÇOIS which leads itself to the most crowded part MITTERRAND STATION of the platform. Bibliothèque François Mitterrand is a station located in the southwestern Paris, in a neighbourhood with recent urban developments, important density and mixed functional usage (national library, cinemas and shops, offices, university, housing). In its three underground levels can be be found two important public transport routes:  At the level -3 can be found the metro 14 line, opened in 1999. This automatic line is the fastest of the Parisian metro lines, and passengers are able to reach the center of Paris in less than 10 minutes. It operates at a very-high frequency (with 90 seconds Figure 1: Concourse of the Bibliothèque François intervals) and offers an impressive capacity Mitterrand in 2020. of more than 30,000 passengers per hour and per direction (pphpd). 3. ISSUES AND CONSTRAINTS FOR  At the level -1 can be found the RER C, a PEDESTRIAN FLOW suburban line running via Paris and opened MANAGEMENT WITHIN AN in 1979 (while the Bibliothèque François URBAN RAIL MASS TRANSIT Mitterrand itself was opened in 2000). RER STATION C capacity is over 40,000 pphpd, but the intervals are much higher between trains, We’ve identified different key issues for pedestrian with an average around 3 minutes between flow management, basing on an analysis of the trains on peak hours, and 5-15 minutes for station system, ground observations and interviews the same stations served. Bibliothèque with stakeholders. François Mitterrand is the last station Flow management is deeply constrained by different within Paris and constitues a pivot-station elements : between a slow, omnibus Parisian section of the line, and a section served by both  Stations are part of a complex railway express and local trains, with potentially system. Flows of trains must not be disrupted high speed, up to 140 kmph. by flows of passengers, but safety must be  The intermediate -2 level is a vast guaranteed at all times. concourse, with accesses to both metro 14,  Moreover, railway systems offers a very low RER C, and most of the station exits. While flexibility : it is difficult to change the train vast, it is not used by travellers for waiting, times and even more the intervals, as it is a as they prefer to reach the platforms timetable based system. immediately. In the mean time, only two of  Few human resources are present in the three available platforms are used by the stations, and they are not specifically trained RER C (one in each direction). These to flow management. platforms, despite their recent construction,  It is very difficult for station managers to are relatively narrow. have the global picture (line or network  This situation leads to different pedestrian state), and it is very difficult for line flow issues in the station: managers to know the current state of the  The difference of intervals between Metro stations. 14 and RER C leads to an accumulation of  Data collection is very limited by legal passengers coming from the metro on RER constraints, and the station itself is a very C southbound platform, particularly on the challenging environment. evening peak hours.  Data predictions are also challenging, due to  This problem is reinforced by the unequal unpredictable events having huge impacts. distribution along the 200 m-long platform, 47 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia Flow in station are a mix of continuous processes and discrete events. Building a system that helps train station operators to manage flow is also complex:  It needs to be reliable, while data collection quality can hugely vary.  It needs to be understood by the different actors , while flow phenomena are themselves challenging to understand.  It also needs to be quick, as the decisions have to been taken in real-time. Figure 3: Double loop of crowd management during  events, proposed in (Martella et al. 2017) While the pedestrian levers themselves can be relatively cheap, the system itself must not The most advanced system proposed in research be too expensive, as investment capacities in seems the one described in (Molyneaux et al. 2020). transportation are limited. Adapted from road traffic management systems, it proposes an estimation of the current state of the 4. PROPOSED SYSTEMS IN station based on several datasources. This estimation RESEARCH is used for a prediction of the future state of the system. Both are used to compute KPI, on which rely Several research teams have already proposed the decision making, with is determined by a control different architectures for pedestrian flow policy. management systems. We’ve found three of them particularly relevant to the needs we identified. In (Wagoum et al. 2012), a simple architecture is identified for crowd management in real-time. Using automated counting and a safety/security management system, it uses a simulation to display key indicators and a visualisation of the crowd. The originality of this work is its ability to show than a real-time simulation is usable in an actual case. Figure 4: Pedestrian traffic management system from (Molyneaux et al. 2020) While the double loop system highlights the importance of humans and a learning process, the two others are more focused on a technical approach, showing the complexity of flow management. There could be a need for integrating human Figure 2: Simple architecture using real-time learning in technical solutions, that’s the reason why simulation (Wagoum et al. 2012) we propose a different architecture in this paper. (Martella et al. 2017) have identified two loops of 5. AVAILABLE DATA COLLECTION crowd management by interviewing 10 Dutch TECHNIQUES professionals of crowd and flow management, including stations. This paper underlines the Several techniques can be used to collect data about importance of planning, with an output constituted pedestrian movements in stations. For instance, the by scenarios and plans who are used in decision system introduced previously and described in making. Data collected is then collected between (Wagoum et al. 2012) used camera automated events to improve the planning itself. The second person counting (APC), which is a very common loop takes places within the event itself, with the technique. execution of action plans and the measure of the crowd. We want to present here a very brief survey of such techniques, as it’s useful to know them for proposing a global architecture. Most of these techniques are rapidly evolving, so this survey has not any pretension to furnish a complete state-of-the art. We will simply propose a rapid overview, then see 48 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia which see which of these techniques are available in may need some training and normalisation of the the case of Bibliothèque François Mitterrand station. crowds behaviours to observe. 5.1. Brief overview of the techniques 5.1.2. Hybrid techniques An important part of data collection about pedestrian These techniques involve both the physical presence movement in stations is automated pedestrian of the pedestrian in the station, and for they to use or country (APC). In this category, we make an possess a device. These are mostly dependant of the important distinction between country the number of possession rate of the device. pedestrians in a space (the stock of passengers) and the number of pedestrians passing a line (the flow of GSM footprints are not very precise and cannot help passengers). Measuring density can be included in to measure pedestrian flows at the scale of the the first category. station, but they can be useful to determine the flow at the scale of the network, and then at least know Despite the huge importance of pedestrian counting, the origin-destination pairs in station (Mun et al. other measurements can be made, such as travel 2008). times, speed, trajectories or even qualitative observations about the crowd behaviour. WiFi (Kurkcu et Ozbay 2017) and Bluetooth (Liebig et Wagoum 2012) can be more precise and help to We consider that the sources used for pedestrian data determine the number of pedestrians in a space, and collection collection can use three kinds of even the trajectory using several hotspots or footprints: physical ones, hybrid (both physical and beacons. They’re heavily dependant on the digital) ones and purely digital ones. possession and activation rates, quite high for Wifi but lower for Bluetooth, and potential interference. 5.1.1. Physical techniques Another very commun hybrid source is the Use of cameras is the best known and probably the automated fare collection (AFC). Ticket gates are most used technique. Automated image processing very common in stations, and very precise as the has known great process in the past years, notably only errors are linked to fraud or the dysfunction of by using artificial intelligence tools such as neural the equipment, generally short due to the crucial role networks in important networks, and thanks to the of AFC in operators revenue. They provide improvement in calculation capacity. It enables both interesting data about pedestrian flow but only at stock and flow pedestrian counting, but it’s hard to some points of the station, and are not always deduce one from the other (Duives et al. 2018). One designed for real-time data transmission. interesting things with this technique is that stations often already have an expansive video collection 5.1.3. Digital techniques system, even if the use of this system can be restricted for legal reasons. However, the position of By using massively digital devices, travellers can cameras in stations is generally suboptimal for produce, voluntarily or not, digital footprints. Some automated counting, notably with problems of of these footprints can be used to predict flow in occlusion and luminosity variation. Associated station, for example using journey requests on a technologies can also be expensive. An calculator and deducing the demand at the station. improvement of the technique is the use of 3D GPS tracking could also be used, but can be as cameras, or thermic cameras. imprecise as GSM footprints, needs to be shared by the user, and are frequently not available in A common simple technique is the use of infrared underground spaces, generally the most crowded sensors, but in the context of stations it can only spaces in stations (so the ones needing pedestrian count the pedestrian flow in very narrow spaces, due management and data). to the importance of occlusion in that case. Another technique we characterise as digital is Some other physical techniques include counting crowdsourcing : lots of passenger themselves will floor carpets (measuring the pressure from walking report their own observations about the flow. pedestrians), CO2 concentration measurement (Han Therefore, this is actually a physical technique, but et al. 2013). or ultra wide band counting (Yang et al. with the point of view of the data collector, this is a 2019). The latter two techniques do not seem purely digital technique. It has already been largely sufficiently mature for a use out of the field of implemented by tech companies and train research. operations, especially for on-board load. Manual data counting and observation also belong A last possibility is to use social networks to detect to physical techniques. If humans are rapidly abnormal situations. Watching the networks could overwhelmed when the number of pedestrians not be very useful in normal times, as users are not grows, they are difficultly replaceable for qualitative very keen to share information about a banal observations. Notably, presence of agents in station situation in a situation, but on the contrary it could may be used to collect such observations, but that 49 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia be very useful to be alerted when an abnormal simplicity again, we will consider that the entire situation is reported on these networks. station has been equipped with these new gates. 5.2. Available techniques in the case of Therefore, we will use two main real-time data Bibliothèque François Mitterrand sources for our proposed architecture at station Bibliothèque François Mitterrand: camera counting for both flows and stocks of passengers within the Some on the previously cited techniques are station, and ticket validation for entrance and exit available on our use case station, Bibliothèque flows. François Mitterrand. The stations as has an extensive cameras network, who can be used for counting the 6. AN ARCHITECTURE INCLUDING number of passengers along the different platforms, DATA COLLECTION, DECISION in the concourse. This camera-based counting MAKING AND HUMAN LEARNING system has already been deployed previously in the station (Dubroca-Voisin & Bertaux 2020). This In this paper, we want to propose an architecture system could also be used to count the flow of which includes technical elements and human pedestrian within the stairs and the escalators, and learning as well. This architecture is schematised in we will use this possibility in our proposed the Figure 5. architecture. Four categories of humans are involved in this Wifi hotspots are available in the station and can be proposal. First, the data manager is responsible of used to have the global of passengers within the maintaining the quality of the data collected. Both station. As it is less precise than the camera system, automatic and manual data collection (including redundant, and for the sake of simplicity, we will not crowdsourcing) can suffer from different biases, consider this data source for this paper. errors and problems. An human intervention can be very useful to check and potentially correct this data The station is fully equipped with ticket gates at its when processing it. three entrances. In 2020, the existing gates started to be replaced with new ones, able to transmit in real- Levers managers have a key role in managing the time the number of pedestrians at each gate (even if flows: they decide of not to activate the different for some reason the gate is open). For the sake of flow management levers available. To do, they Figure 5: Global proposed architecture of the system, including human-data interactions 50 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia consult the flow supervision systems, which provide could involve real-time simulation, as proposed in them useful KPI and flow visualisation. They're able (Wagoum et al. 2012), but simple fusion techniques to approve the decisions proposed by the system or could be sufficient to ensure a sufficient quality. to take their own decisions. They help to apply a The processed data is used for both supervision flow management plan, but they also take a key role systems and decision making. These decisions are in designing this plan, using their experience. tracked, which is important for future data This plan is dependant on the policy, which can be processing. forged taking into account the desires from the different stakeholders (train operating companies, The flow management plan is a key data provider for public regulators, passengers, etc.). decision making, which appeals to very structured plans. It it itself base on the policy and an the Finally, all agents, including levers managers but available skills. also ground flow managers, need training. The skills gained can be tracked by an adequate system. Once the flow has been managed, the analysis provides new knowledge, which needs to be These humans interact with data in different states. encapsulated in a knowledge management system. It Data is first collected, using different techniques therefore influences policy production itself. such as image analysis, weight counting, numerical As this system is quite conceptual, we’ve decided to footprints detection. apply it the actual case of Bibliothèque François Data is then processed to ensure its consistency and Mitterrand station. make it usable for KPI and visualisations. This step Figure 6: Functional architecture applied to the case of Bibliothèque François Mitterrand station 51 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia 7. APPLICATION ON BIBLIOTHÈQUE understandable and sensitive on the number FRANÇOIS MITTERRAND of pedestrians on the platform. STATION USE CASE  K3: proportion of passengers using the stairs, the middle escalator and the last escalator. Bibliothèque François Mitterrand station offers an  However, some other KPI could be useful for interesting use car, due its current pedestrian issue the levers manager: described above, and its relative simplicity. To apply  K1.1: number of minutes before reaching the our proposed architecture, we first want to identify saturation threshold at the current pedestrian the local stakeholders then to propose relevant KPIs flow rate. This indicator could be useful to for this case. Finally, we propose both functional and alert the manager than an action is needed. technical architectures for that case.  K1.2: maximal density on the platform. This 7.1. Local stakeholders could indicate a local dangerous situation even if the number of pedestrians on the The station is managed by a local team of the train platform is still normal. operating company: station staff, train drivers or  In the proposed system, the precise KPIs to ticket inspectors work for the same company. use, their ideal values and the alerts threshold However, this company also outsources pedestrian should be defined by the different flow management on the southbound platform at stakeholders during the policy production peak hours, creating a second team with different process. In our example, avoiding accidents working conditions. and ensuring the best confort possible for As of 2020, there is no specific staff known as customer are global goals, while retaining ‘levers managers’ or even ‘flow manager’. The pedestrians for doing so is an acceptable function is however taken by the station staff lever. Some global KPIs could also be used; manager who’s present at the station. He’s a key they’re not KPIs dedicated to the real-time stakeholder as he will have the mission, in our operation of the system, but instead to ensure system, to monitor the situation, participate in the the global efficiency of the system, such as: flow management planning, and decide in real-time  K4: number of accidents linked to pedestrian which levers to apply. crowding (the target value for this value should of course be 0). He will be in direct link with the agents who will  K5: average level of comfort experienced by enforce his decisions, give him direct feedback, etc. each traveller, for instance by using Fruin and also with control centres (mainly the railway level of service scale. operation center of RER C, the RATP line 14 control  K6: total duration when pedestrians had to be center, and potentially the signalling cabin) to retained (as this is an acceptable but not exchange information. desirable solution). Other stakeholders will mainly be involved in These KPIs are important for the decision analysis, previous steps, notably elaborating the flow knowledge production, who could then be used for management policy. The stakeholders notably policy production and formation. include the passengers, who could have a key role in helping to determine which kind of levers and 7.3. Functional architecture regulations are acceptable or not. We so applied the global architecture to the case of 7.2. Relevant KPIs Bibliothèque François Mitterrand station. This application is summarised in Figure 6, using the As we explained above, there are three main same structure than Figure 5. Each one of the problems in the station : accumulation of passengers processes is detailed. in the southbound platform, unequal repartition on the platform and unequal use of stairs and escalator. Data collection is made using cameras counting We can define three immediate KPI: passenger stock on each section of the platforms and  in the concourse, and passenger flow at stairs and K1: number of passengers on the southbound escalators, while ticket gates are counting entrance platform.  and exit flows. K2: number of passengers who would need to move along the platform to have an equal Data processing uses two modes of verification : repartition between every section of the ticket gates data is used to check consistency of the platform. Unequal repartition could also be cameras data, while human checks are also represented by indicators such as Gini performed by the station staff. This enables to coefficient, but using number of passengers compute KPI such as those described in the previous makes the indicator both more directly section. 52 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia The supervision process could here take the form of Our technical solution is schematised in Figure 7. a map, showing the flows and the stocks of The schema is slightly modified to take into account passengers in the station, along with the relevant the fact that the data and levers managers are KPI. This map would be updated frequently and ‘technically’ the same person. consulted by the station staff manager to help him The two data collection systems are complicated take decisions. themselves and could be described in depth, but the These decisions would need to be communicated use of proprietary solutions reduce the complexity o and saved, ideally using the same tool. They would the global implementation. be made to ensure the goals defined by flow Data is then collected via APIs and computed trough management planning, and particularly the main a small custom solution. The implementation itself one, which is to limit the pedestrian accumulation on is not complicated but the algorithms can be, their platforms. ability to ensure the data consistency is key in the As the levers are not always activated, a good way success of the system. to perform an analysis is to compare data between This data comes to a global tool, enabling to monitor two days, one where the levers were activated and one where they weren’t. This comparison will help (via map and KPIs) but also to interact by indicating if data looks consistent with human observations or to determine which levers are the most efficient and not, but also the decisions taken. thus gaining knowledge. Decisions and data are stored with simple CSV Producing and obtaining this knowledge is an databases, enabling analyses made on notebooks. important part of the formation and training of the agents, including the station staff manager who’s This feeds a knowledge database listing the different levers used and their effects. responsible for flow management. This can either be relatively informal or supported by a tool. Policy and planning are made simply using unstructured documents, describing the global All this functional architecture must be completed policy wanted by stakeholders and its local by a technical one, in order to be applied properly. application to the station. This is technically 7.4. Technical architecture transformed into a config file, necessary for processing data and computing KPI. In this last section, we propose a very simple technical architecture for the case of Bibliothèque A simple technical solution would git to the François Mitterrand station. This solution is functional needs at the station. Its most complicated voluntarily simple if not simplest in order to propose elements are the map and data processing, due to the a first cheap and feasible solution for the need of precise algorithms and very powerful stakeholders. However, this solution still needs to be visualisation. complete and provide all the needed tools. 53 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia Figure 7: Architecture applied to the case of Bibliothèque François Mitterrand station [2] Duives, D. C., Daamen, W., & Hoogendoorn, S. 8. CONCLUSION P. (2018). Monitoring the Number of Constraints are high when designing a pedestrian Pedestrians in an Area: The Applicability of flow management system. In this paper, we’ve tried Counting Systems for Density State Estimation. Journal of Advanced Transportation, 1–14. to propose a novel architecture taking into account https://doi.org/10.1155/2018/7328074 both technical constraints and necessity of training. [3] Han, H., Jang, K.-J., Han, C. & Lee J. (2013). This architecture relies on efficient data collection, Occupancy estimation based on CO2 for instance with automatic pedestrian counting concentration using dynamic neural network using cameras or automated fare collection. It is then model. Proceedings of the 34th AIVC - 3rd necessary to propose a monitoring system, which TightVent - 2nd Cool Roofs' - 1st venticool favours taking adequate decisions. Those needs to be Conference , 25-26 September, Athens 2013. traced to enable analysis and so the production of https://www.aivc.org/resource/occupancy- knowledge concerning the levers at the station. estimation-based-co2-concentration-using- dynamic-neural-network-model We’ve applied this architecture to a simple case, [4] Kurkcu, A., & Ozbay, K. (2017). Estimating Bibliothèque François Mitterrand station in Paris. Pedestrian Densities, Wait Times, and Flows This station has recurrent issued, and we proposed a with Wi-Fi and Bluetooth Sensors. simple data collection system to help tackle these Transportation Research Record: Journal of the problems. Transportation Research Board, 1, 72–82. https://doi.org/10.3141/2644-09 REFERENCES [5] Liebig T. and Kemloh Wagoum A. (2012). [1] Dubroca-Voisin, C.-M., & Bertaux, P. (2020). Modelling microscopic pedestrian mobility TCO Voyageurs : une visualisation des using Bluetooth. Proceedings of the 4th voyageurs au services des voyageurs (Revue International Conference on Agents and générale des chemins de fer, pp. 62–73). Artificial Intelligence - Volume 2: ICAART, ISBN 978-989-8425-96-6, pages 270-275. DOI: 10.5220/0003833802700275 54 Capucine-Marin Dubroca-Voisin YRS 2021 A NOVEL GLOBAL ARCHITECTURE FOR MANAGING PEDESTRIAN FLOWS IN TRAIN 15. -17. September STATIONS Portoroz, Slovenia [6] Martella, C., Li, J., Conrado, C., & Vermeeren, [8] Wagoum, A. U. K., Chraibi, M., Mehlich, J., A. (2017). On current crowd management Seyfried, A., & Schadschneider, A. (2012). practices and the need for increased situation Efficient and validated simulation of crowds for awareness, prediction, and intervention. Safety an evacuation assistant. Computer Animation Science, 381–393. and Virtual Worlds, 1, 3–15. https://doi.org/10.1016/j.ssci.2016.09.006 https://doi.org/10.1002/cav.1420 [7] Molyneaux, N., Scarinci, R., & Bierlaire, M. [9] Yang, X., Yin, W., Li, L., & Zhang, L. (2019). (2020). Design and analysis of control strategies Dense People Counting Using IR-UWB Radar for pedestrian flows. Transportation. With a Hybrid Feature Extraction Method. IEEE https://doi.org/10.1007/s11116-020-10111-1 Geoscience and Remote Sensing Letters, 1, 30– 34. https://doi.org/10.1109/lgrs.2018.2869287 55 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Olga Maria Giannakari, Civil Engineer MEng Hellenic Institute of Transport 6th km Charilaou-Thermi Rd., Thessaloniki, Greece giannakari@certh.gr ABSTRACT Autonomous Vehicles (AVs) constitute an emerging area of research and development in the automotive industry. Automation in vehicles refers both to automated applications and features, such as the autopilot, Adaptive Cruise Control (ACC) and automated parking, and to fully autonomous systems where a physical driver is not necessary (SAE L4-L5 automation). So far, fully automated shuttles have been deployed in some countries, whereas several automated features are deployed in passenger vehicles already available on the market. In parallel, user acceptance of such systems is being studied, aiming to capture the end users’ needs, wants and priorities concerning automation in transport. Undeniably, these features and technologies are being developed having as ultimate aim to enhance traffic safety and decrease consecutive accidents and deaths. Yet, this does not appear as an adequate reason for convincing people to accept and use autonomous vehicles. Users seem to be reluctant, mainly due to the lack of trust towards a vehicle operating without a supervisor/driver. In this paper, a review of several surveys on user acceptance is being conducted, aiming at investigating users’ opinions and identify – where relevant - the source of their lack of acceptance towards AVs. Moreover, the usability of autonomous vehicles is being addressed, tackling issues that are commonly understated, such as the possible increase in mobility, as a consequence of the wide accessibility of AVs, as well as the expected increase in repair and replace costs of these automated, advanced safety systems. Keywords: autonomous vehicles, automated features, user acceptance, usability 1. ACCEPTANCE OF AUTONOMOUS in order to reinforce connected vehicles. Automated VEHICLES FOR PRIVATE USE and connected vehicles shape the future of transport, which is expected to be safer, based on the Autonomous vehicles (AVs) are emerging in our elimination of human error. Furthermore, AVs will lives. Specifically, driving assistance functions are be accessible to a wider audience, including elderly, already being integrated in new vehicles, such as children, people with disabilities that are incapable automated pilot, park assist, cruise control and more. of driving a vehicle, and other vulnerable road users Vehicles are also increasingly connected to the (VRUs). It seems that it will take some time to internet and are constantly engaging digital ensure passengers and drivers about automated technologies, which are moving away from the pure driving, but the benefits will always outstand their “electro-mechanical” world. All of these new worries. features and functionalities aim at providing safer, more accessible, cost and fuel efficient travels, while 1.1. Review of acceptance of private contributing to a more sustainable environment. autonomous vehicles by end users Although the general public is adapting in these End users, such as passenger vehicles’ drivers, changes slowly, they seem to be particularly hesitant passengers and VRUs, are seeking for improved, when it comes to fully automated vehicles or even affordable and green means of transport nowadays. vehicles operating with a minor assistance from the Electrified and automated mobility comes as a driver. State of the art technology, artificial solution to a lot of everyday problems commuters intelligence and machine learning are used to equip face. Yet, people’s main concern is road safety. autonomous vehicles, while communication Autonomous vehicles are designed to offer an infrastructure, such as 5G, G5 and IoT, is developing unambiguous way of interaction between the vehicle 56 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia and the driver/passenger. But does the technology operation of the vehicle is needed. This interface and the infrastructure meet users’ needs and wants? also assists the communication between the The prioritized needs and priorities from the part of driver/operator of the vehicle, the other road users the involved end users include the sense of perceived and the authorities (e.g. the police) (Kulmala, safety while using automated driving solutions. Jääskeläinen, & Pakarinen, 2019). The use of Automated vehicles are expected to improve road adapted HMI aims to mitigate risks that can occur in safety since they are designed to communicate with driverless or assisted driving modes. In the context the environment, including pedestrians, other of the EU project Adas&Me1, an HMI application vehicles – equipped or not, VRUs, and to act quicker stated is detecting when the driver is not capacited than a driver, whose attention may be destructed or to drive, due to fatigue, stress and distraction. his reaction time may be slower. Yet, autonomous Particularly, a survey was conducted to assure that vehicles as of today operate only with the assistance the developed systems are able to accurately detect of a driver, at levels 2-4 as the Society of when the driver/rider is not capacitated to drive, and Automotive Engineers (SAE) has indicated. This consequently mitigate these states and avoid fact raises new concerns about road safety such as dangerous situations through the use of adapted risks of driver confusion/distraction, misuse of the HMI and automation mode (Cocron, et al., 2018). systems and liability issues, which have to be The evaluation focused on: addressed in order to ensure introduction and use of  verifying the effectiveness of the systems to automated vehicles in our transportation system recognize the driver’s state, (Kulmala, Jääskeläinen, & Pakarinen, 2019).  Furthermore, in order to encourage the acceptability attesting the capacity of the HMI to display of automated solutions and potential replacement of clear and unambiguous information,  the current status of provided services, all of the end evaluating the driver behaviour following a users, passengers, drivers, need to be properly system warning/suggestion, informed about of this new era of automated and  collecting the driver’s opinion on the connected mobility. According to a survey system’s usability, conducted between the 28 Member States of the EU  knowing the driver’s trust and acceptance in 2019 and published in the Special Eurobarometer levels regarding the ADAS&ME functions. 496, “Expectations and concerns of connected and In an automation-ready survey conducted for the automated driving Europe” (Joint Research Centre, project CoExist2, which aims at preparing the 2020), 58% of the people questioned, have heard, introduction of the automated vehicles, the read or seen something concerning automated preliminary results showed that respondents believe vehicles during the past twelve months. This that the most relevant actions to take are the percentage varies at national level, with the development of regulations for CAVs (96%) and the Netherlands and Sweden concentrating above 90% definition of responsibilities regarding data positive replies, while in Slovakia, Malta and Czech management (85%). Surprisingly, the adaptation of Republic positive answers are just above 50%. local mobility plans (56%), the involvement of Another challenging aspect to look for is people’s citizens (56%) and capacity building (40%) are the willingness to pay for autonomous vehicles (AVs) or measures considered the least relevant to prepare for services they provide. Surveys show that opinions the introduction of CAVs on our roads (CoEXist, differ depending on age and gender: younger people 2019). According to Alonso Raposo et al., four key are more eager to use or pay for AVs or different game changers are shaping the future of road services offered in AVs, such as communicating and transport: automation, connectivity, decarbonisation productivity (Alonso Raposo et al., 2019). Also, and sharing. Preliminary studies on user willingness male users are more open to use AVs than female to use (or pay for) AVs seem to reflect an overall ones (Alonso Raposo et al., 2019). Age is also a positive acceptance of these new systems. Gender factor affecting the perception of accessibility in and age differences appear: male users seem to be both the phase of design and more willing to use AVs than female counterparts implementation/operation. Internet illiteracy and young people tend to show greater willingness addresses most elderly people, let alone IoT and C- to use or pay for AVs compared to elderly people ITS services. Therefore, the deployment of AVs (Alonso Raposo et al., 2019). Users have also needs to be specially adapted to vulnerable road expressed willingness to pay for different services users, disabled, elderly etc. To begin with, Human- offered in AVs, with those relating to Machine Interface (ΗΜΙ) is vital for the deployment communication (e.g. social networks) and of autonomous vehicles operating with the productivity ranking highest in their ratings assistance of a driver. HMI has been developed in compared to, for example, entertainment-related autonomous vehicles operating at SAE L2-L4, services (Alonso Raposo et al., 2019). However, a where a driver operating or supervising the 1 https://www.adasandme.com/ 2 https://www.h2020-coexist.eu/ 57 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia significant portion of the population still has a deployment of connected vehicles. Overall the negative attitude towards driverless vehicles. On top results suggest respondents are not yet ready to fully of that, another JRC’s report, “The r-evolution of adopt connected and automated vehicles. Another driving: from Connected Vehicles to Coordinated interesting survey was conducted for the EU project Automated Road Transport (C-ART)” , states that AUTOPILOT4. In the project’s deliverable “User overall users’ acceptance towards AVs is slowly requirements analysis” one of the scenarios improving as people are starting to experience more developed, addressed carsharing of an AV as a and more with the technologies through the existing touristic experience. The survey was online and it demonstration and marketing activities of the was conducted in eight countries: Finland, France, various manufacturers and technology companies Germany, Greece, Italy, Netherlands, the UK and (Alonso Raposo, Ciuffo, M., & Thiel, 2017). Spain. Participants willingness to use the service Furthermore, in the context of the British project UK was generally positive, with two thirds of them Autodrive3, the researchers at Cambridge agreeing that they would use the service if it was University’s Engineering Department and the available (Aittoniemi, et al., 2018). Furthermore, in Department of Psychology carried out a national the deliverable named “User acceptance survey of public attitudes towards self-driving assessment”, another user acceptance evaluation vehicles, gathering 2850 responses. Results show was carried out for five European pilot sites, where that more than three quarters (76%) of those four use cases were tested, using autonomous surveyed had heard of driverless vehicles, vehicles: Automated valet parking, Highway pilot, comparing favourably to driver assistance Platooning and Urban driving. In total, 199 public technologies that are already commercially participants took part in user tests across the available, such as adaptive cruise control (familiar to AUTOPILOT project. Some interesting results 40% of those surveyed), automated emergency detected in all pilot sites were that a) the majority of braking (38%) and lane-keeping systems (34%) (UK the users (> 88%) found it (very) important to have Autodrive, 2017). Another survey conducted for the the option to take back control over the vehicle at Eurobarometer report aims at assessing EU citizens’ any time, b) information that makes the trip safer and awareness of automated vehicles and their comfortable, such as information about possible experience with automated or semi-automated hazards or waiting time was evaluated as an driving functions; measuring attitudes towards important factor and c) requirements on driving in or interacting with automated vehicles on customization options, especially that the the road; evaluating citizens’ willingness to information is provided in the own language, were purchase and use automated vehicles; determining also an important feature from the user’s point of what citizens expect with regard to automated view. Overall, participants’ concerns were about: vehicles. The survey was carried out in the 28  the functionality of the system (worried about Member States of the EU, with a total of 27656 system failure e.g. in detection of objects, by respondents from different social and demographic hitting pedestrians), groups being interviewed face to face. A majority of  respondents say they would not feel comfortable in the unexpected or harsh a fully automated vehicle without the supervision of brakings/accelerations,  a human operator in it, but 7/10 would feel the otherwise uncomfortable driving style comfortable travelling in a fully automated vehicle (e.g. cuts-in, lane change, jerks, swaying, with the supervision of a human operator in it. In the slow speed, driving close to objects), question “To what extend would you feel  the technical failure of the system (e.g. comfortable or not travelling in a fully automated navigation, take-over, manual braking) and vehicle with the remote supervision of a human  the properties of vehicle or operator?”, 30% answered that they would feel not automation/service (e.g. uncomfortable seat at all comfortable, while 31% would feel not very belts, turning of steering wheel, HMI, comfortable (Joint Research Centre, 2020). automatic gear shift). However, it is interesting that in a 2017 Generally, although there were a lot of differences Eurobarometer survey, between 52 and 63% of users between the services tested and the way in which would feel uncomfortable “Being driven in a they were tested, the outcomes were rather positive driverless car in traffic” (Alonso Raposo et al., and, in many cases, similar. Levels of acceptance 2019). This shows an encouraging aspect as the varied, but there was no real rejection of the services more aware people become towards vehicle (Aittoniemi, et al., 2019). Zoellick, J. et al measured automation, the more prone they are to accept it. : 1) acceptance of AV, 2) perceived safety, 3) trust, Still, a relative majority of respondents are not in 4) intention to use and emotions associated with favour of the deployment of fully automated AVs surprise, fear, boredom and amusement upon a vehicles on roads, but the majority is in favour of the 3 http://www.ukautodrive.com/ 4 https://autopilot-project.eu/ 58 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia pilot test with 125 participants who have gaps in establishing autonomous vehicles for private experienced a ride in an electric AV on a large clinic use. area in Berlin, Germany. Respondents had Table 1: Needs, wants and priorities of end users and previously taken a ride with an AV on a private existing gaps for automation in private vehicles terrain along 2 campuses in mixed traffic. Results state that respondents were amused, surprised and Gaps Needs Wants Priorities not afraid after the experience. The results stand in 1 Familiarise User Driver Safety, with friendly- to be Willingness contrast to critical comments addressing automation HMI, able to to pay uncomfortable interior, slow driving and abrupt Accessibl take braking (Zoellick, Kuhlmey, Schenk, Schindel, & & e, control Blüher, 2019). Another report on autonomous of the vehicle vehicles was conducted from a customer perspective 2 Costumer Environme from the research institute CAP Gemini. The survey, training based ntally addressing over 5500 consumers from around the on age groups sustainable world and executives at 280 companies, from (with VR/AR) automotive OEMs to technology players states that automotive consumers seem to be ready for the It seems that one of users’ main concern is a possible future and for self-driving cars operating at SAE system failure in fully automated vehicles. level 4. Consumers seem very clear about their expectations, concerns, and desires, and they are 2. ACCEPTANCE OF AUTONOMOUS willing to pay a premium to receive them. In VEHICLES FOR PUBLIC response, automotive and tech organizations are TRANSPORT making significant investments in hardware and Automation in public transport has long been along software and racing to test and pilot vehicles. The of the curve. Metro lines in London and Paris are recommendations for accelerating the journey operating autonomously for the past decades towards a self-driving future include: informing the already. Automated Road Transport Systems consumers, understanding and reassuring them, (ARTS) though, are still developing. Automated building an ecosystem of services, developing of vehicles in road transport are clearly more software skills. As results indicate, 52% share of challenging to be developed and operate efficiently consumers would prefer self-driving cars five years than track-based transport modes. Automation and from now, while 56% would be willing to pay a connectivity are rapidly increasing in road transport premium of up to 20% over their current budget for and mobility, making every imagination of a self-driving car (Winkler, et al., 2019). seamless, public transport feasible. Cities suffer 1.2. Gaps, needs, wants and priorities of from traffic congestion due to the excessive number automation in private vehicles of private vehicles and consequently travel time is high. Although cars offer a more pleasant and safe As it is suggested by the surveys and reports environment and a personalised commute, door-to- reviewed, people are not ready to accept fully door, the vast number of them has negative impacts automated vehicles in their lives, either as on the environment, causing air pollution, and drivers/passengers or as pedestrians in environments people’s social and psychological status, since they where autonomous vehicles operate. This hesitance consume a lot of time travelling instead of having is justified in the lack of trust towards a vehicle this time for personal pleasure. Public transport operating without a supervisor, as many issues always aimed at alleviating this lifestyle, but failed accompany this situation. A critical issue raised is to succeed many times. Autonomous vehicles are the assignment of the responsibility for the vehicle designed to be convenient, functional, accessible to operation; whether this should be aligned to the all users, including VRUs, and fuel efficient since vehicle manufacturer, the providers of the system or AVs used for public transport will be electric. an insurance company that may be in charge of the Therefore, the deployment of autonomous vehicles vehicle. The same issue also applies to lower level for public transport is expected to lead to a reduction of automation vehicles, that do equip a of vehicle ownership, with public transport driver/supervisor to perform specific tasks while becoming an efficient and convenient mode of driving (SAE L4). It seems that a common and transport for all, while providing a safe, automated, operational legislative framework needs to be time saving option for commuting. Road transport developed prior to introducing semi and fully and mobility of the future are characterized by automated vehicles in order to promote both safety increased automation and connectivity. User safety, and harmonization of transportation and to reassure energy consumption and efficiency, traffic users of automated vehicles. The matrix in Table 1 congestion and drivers’ and passengers’ comfort and summarises the findings relating to the needs, wants convenience are expected to be assessed by and priorities of end users, as well as the existing automation and connectivity. In public transport, 59 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia autonomous vehicles can result in a cost reduction people would prefer fully autonomous over manual of approximately 50%. Instead of a driver in each driving and would use an autonomous vehicle vehicle, one person in a control centre may or shall (Portouli, et al., 2017). Similar to the shuttle bus in monitor and, if needed, manoeuvre, several vehicles Trikala, another self-driving bus called Digibus, (Kulmala, Jääskeläinen, & Pakarinen, 2019). At the operating at SAE L2-L3, was tested on a public road same time, new technologies and features, such as in Austria. The shuttle from the French company sensors, wireless communications (G5), 5G Navya Tech has been tested on a 1.4km long track networks, IoT and big data will be largely developed in the village of Koppl. The trial in Koppl was one and will provide new business models for the of the first trials of autonomous public transport automotive sector and their cost is expected to vehicles worldwide on public roads with mixed decrease. Therefore, plenty stakeholders are affected traffic in a rural area. The focus of this trial was on by this new era in transport. Public transport the real-world evaluation of a self-driving shuttle for operators need to provide efficient, safe and bridging the first/last mile in public transport. A connected transport with low cost. Cities promote questionnaire survey was conducted to collect data automation as it constitutes a potential alternative on passenger experience. The majority of passengers solution to current and future problems, such as were between the ages of 21 and 50. 15.1% were increased traffic congestion, while also providing between 51 and 60, 9.6% were over 60. Just over 5% for environmental sustainability, by increasing the of respondents were between 13 and 20. Concerning use of public transport. Besides, a sustainable, smart employment status, 65.4% of passengers were city, comprising automated and connected public employed, 11.0% were in training or were self- transport, not only buses, but metro and tram lines employed and 8.2% were retired. Almost 80% of as well, is more appealing to residents and business. passengers own a private car, while 20.5% of them The combination of high-capacity public transport do not. Almost 40% of those questioned, claimed with automated car sharing services are expected to they could imagine that a self-driving shuttle bus in result in a considerable reduction of the average their municipality could remove the need for a travel time comparing to using a private vehicle (second) car. Around 59% do not think a self-driving (Kulmala, Jääskeläinen, & Pakarinen, 2019). Also, shuttle bus is a realistic option and may not replace by investing in the enhancement of public transport a private (second car). Regarding prior knowledge networks, systems and services, policy makers and on the topic of self-driving vehicles, 13% of authorities may accomplish the elimination of passengers indicated they had none. Almost 43% vehicle ownership. That is the reason why research had heard of self-driving vehicles, 44% were more institutes and academia focus their research interests familiar with the topic. This high percentage is due in the integration of automated vehicles, in all to the fact that many of the test passengers were part aspects of road transport. of company delegations who had relevant knowledge in this area. For the majority of 2.1. Review of acceptance of autonomous passengers (84.7%), their Digibus test drive vehicles for public transportation by represented their first ride in a self-driving shuttle. end users 9.2% indicated that they had been on a self-driving Quite a few pilot projects have taken place across shuttle before and 5% indicated that they had ridden Europe. A questionnaire survey conducted in the on the Digibus in the past. The reasons for taking city of Trikala, where an autonomous mini bus was part in a test drive varied and were equally operating at SAE level 4, for a period of six months, distributed: 29.4% took part out of professional or in mixed traffic in the city centre. The pilot was part scientific interest, 25.5% out of interest in an of the CityMobil2 EU project. The questionnaire innovative means of transport, 23.0% were was addressed towards two different audiences; the motivated by curiosity, 20.4% were interested in the first part of the survey was intended to passengers at technology. Around 92% reported enjoying or very the moment when they descended from their trip and much enjoying their ride on the Digibus. According the second one was a citizens’ offline survey, during to passenger testimonies, they especially appreciated demonstrations were taking place. The first part of the vehicle's "comfortable driving style", "feeling of the survey, which gathered 200 responses from safety", "automated nature of the driving" or its passengers, indicated that the mini buses were "reliable detection of other road users or obstacles". widely accepted and integrated in citizens’ everyday A little more than 6% of passengers indicated that life without highlighting any significant concerns they had not particularly enjoyed their Digibus ride regarding the safety and security arose. The second and 1% said they did not enjoy their ride at all. The part of the survey among citizens, concentrating 519 reasons for this included, for example "the sharp responses, revealed that age is a significant factor braking or frequent braking", "the prototypical when it comes to using the autonomous mini buses, nature of the technology" or "the lack of smooth driving behaviour“. The passenger survey also as younger people were identified as regular users. Experienced with driving automation, younger revealed very positive results regarding perceptions of safety on board. Almost 90% of passengers felt 60 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia safe or very safe on board the Digibus. It must be transport. In general, user acceptance increases to assumed, however, that passengers' feelings of the same extent as the autonomous vehicles are security would probably be lower if the shuttle were optimised and adapted to fit the different user completely driverless. As per the reasons passengers requirements (Dubielzig, Reisch, & Panou, 2018). gave for not feeling safe on board, they were "abrupt Also, as part of the CityMobil2 project, a study or jerky braking", "lack of confidence in the new focused on the users’ expectancies which might technology", "lack of experience", " poor sensor influence behavioural intentions regarding the use of system" or that "the shuttle cannot differentiate Automated Road Transport Systems (ARTS). A between people and vehicles". When asked what total of 349 valid responses were collected. Results passengers could imagine using the Digibus for in show that performance expectancy is the strongest their neighborhood, 28.3% answered they would use predictor, suggesting that the most important factor it to commute to work or school or as a shuttle to the that people will consider in deciding whether or not next public transport stop. 20.7% of them indicated to use an ARTS is how well they believe it will they would use a shuttlebus like this for daily perform in comparison to other public transport errands, such as going shopping, attending doctor's systems. Social Influence and Effort Expectancy appointments, visits to local government offices etc. also had an impact on behavioural intentions, Around 16% of passengers could imagine the indicating that the influence of other people, and Digibus being used for either leisure activities (trips perceptions of how difficult the system is to use will to football training or music lessons, as a shuttle to also both influence the decision to use an ARTS. ski-lifts or hiking trail start points etc.), as a delivery Apart from that, other factors such as perceived service for parcels, shopping etc. or as work-site safety or on-board comfort should be considered in transportation on enclosed company grounds. 1.4% future work in this area (Madigan, et al., 2016). A of the respondents could not imagine an application similar survey was conducted in order to investigate for a shuttle bus of this kind (Rehrl & Zankl, 2018). the factors that influence users’ acceptance of Another questionnaire was conducted in the context automated road transport systems (ARTS). The of the HORIZON 2020 project AVENUE5, focusing audience was the users of the CityMobil2 pilot on people with disabilities. Yet, the methodology, including a shuttle bus in the city of Trikala. In total, questionnaires and replies collected are generic, 315 participants completed the questionnaire. All of targeting all types of population. People with some them had used the shuttled bus at least once, with 14 kind of impairment (hard of hearing, reduced participants having used it more than 5 times. The mobility or visually impaired) were 57% of the total main aim of this study was to use an adapted version interviewees. None of the participants with hearing of the UTAUT framework to investigate the social- impairment would use an autonomous bus, while psychological factors that influence users’ 57% of the visual impaired would use it. However, acceptance of an automated road transport system there might not to be a correlation between hearing (ARTS). In particular, users’ enjoyment of the impaired and acceptance of autonomous vehicles, system plays a big part in their desire to use it again, because all of them are over 60 years of age. In the while the performance of the system, the resources age group of over 60 only 44% (15 Interviewees) provided to support its use and the social popularity would use an autonomous bus, while 10 would not of the system all appear to be important factors do it. In the group of users without any impairment (Madigan, Louw, Wilbrink, Schieben, & Merat, only 32 % would not use autonomous vehicles. It has 2017). also been noticed that almost 60% of the participants 2.2. Gaps, needs, wants and priorities of talk with the driver, half of them for conversation, regardless the age. Generally, this survey shows that, automation in public transportation although results suggest a high acceptance and Results occur out of the many pilots that have taken expectation from the interviewees, a hesitant attitude place around Europe, (e.g. Finland, Greece, towards autonomous vehicles is manifested. This Austria), in larger and smaller cities. Mini hesitance is due to empathy for drivers who will lose autonomous buses (shuttles) were occupied in city their jobs and, at the same time, due to the fear of the centers, with users welcoming the concept and using absence of a “supervisor”. Interview partners state the service for their commute. The autonomous that they would use autonomous busses “only if they vehicles demonstrated required a supervisor on were in service for a long time without accidents”. board, since they are not yet qualified for fully So, even the interviewees with the most welcome autonomous driving. Along with the pilots, many towards autonomous buses attitude, would prefer a surveys on users’ acceptance and perception of the security supervisor in autonomous public transit. automated public transport were conducted. Even if passengers’ needs differ according to their Pertaining to those surveys, it appears that the disability, all of the user groups expect a significant autonomous mini buses were widely accepted and added value in the use of autonomous buses in public used for small commutes. A significant factor 5 https://h2020-avenue.eu/ 61 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia affecting users’ acceptance is age, since younger 3. USABILITY OF AUTONOMOUS users were identified as more frequent users. Safety VEHICLES was mainly inspired from the presence of the supervisor inside the vehicle. Passengers declare 3.1. AVs impact in mobility that they would be uncomfortable riding on a bus As a matter of fact, autonomous vehicles will with no physical supervisor to be in charge of the broaden the landscape of potential drivers/operators, vehicle and take control when/if needed. In the since underserved population, such as non-drivers, future plans for automated public transport services, teens, elderly and people with disabilities, will be the supervisor of the vehicle will operate the vehicle able to use AVs for their commuting. Therefore, remotely from a control centre. Although the vehicle automation is subject to increase mobility, a feedback was positive, as to if users would use again side effect that is not desirable. The development of an autonomous bus for their commute, a fear for the autonomous vehicles aims to mitigate traffic driver’s absence emerged. Therefore, passengers’ accidents, enhance traffic safety and reduce psychological status when riding on an autonomous emissions. An increase of the number of vehicles vehicle needs to be estimated so as to detect which will further deteriorate traffic congestion. In order to circumstances make them feel comfortable and avoid such an unwanted consequence, autonomous which not. Public transport users’ needs and vehicles can be used by more than one family priorities have to do with the safety during their trip, member or among strangers, through a car sharing the comfort in the vehicle, the frequency of the scheme. Also, autonomous vehicles are equipped scheduled routes and the price of the corresponding with technology infrastructure that allows them to be services. Overall, automation in public transport will connected as well. Connected and autonomous have an impact on travelers’ behavior, their social vehicles (CAVs) can communicate between them, and psychological attitude towards automation, but improving the efficiency of transportation systems also on urban development, environment, (Elliott, Keen, & Miao, 2019). As a result, entertainment and commerce, growth and jobs. The developing a mobility web instead of promoting the findings from the literature review indicate the most private use of AVs can efficiently reduce the number important needs, wants and priorities of end users, of cars in the streets (Duarte & Ratti, 2018). A along with the existing gaps in automated public mobility web will also impact parking spaces. Since transportation. A table collecting these is presented AVs will be user more frequently as a matter of car below. sharing, fewer cars and therefore parking spaces will Table 2: Needs, wants and priorities of end users and be necessary (Duarte & Ratti, 2018). According to existing gaps for automation in public transportation Hannon et al., the boundaries among private, shared, and public transport will be blurred, given that Gaps Needs Wants Priorities connected and autonomous vehicles can operate 1 Safety Frequent Accessibility Safety-need assurance service to all of a upon request, as well as plan trips upon request, (including security while offering travelers flexible ways to commute VECs*), personnel, and keeping a low cost (Hannon, Knupfer, Stern, & Affordable Clean Nijssen, 2019). Replacing private car ownership is tickets interior 2 User- Access expected under the utilization of Shared friendly remote areas Autonomous Vehicles (SAVs), by offering an app, efficient, flexible and affordable on‐demand Smooth mobility (Mohammadzadeh, 2021). Eventually, driving style-no researchers acknowledge the possible increase in harsh private vehicle’s ownership and have already come brakes or up with strategies, technologies and plans in order to impatient tackle the threat and actually promote a sustainable driving 3 On demand Clear Door to door Environme transportation solution. service: announcem service ntally- Door to ents on friendly 3.2. Advanced driver assistance systems door board, commute (ADAS) Comfortabl e interior of Autonomous vehicles acquire a lot of advanced vehicles systems in order to operate all of the automated * Vulnerable to Exclusion, e.g. elderly, disabled, tourists, functions. Already, vehicles of SAE level 2 and 3 are internet/language illiterate, unemployed, unprivileged, refugees equipped with plenty of advanced systems, such as the Adaptive Cruise Control (ACC), Lane Keeping Assistance, Blind Spot Warning, Parking Assistance and more. Obviously, vehicles of higher automation are equipped with even more advanced systems so as to offer safe transportation. According to AAA 62 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia (American Automobile Association), “at least one different types and models. The results are presented ADAS feature is available on 92.7% of new vehicles in Table 4. available in the U.S. as of May 2018” (American Table 4: Estimated repair costs per part Automobile Association, 2019). Some of the most Technology Repair cost important advanced safety systems integrated in vehicles are presented in Table 3, along with the Front Camera Sensors $850-$1900 corresponding technologies required for their Front Radar Sensors $900-$1300 operation (Edmonds, 2018). Side Mirror Sensors $500-$1100 Table 3: Technologies for advanced safety systems Rear Radar Sensors $850-$2050 Technology ADAS feature Ultrasonic Sensors $500-$1300 Automatic Emergency Source: (Edmonds, 2018) Braking Front Camera Sensors Adaptive Cruise Control 4. CONCLUSION Lane Departure/Keeping Automatic Emergency Focusing on the passenger acceptance of AVs, Front Radar Sensors Braking Adaptive Cruise Control safety is the driving force in establishing AVs, with Side Mirror Sensors Around View Monitoring the "fear of the unknown” or “the lack of Blind Spot Monitoring confidence" pointed out as the main reason of Rear Radar Sensors Rear Cross Traffic Alert concern, inconvenience and doubt. This endogenous Ultrasonic Sensors Parking Assistance mistrust is empowered by the fact that until now the Source: (Edmonds, 2018) legislative framework is a little obscure pertaining to All of these systems are developed to provide a safe the share of responsibility in case of accident and driving experience, reduce traffic accidents and save who pays except for the insurance companies. In lives. However, their invasion in the automotive addition, the willingness to pay is also a key issue market did not allow for drivers to get familiar with for everyone, according to who is going to be using these systems, and as a result many buy new burdened with the initial investment and how cars, not knowing how to operate the new, advanced affordable is the new automated passenger safety systems. This situation may cause even more transportation system is going to be for the average significant accidents, if drivers trust that these user with or without public subsidy and / or other systems are capable of avoiding any possible threat support or participation from the private domain or without their own involvement in the driving through public private partnership schemes. In fact, procedure. To conclude, specialized training should the potential costs versus the expected revenues be offered to past, new and future drivers, focusing constitutes one of the most important parameters on advanced safety systems operation and taken into consideration in the frame of spatial capabilities. planning at strategic and governmental decision making level, but it is equally significant at tactical 3.3. Repair cost and operational level as well. Last but not least, there While technology is rapidly evolving and safety is also the accessibility matter which is twofold: of systems are designed with a view to exclude the course there is the issue of accessibility in the phase driver from actually operating the vehicle, a field of design concerning the VRUs, but on the other commonly neglected is the repair cost of these hand there is the need for easy and equal access from systems. According to an AAA research from 2018, all the users no matter the level of their “vehicles equipped with advanced driver assistance familiarization with the new technologies, the IoT, systems (ADAS) such as automatic emergency the C-ITS and the smart mobility concepts and braking, blind spot monitoring, lane departure services, also meeting the expectations for the warning and others, can cost twice as much to repair minimization of the required equipment and following a collision due to expensive sensors and respective services. This means that any service or their calibration requirements.” (Edmonds, 2018). application developed towards the integration or The cost of repairing or replacing these new features upgrading of the provided services is desired to be is unfortunately unbearable for most users. equally accessible by all, irrespective of the Repairing these systems can also occur from minor socioeconomic status of the user, the available incidents, related to crashing a door or even a mirror. equipment - hardware or software (e.g. smart phone Furthermore, repairing or replacing of some typical or tab running a simple easy to download and car features, such as the windshield, will also require operate app) and the level of knowledge, know how, the replacement of the camera that is behind the technology illiteracy and familiarization. What is glass in an equipped vehicle, multiplying the repair more, another issue is the difficulty from the part of cost to even three times the cost without the the users towards the sharing and compilation of advanced system (Edmonds, 2018). AAA estimated their personal data. Furthermore, factors that affect repair costs of sensors, taking into considerations the users’ preference towards automated road transport systems (ARTS) over other public transport 63 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia systems is each one’s performance, as well as the [5] American Automobile Association. (2019, social influence from other people, depending on January). ADVANCED DRIVER their recommendations, preferences and indications. ASSISTANCE TECHNOLOGY NAMES - AAA’s recommendatio A user-friendly service is appealing to a larger n for common naming of advanced safety systems. Retrieved from audience, including the elderly, people with NewsRoom.AAA.com: disabilities and other VRUs. A balance between https://www.aaa.com/AAA/common/AAR/file private autonomous vehicles and AVs for public s/ADAS-Technology-Names-Research- transportation is expected to occur, taking into Report.pdf consideration the tendency to shared mobility [6] Cocron, M. P., Vallejo, A., Delgado, M. B., researchers focus on, along with the need for Wilbrink, M., Anund, A., Krupenia, S., & accessible, convenient and low cost transport. The Zanovello, L. (2018). D7.1: Evaluation cost accompanying owning and maintaining an Framework. ADAS&ME. autonomous vehicle could be a suspending factor for increased vehicle ownership as well. Sustainable [7] CoEXist. (2019, December 4). CoEXist Automation-Ready survey – preliminary results cities need sustainable solutions to combat traffic available. Retrieved from CoEXist: problems. More lanes will not fix the problem of https://www.h2020-coexist.eu/coexist- traffic congestion. Neither will more vehicles. automation-ready-survey-first-results- available/ REFERENCES [8] Duarte, F., & Ratti, C. (2018). The Impact of [1] Aittoniemi, E., Barnard, Y., Chen, H., Ertl, D., Autonomous Vehicles on Cities: A Review. Harrison, G., Kolarova, V., . . . Touliou, K. Journal of Urban Technology, 3-18. (2018). D4.7: User requirements analysis. doi:10.1080/10630732.2018.1493883 AUTOPILOT EU. [9] Dubielzig, M., Reisch, M., & Panou, M. (2018). [2] Aittoniemi, E., Barnard, Y., Ertl, D., Harrison, D2.4: First Passenger needs analysis and G., Kolarova, V., Malin, F., & Touliou, K. specifications. AVENUE project. (2019). D4.8: User requirements analysis. [10] Edmonds, E. (2018, October 25). New Vehicle AUTOPILOT EU. Technologies Double Repair Bills for Minor [3] Alonso Raposo, M., Ciuffo, B., M., & Thiel, C. Collisions. Retrieved from (2017). The r-evolution of driving: from NewsRoom.AAA.com: Connected Vehicles to Coordinated Automated https://newsroom.aaa.com/2018/10/new- Road Transport (C-ART) Part I: Framework for vehicle-technologies-double-repair-bills-minor- a safe &. doi:10.2760/225671: EUR 28575 EN. collisions/ [4] Alonso Raposo, M., Ciuffo, B., Alves Dias, P., [11] Elliott, D., Keen, W., & Miao, L. (2019). Recent Ardente, F., Aurambout, J., Baldini, G., advances in connected and automated vehicles. Baranzelli, C., Blagoeva, D., Bobba, S., Braun, Journal of Traffic and Transportation R., Cassio, L., Chawdhry, P., Christidis, P., Engineering (English Edition), 109-131. Christodoulou, A., Corrado, S., Duboz, A., doi:https://doi.org/10.1016/j.jtte.2018.09.005 Duch Brown, N., Felici, S., Fernandez Macias, [12] Hannon, E., Knupfer, S., Stern, S., & Nijssen, J. E., Ferragut Martinez Vara De Rey, J., Fulli, G., T. (2019, January 16). The road to seamless Galassi, M., Georgakaki, A., Gkoumas, K., urban mobility. Retrieved from Grosso, M., Gomez Vilchez, J., Hajdu, M., McKinsey&Company: Iglesias Portela, M., Julea, A., Krause, J., https://www.mckinsey.com/business- Kriston, A., Lavalle, C., Lonza, L., Rocha Pinto functions/sustainability/our-insights/the-road- Lucas, A., Makridis, M., Marinopoulos, A., to-seamless-urban-mobility Marmier, A., Marques Dos Santos, F., Martens, B., Mattas, K., Mathieux, F., Menzel, G., [13] Joint Research Centre. (2020). Special Minarini, F., Mondello, S., Moretto, P., Mortara, Eurobarometer 496 - Expectations and concerns B., Navajas Cawood, E., Paffumi, E., Pasimeni, of connected and automated driving Europe. F., Pavel, C., Pekar, F., Pisoni, E., Raileanu, I., ISBN: 978-92-76-17323-6. Sala, S., Saveyn, B., Scholz, H., Serra, N., Tamba, M., Thiel, C., Trentadue, G., Tecchio, [14] Kulmala, R., Jääskeläinen, J., & Pakarinen, S. P., Tsakalidis, A., Uihlein, A., Van Balen, M. (2019, June). The impact of automated transport and Vandecasteele, I., The future of road on the role, operations and costs of road transport, EUR 29748 EN, Publications Office operators and authorities in Finland. ISSN of the European Union, Luxembourg, 2019, (online) 2342-0294, ISBN (online) 978-952- ISBN 978-92-76-14318-5 (online),978-92-76- 311-306-0: Finnish Transport and 14319-2 (print), doi:10.2760/668964 Communications Agency Traficom. Retrieved (online),10.2760/524662 (print), JRC116644. from TRAFICOM: https://www.traficom.fi/sites/default/files/medi a/publication/EU_EIP_Impact_of_Automated_ Transport_Finland_Traficom_6_2019.pdf 64 YRS 2021 Olga Maria Giannakari 15. -17. September ACCEPTANCE AND USABILITY OF AUTONOMOUS VEHICLES OVER THE LAST YEARS Portoroz, Slovenia [15] Madigan, R., Louw, T., Dziennus, M., [19] Rehrl, K., & Zankl, C. (2018). Digibus©: results Graindorge, T., Ortega, E., Graindorge, M., & from the first self-driving shuttle trial on a Merat, N. (2016). Acceptance of Automated public road in Austria. Eur. Transp. Res. Rev. Road Transport Systems (ARTS): An 10, 51, https://doi.org/10.1186/s12544-018- Adaptation of the UTAUT Model. 0326-4. Transportation Research Procedia Vol. 14, [20] UK Autodrive. (2017, August 8). Survey finds 2217–2226. UK public still “open minded” about self- doi:https://doi.org/10.1016/j.trpro.2016.05.237 driving vehicles. Retrieved from UK Autodrive: [16] Madigan, R., Louw, T., Wilbrink, M., Schieben, http://www.ukautodrive.com/survey-finds-uk- A., & Merat, N. (2017). What influences the public-still-open-minded-about-self-driving- decision to use automated public transport? vehicles/ Using UTAUT to understand public acceptance [21] Winkler, M., Mehl, R., Erander, H., Sule, S., of automated road transport systems. Buvat, J., KVJ, S., . . . Khemka, Y. (2019). The Transportation Research Part F: Traffic autonomous car A consumer perspective. Psychology and Behaviour, 50, 55–64. Capgemini Research Institute. doi:https://doi.org/10.1016/j.trf.2017.07.007 [22] Zoellick, J. C., Kuhlmey, A., Schenk, L., [17] Mohammadzadeh, M. (2021). Sharing or Schindel, D., & & Blüher, S. (2019). Assessing owning autonomous vehicles? Comprehending acceptance of electric automated vehicles after the role of ideology in the adoption of exposure in a realistic traffic environment. autonomous vehicles in the society of https://doi.org/10.1371/journal.pone.0215969: automobility. Transportation Research PLoS ONE, 14(5). Interdisciplinary Perspectives, Volume 9. doi:https://doi.org/10.1016/j.trip.2020.100294 [18] Portouli, E., Karaseitanidis, G., Lytrivis, P., Amditis, A., Raptis, O., & Karaberi, C. (2017). Public attitudes towards autonomous mini buses operating in real conditions in a Hellenic city. 2017 IEEE Intelligent Vehicles Symposium (IV) (pp. 571-576). Los Angeles, CA: doi: 10.1109/IVS.2017.7995779. 65 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH REPUBLIC Blanka Hablovičová, M.Sc. Vítězslav Křivánek, Ph.D. Petra Marková, M.Sc. Transport Research Centre (CDV) Líšeňská 33a, 636 00 Brno, Czech Republic blanka.hablovicova@cdv.cz ABSTRACT Changes in acoustic characteristics of the road’s wearing courses during years of using have an immediate influence on the total noise level from the road traffic. According to the CNOSSOS-EU for the calculation of strategic noise maps, protection of the environment against traffic noise is very important. Long-term noise measurements by the CPX method are performed by only one company in the Czech Republic (CZE): Transport Research Centre (CDV). Low-noise wearing courses began to be used a few years ago (from mixture PA, through BBTM, to SMA NH (NH is the Czech abbreviation and marking for low-noise)). CDV cooperated on the elaboration of the technical specification TP 259 defining the rules for designing asphalt mixtures for noise reducing wearing courses and issued the certificated method for assessing their long-term noise performance, both of them were issued in 2017. Noise reduction increased slowly until documents were announced but rapidly increased after this because everyone had to start following the rules that had not been there before. Due to constantly changing low-noise mixtures, it is necessary to observe and evaluate acoustic characteristics of road surfaces during the lifetime of wearing courses and the whole roadway. Tyre/road noise very much depends on the asphalt mixture and condition of the wearing course. Low-noise and common asphalt mixtures are laid in some locations in the CZE right next to each other so that their behavior over time can be studied. Practical results measured in the CZE are described in this study. Keywords: Low-noise wearing course, asphalt mixture, road noise reduction, CPX. 1. INTRODUCTION directive implies requirements for Strategic Noise Mapping (SNM) with regular updates in five-year Noise is one of the most important (and still rising) cycles in all EU countries. The methodology pollutant affecting human health (World Health CNOSSOS-EU (Kephalopoulis et al., 2012) for Organization, 2018; Klæboe, 2011) with impact calculating SNM is mandatory for EU members such as sleep disturbance (Pirrera et al., 2014; since 31 December 2018 (in legislation recorded as Ohrstrom et al., 2006), cardiovascular and Commission Directive 2015/996 (2015)). The physiological effects (Babisch, 2008; Bluhm et al., Environmental Noise Directive also requires to 2007), mental health effects (Klompmaker et al., implement action plans in which noise reduction 2019; Stansfeld et al., 2009), annoyance (Okokon et solutions are needed. The most used solutions for al., 2018) or cognitive impairment (Jafari et al., traffic noise reduction are noise barriers which can 2018; Jafari et al., 2020). Traffic noise has also achieve high noise reduction rate. Their construction impact on wildlife (Francis et al., 2009; Castaneda and maintenance costs are unfortunately high and et al., 2020; Finch et al., 2020), on economy and they cannot be implemented into the urban areas in social life (Fyhri et al., 2006; Kim et al., 2019) many cases as applicability in a street canyon is (detail study about noise costs is contained in the problematic, e.g. possible non-compliance with the chapter 6 of handbook (Van Essen et al., 2019)). conditions for daylighting of residential units, noise The growing trend of noise pollution has led to barrier would not be effective due to interruptions by strategic and operational legal measures through side streets etc. Replacement of the road surface for EU, namely with the adoption of the Environmental low-noise wearing course is one of the approaches Noise Directive (Directive 2002/49/EC, 2002). The to reduce traffic noise, especially in urban areas, 66 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia with lower cost than noise barrier or façade acoustic behavior of wearing courses (to compare insulation (Larsen et al., 2001; Krivanek et al., the results from different locations) is the close- 2020). proximity (CPX) method in accordance with the results of the European project ROSANNE (Haider The aim of this paper is to present the equipment for et al., 2016) since 2015. CPX measurements are monitoring of wearing courses in the Czech carried out by various types of measuring systems Republic, to introduce the most important across the world, such as enclosed CPX trailers documents about low-noise wearing courses at the (Kleiziene et al., 2019; Miljkovic et al., 2012; national level, and to show some partial results from Vazquez et al., 2019), open CPX trailers (Tonin et tire/road contact noise measuring in the Czech al., 2014; Van Leewen et al., 2007; Kragh et al., Republic. 2013). Test tire with microphones can be also assembled directly on a test vehicle (Cesbron & 2. LOW-NOISE WEARING COURSES Klein, 2017; Licitra et al., 2014; Campillo-Davo et AND NOISE MONITORING al., 2019). Round robin test between nine CPX The properties of the road wearing course are highly systems from four countries performed by Peeters et influencing the total road traffic noise emission. The al. (2018) concluded that all systems met importance of surface and condition of road will requirements, which are differences from average continue to increase as a consequence of the reference values, for random errors (range 0.11– development of electric vehicles and 0.27 dB) and repeatability (range 0.07–0.22 dB), electromobility. Noise generated by tire/road two didn’t fulfil the requirements for systematic contact, whose mechanism is described in (Sandberg error, differences were above 1 dB. Vieira and & Ejsmont, 2002), is prevailing from speed approx. Sandberg (2019) presented similar conclusions from 40 km/h for current passenger cars with combustion round robin test in Sweden (two open trailers and engines and for heavy vehicles it is around 70 km/h two enclosed trailers were tested) with some (Sandberg & Ejsmont, 2002), whereas for electric recommendations for the future of CPX method. So vehicles, tire/road contact noise dominates from the results obtained with different CPX trailers speeds as low as 20 km/h due to very quiet engines which meets ISO 11819-2 (2017) are comparable. (Czuka et al., 2016). Basic structural properties of low-noise pavements for appropriate tire/road noise 3. SITUATION IN THE CZECH reduction are: high voids content, minimal REPUBLIC megatexture, and not a very smooth macrotexture Long-term road noise measurements are performed (Sandberg, 1999). by Transport Research Centre (CDV) in the Czech Sandberg (1999) described the history in short about Republic (CZE). SPB method was used for road the low-noise pavements before year 2000. noise measuring until 2011 (Cholava et al., 2009). Descornet (2000) introduced low-noise techniques CPX method was tested and compared with SPB for many materials such as cement concrete, resin- during the project CG712-102-120 (Transport bound surface, porous asphalt, thin layers and some Research Centre, 2011). Important output of the specific structures. Aksnes et al. (2009) presented project was the approval of the CPX trailer for road results from development and testing of low-noise noise measurements in the Czech Republic. The pavements adapted to Norwegian conditions, comparative noise measurement were proceeded Anfosso-Lédée and Brosseaud (2009) monitoried between CPX systems of CDV and Eurovia porous asphalt concrete and very thin asphalt (Krivanek et al., 2014) in the years 2013–2018 concrete in France. Vaitkus et al. (2014) performed within project CESTI (2018). So, the regular laboratory testing (acoustic properties included) of monitoring of road noise surfaces by CPX method stone mastix asphalt and porous asphalt used in has been carried out by CDV since 2012 (Krivanek, Lithuania according to Lithuanian standards. Markova & Spicka., 2016) and measurements along Acoustic ageing of rubberized pavements was distances over 5,500 km were performed until 2020. modelled by Licitra et al. (2019). PIARC published Methodology from 2014 (Transport Research a summary (PIARC, 2019) of global knowledge of Centre, 2014) is an effective CZE tool for traffic noise at the end of 2019 – noise limits, impact independent verification, evaluation and assessment of noise on human health, environmental aspects, of noise properties of roads since 2014. Based on the economic evaluations, prediction, measurements, obtained data since 2012, two important documents evaluation, anti-noise measures and maintenance of were published in 2017: the first was certified road surfaces. methodology (Transport Research Centre, 2017) which presents values of acoustic emission of road Implementation of reliable and comparable surfaces and their change over time, and the second measurements of wearing courses over a long period was Technical Specification No. 259 (TP 259) is a necessary condition for the correct evaluation of (Valentin et al., 2017) which e.g. defines the the influence of road surfaces on traffic noise. requirements for materials of low-noise pavements, Recommended method for long-term monitoring 67 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia specifies the method of marking, determines the 4. MEASURING PROCEDURE technical requirements as void content, and construction conditions as compaction. Both 4.1. CPX method documents are fully integrated into the quality Measurements have been performed on an open system in the field of roads. CPX trailer designed by CDV with approximately According to the National Reference Laboratory, width 2.5 m and length 5 m (see Fig. 1) which meets CNOSSOS-EU is intended as a decisive method of requirements of the international standard traffic noise determination for the Czech Republic in ISO 11819-2 (2017) describing close-proximity the future. It defines road surface noise corrections method. Standard is supplemented by ISO 11819-3 for entire EU globally but these pavements don’t (2017) that defines reference tires for use in the CPX correspond to wearing course mixtures in individual method (the standard was revised in January 2021). EU member states, not only by their parameters but Locations of individual measuring microphones on also by their composition or noise levels (each a CPX trailer according to ISO 11819-2 are shown country uses mixtures adapted to local conditions). in Fig. 2 and the real placement can be seen in Fig. 3. So implementation of CNOSSOS-EU in national The standard ISO 11819-2 specifies two mandatory Czech conditions is ongoing. microphones for CPX measurements, but more microphones can be used where more complete Average noise level of the typical road section processing of tire/road noise is required as for low- (ACO 11) in the CZE at the second year of lifetime noise surfaces. So, the system was upgraded from 5 was determined as reference value for tire/road microphones to 6 microphones (supplemented, contact noise of wearing courses in the CZE. tested and verified) according to the standard during Reference value is set at 90 dB for velocity 50 km/h winter 2018/2019. and 98 dB for velocity 80 km/h (for tyre P1 according to ISO 11819-3 (2017)) based on measurements performed in the Czech Republic and foreign experiences. Pavements with lower values have noise reducing properties. The low-noise surface is a surface that can reduce tire rolling noise by at least 3 dB compared to an average noise level of typical road surface. Newly laid low-noise wearing courses can reduce noise by up to 5 dB. They are divided into low-noise wearing courses which meet TP 259 (with abbreviation NH) and into wearing courses outside of TP 259 but still reducing noise due to small aggregate fraction since 2017 when TP 259 was issued. The first experimental Figure 1: Open CPX trailer designed by Transport implementations of low-noise wearing courses in Research Centre. CZE were realized in 2010, approximately twenty of these test sections were layed until 2012 (porous asphalt (PA) primarily). PA has excellent properties immediately after laying but clogging and loss of surface macrotexture which decrease noise reduction properties, occur very quickly, especially in urban areas, PA wearing course is also damaged by freeze-thaw cycles (Kwiatkowski et al., 2014). Wearing courses SMA LA (from originally Austrian or German “lärmarm”) or BBTM have been used since 2014 based on foreign experiences (especially from Germany) which were incorporated to TP 259 Source: ISO 11819-2 (2017). so they are named SMA NH and BBTM NH (NH is Figure 2: Diagram of the locations of individual Czech abbreviation for low-noise and has the same measuring microphones on a CPX trailer. meaning as LA) since 2017. Grinding technology based on diamond grinding of concrete surface is being tested but for low-noise wearing courses, asphalt mixture are prioritized. 68 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia tire/road noise emissions. Only one tire should be selected for measurements. Commonly manufactured tires don’t have adequate production stability and the results must be comparable during time. Tire Uniroyal Tigerpaw 225/60 R16 SRTT (in ISO 11819-3 marked as P1) is recommended for passenger cars in the automotive industry as a standard tire for reference tests according to ASTM F2493 (ASTM, 2020), so the tire SRTT P1 was used for measurements, see Fig. 5. Every year is used newly acquired reference tire in order to minimize the effect of degradation (measurements are performed in the period from about April to Figure 3: Real placement of individual measuring October). Impurities in the tread were removed microphones on the CPX trailer. before the actual testing. Hardness has been tested every two months. Measurements were performed at a speed of 50 km/h for rural areas, and at 80 km/h for other categories of roads (eg. highways, bypass). Velocity of vehicle with CPX trailer (GPS module UA-9004 RLVBSS1), air temperature (sensor SENECA PT100 probe) and surface temperature of measured road (infrared sensor CALEX – PC21MT-0) were continuously recorded during each measurement, see Fig. 4. The measuring multianalyzer Brüel&Kjær was composed of PULSE LAN-XI (module 3050-B-060 and module 3056-A-040). The software PULSE LabShop which can work and synchronize all signals in one timeline, even if they have different sampling frequencies, was used. Pre- polarized free field microphones Brüel&Kjær type Figure 5: Used reference tire Tire Uniroyal Tigerpaw 4189, in accuracy class 1 and with sensitivity 225/60 R16 SRTT. 50 mv/Pa, were used for self-sensing noise. 4.2. Measured data The main output of the measurements is the equivalent sound pressure level of tire/road contact and third-octave characteristic of section according to ISO 11819-2. All measurements have been adjusted by A-weighting filter. Data obtained by synchronic measurements from the 6 microphones postprocessed within the PULSE LabShop program. CPX measurement has prescribed spectrum range that is recorded but only the required part of spectrum is evaluated. Disturbing influences, as crossing over the sewerage cover or high peaks caused for example by the rescue siren, were removed manually. Arithmetic average of noise levels values at the 6 microphones was performed by Excel for all valid measurements, it is given as the Figure 4: LAeq, velocity and temperature on a selected final result rounded to one decimal place. section on synchronized timeline. The effect of the measuring tire represents the 5. RESULTS AND DISCUSSION greatest uncertainty in the noise measurement at the tire/road contact (Licitra et al., 2017; Buhlmann et Results from two different measurement sites are al., 2007), which means the same type of tire should selected as partial examples. Comparison of be used for long term measurements. Rubber common and low-noise asphalt mixture (wearing hardness and temperature of the tire have to be course) from the same locality is made. The sections within a specified range (Sandberg et al., 2016) and follow immediately after each other, i.e. sections are the standard ISO/TS 13471-1 (2017) determines affected by almost the same meteorological procedures for defining effect of temperature on conditions, traffic intensity and composition as well as the velocity of the traffic flow. Periodic 69 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia measurements of noise verification were after laying and after two years it was 90.2 dB, it is madein summer annually, the wind speed was below an increase of only one decibel. Compared to that, 5 m/s and obtained noise values were recalculated noise increase of PA 8 CRmB during two years of according to the standard thus all results were lifetime is 4.2 dB, from 85.5 dB to 89.7 dB. Low- referenced under comparable conditions; e.g. noise noise asphalt mixtures are mostly surfaces with high value measured under condition of velovity void content, especially porous asphalt has higher 78.9 km/h and temperature 23.8 °C was corrected to void content then other low-noise asphalt mixtures. noise value under conditions 80 km/h and 20 °C. In order to achieve the maximum void content of Year 0 of wearing course in legend of results (Fig. 6 asphalt, the individual grains are bonded in points and Fig. 9) represents the year of laying this and there are high requirements for the binder, e.g. pavement, eg. if year 2015 is year 0, then 2016 is apply suitable additives such as mineral or cellulose year 1, 2017 year 2 etc. fibers (Valentin et al., 2017). Nevertheless, there is a risk of gradual weakening of the contact surface The acoustic lifetime of the low-noise wearing which leads to a higher susceptibility to damage of course is characterized as the period during which the pavement – ripping out of aggregates from monitored type of wearing course achieves an wearing course (Sandberg & Ejsmont, 2002). In improvement in acoustic attenuation compared to addition, clogging occurs at these surfaces (Fig. 7). reference value. When the wearing course reaches its acoustic life limit, this doesn’t mean that the life The self-cleaning effect by the traffic is minimal at lower speeds, so a specific treatment is required limit of such wearing course is reached. Acoustic (Krivanek, Pavkova et al., 2016). Regular life of wearing course can be very different and can maintenance of all low-noise wearing courses is be fundamentally affected by appropriate rather not performed in the CZE. Less significant maintenance. noise increase of the low-noise surface may be due 5.1. Rural road to their a shorter service life compared to common surfaces. Lifetime of low-noise surfaces is set at The first example represents a rural area with a 8 years and lifetime of common wearing courses is velocity limit of 50 km/h. Industrial (heavy freight) 12 years according to the Czech experiences and and agricultural activities are very active in this area. informations from Directorate of Roads and It is a class II road with traffic flow above Motorways. After three years of lifetime was noise 3,000 vehicles per day. Differences between noise of wearing courses ACO 11 and PA 8 CRmB development of low-noise (PA 8 CRmB, section 90.6 dB for both of them and then noise of porous length 780 m) and common (ACO 11, section length asphalt increased faster than noise of common 260 m) asphalt mixture is shown in Fig. 6. The most asphalt ACO 11, so in six year of wearing course significant increase of noise (no road’s defects) lifetime was noise of PA 8 CRmB 92 dB, resp. occurs in the first three years after laying which is 91.3 dB for ACO 11. So, defects of low-noise related to the gradual “taxiing” of the surface. wearing course show earlier than on the common surface, especially in the experimental and the first realized sections of these asphalt mixtures. Figure 6: Comparison of tyre/road noise of mixture ACO 11 and low-noise mixture PA 8 CRmB in rural area under heavy transport and agricultural machinery during six years of lifetime (measured with velocity 50 km/h). Figure 7: Clogging of road surface. Red line in Fig. 6 symbolizes the reference Third-octave frequency spectrum of ACO 11 and equivalent sound pressure level 90 dB for velocity PA 8 better expresses noise emission changes 50 km/h. Common (ACO 11, blue color) and low- (Fig. 8). More detailed differences between the noise (PA 8 CRmB, yellow color) surfaces have surfaces can be traced by this. PA 8 CRmB after approximately the same noise level about 2–3 years laying (red color) has the lowest values and they are after laying without regular maintenance in this significantly lower than others for frequencies above measured location. Noise of ACO 11 was 89.2 dB 800 Hz. In porous asphalt, sound wave propagates 70 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia into the gap, releases part of its acoustic energy (with laying the new pavement (the same asphalt mixture) the generation of heat) and thus reduces the noise at the same place so it is a decrease of 3.1 dB. The reflected back out. Frequencies of ACO 11 (orange renewing of wearing course even with a common color) and PA 8 CRmB (green color) after two years asphalt is also a good solution for road noise of lifetime have very similar course which is in reduction. This exchanged surface reaches reference agreement with the measured noise values in the value 98 dB after two years and then the noise is Fig. 6. The situation is opposite to the previous case slowly increasing to 98.6 dB in five year of lifetime of PA, the gaps are clogged, sound waves don’t get of pavement ACO 11. Low-noise surface has the into them and are immediately reflected. biggest influence on noise reduction in this location. Noise level of noise is 94.1 dB after laying which is 3 dB lower than common pavements after laying. Noise continuously increases to value 96.6 dB in five year of lifetime, so even after five years the noise level doesn’t reach the reference level. The most significant increase in wearing course noise occurs in the first few (2–4) years after laying. Subsequently, the increase of noise gradually slows down provided that no distresses occur. Behavioral curve about the long-term development of road surface noise corresponds to logarithmic rather than a linear course, which is also generally known fact (Haider et al., 2016). Initial void content of asphalt mixtures should be in the range 8–15 vol. % Figure 8: Third-octave frequency spectrum of A- according to TP 259. This value decreases by weighted signal of surfaces ACO 11 (2 years) and gradual presence of dust particles, grease and other PA 8 CRmB (0 and 2 years) for velocity 50 km/h. impurities until the asphalt course approaches void content of a common asphalt mixture with value 5.2. Highway below 4 vol. %. Technically, this asphalt course Fig. 9 shows results for highway road outside of continues to function for several years as a quality rural area with traffic flow above 30,000 vehicles wearing course, but from an acoustic point of view, per day and higher velocity 80 km/h which means its acoustic life will be gradually achieved. So, the that the self-cleaning effect applies here. The results from noise measurements performed in the absence of agricultural machinery on the road Czech Republic correspond with the knowledge represents lower mud pollution and thus less from foreign countries. clogged surface. Red line in Fig. 9 symbolizes the Open wearing course of porous asphalt type is less reference equivalent sound pressure level 98 dB for suitable to use in the Czech Republic due to velocity 80 km/h. There are results of low-noise meteorological conditions (freeze-thaw cycles) with section length 2,500 m (BBTM 8 NH, yellow primarily and in terms of non-mainenance color) and 4,100 m long section of common (clogging). Their initial benefits may be large, but (ACO 11, blue color) wearing course. air gaps become clogged faster and if more frequent regular cleaning is not chosen, they quickly lose their acoustic properties. Asphalt mixture BBTM 8 NH (and also SMA 8 NH) is more suitable for using and higher velocity of vehicles is also benefit due to self-cleaning effect. Tyre/road noise changes over time, so it is not possible to stay at the first measured value as given value for wearing course, but it is necessary to perform repeated measurements. In the area of evaluation of road surface properties at the international level within CEDR the need to Figure 9: Comparison of tyre/road noise of low-noise introduce and monitor noise in the new road mixture BBTM 8 NH and common mixture ACO 11 management system is discussed. At the national during five years of lifetime (measured with velocity level, a decision scheme for the classification of the 80 km/h). overall condition of road is proposed, listing the five main criteria for monitoring condition of the road Dark blue dashed line (100.2 dB) represents noise network: anti-skid properties, roughness, surface level of the damaged common pavement before noise, road load capacity and disorders (damages). replacement, the value 97.1 dB was measured after 71 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia The noise should be monitored but no specific classification levels for evaluation of road surface indicators are prescribed. Determination of noise noise in the Czech Republic. classification of road surfaces is the main aim of the solving project in the national level in the Czech REFERENCES Republic for now. [1] Aksnes, J., Saba, R. G., & Berge, T. (2009). Development and testing of low noise 6. CONCLUSION pavements in Norway. International Conference The CNOSSOS-EU methodology requires for BCR2A’09 (pp. 513–520), Champaign, USA. processing of strategic noise maps to work with Leiden: CRC Press (Taylor & Francis Group). individual types of road surfaces (e.g. SMA, ACO, [2] Anfosso-Ledee, F., & Brosseaud, Y. (2009). BBTM, PA etc.), including its age. Each country has Acoustic monitoring of low noise road its own recipe for each type of asphalt mixture. The pavements. Noise Control Eng., 57(2), 50–62. accompanying table F-4 shows the road surface [3] ASTM International. F2493-20: Standard coefficients which are based on data obtained in the Specification for P225/60R16 97S Radial Netherlands. Many types of these surfaces are not Standard Reference Test Tire. West used in the Czech Republic, the composition of the Conshohocken, PA, USA, 2006(2020). wearing courses used in CZE is similar to those used [4] Babisch, W. (2008). Road traffic noise and in Germany, yet there are differences. However, cardiovascular risk. Noise Health, 10, 27–33. CNOSSOS-EU allows the setting of national coefficients. The exact determination of values [5] Bluhm, G. L., Berglind, N., Nordling, E., & (especially for low-noise wearing course) is up to the Rosenlund, M. (2007). Road traffic noise and hypertension. Occup. Environ. Med., 64, 122– individual EU members due to the high diversity of 126. used technologies and local meteorological conditions. Roads are in the CZE classified into five [6] Buhlmann, E., Schulze, S., & Ziegler, T. (2013). categories (1–5) on the basis of skid resistance, Ageing of the new CPX reference tyres during a pavement surface roughness, a load of road surfaces measurement season. Internoise (pp. 369–376), and damages. Road surface noise has no Innsbruck, Austria. Austrian Noise Abatement Association. classification and there is already a demand for noise assessment for e.g. manufacturers or road managers. [7] Campillo-Davo, N., Peral-Orts, R., Campello- Tire/road noise is a variable parameter and it is Vicente, H., & Velasco-Sanchez, E. (2019). An important to perform many CPX measurements and alternative close-proximity test to evaluate data evaluations of different wearing courses to sound power level emitted by a rolling tyre. Appl. Acoust., 143, 7–18. create classification levels for the CZE. Partial results could then be used as input data for modeling [8] Castaneda, E., Leavings, V. R., Noss, R. F., & within the CNOSSOS methodology, so as for Grace, M. K. (2020). The effects of traffic noise adjustment of the Czech tire/road noise coefficients. on tadpole behavior and development. Urban Ecosyst., 23, 245–253. The noise changes over time and differently for each [9] Cesbron, J., & Klein, P. (2017). Correlation wearing course. These changes of noise in tire/road between tyre/road noise levels measured by the contact can be monitored by regular measurement Coast-By and the Close-ProXimity methods. with the CPX method. Low-noise wearing courses Appl. Acoust., 126, 36–46. have a durable efficiency on highways due to the self-cleaning effect thru the velocity of vehicles. [10] CESTI, Centre for Effective and Sustainable Transport Infrastructure, project Whereas on rural roads, pavements may lose rapidly No. TE01020168 of Technology Agency of the their acoustic properties and behave as a common Czech Republic. Available from: wearing course faster due to clogging which can be http://www.cesti.cz/index.php?page=home&en better observed using the noise spectrum. So, if we =en. want to keep acoustical function of low-noise [11] Cholava, R., & Krivanek, V. (2009). surfaces longer, it is necessary to perform proper Optimization of parameters of road surfaces to maintenance/cleaning, which is not yet the rule in reduce noise burden from transport. the Czech Republic. International Congress on Sound and Vibration Acknowledgments (pp. 1408–1415), Krakow, Poland. Auburn: International Institute of Acoustics and Vibration. This article is financed from the state budget by the Technology agency of the Czech Republic and the [12] Commission Directive (EU) 2015/996 of 19 Ministry of Transport of Czech Republic under the May 2015 establishing common noise assessment methods according to Directive DOPRAVA 2020+ Programme, project 2002/49/EC of the European Parliament and the CK02000121 Determination of values of Council. Off. J. Eur. Comm., L 168, 1–823. 72 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia [13] Czuka, M., Palla, M. A., Morgan, P., & Conter, [26] Kim, K., Shin, J., Oh, M., Jung. J-K. (2019). M. (2016). Impact of potential and dedicated Economic value of traffic noise reduction tyres of electric vehicles on the tyre-road noise depending on residents’ annoyance level. and connection to the EU noise label. Transp. Environ. Sci. Pollut. Res., 26, 7243–7255. Res. Proc., 14, 2678–2687. [27] Klæboe, R. (2011). Noise and health: annoyance [14] Descornet, G. (2000). Low-noise road surface and interference. In J. O. Nriagu (Ed.), techniques and materials. Paper presented at Encyclopedia of Environmental Health (Vol. I, Internoise 2000, Nice, France. pp. 152–163). Amsterdam, London: Elsevier Science. [15] Directive 2002/49/EC of the Europen Parliament and of the Council of 25 June 2002 [28] Kleiziene, R., Sernas, O., Vaitkus, A., & relating to the assessment and management of Simanaviciene, R. (2019). Asphalt pavement environmental noise. Off. J. Eur. Comm., L 189, acoustic performance model. Sustainability, 12–25. 11(10), 2938–2952. [16] Finch, D., Schofield, H., & Mathews, F. (2020). [29] Klompmaker, J., Hoek, G., Bloemsma, L. D., Traffic noise playback reduces the activity and Wijga, A. H., Van den Brink, C., Brunekreef, feeding behaviour of free-living bats. Environ. B., Lebret, E., Gehring, U., & Janssen, N. A. H. Pollut., 263(B), 114405. (2019). Associations of combined exposures to surrounding green, air pollution and traffic noise [17] Francis, C. D., Ortega, C. P., & Cruz, A. (2009). on mental health. Environ. Int., 129, 525–537. Noise pollution changes Avian communities and species interactions. Curr. Biol., 19(16), [30] Kragh, J., Iversen, L. M., & Sandberg, U. 1415–1419. (2013). Road surface texture for low noise and low rolling resistance. Copenhagen: [18] Fyhri, A., & Klæboe, R. (2006). Direct, indirect Vejdirektoratet. influences of income on road traffic noise annoyance J. Environ. Psychol., 26(1), 27–37. [31] Krivanek, V., & Effenberger, K. (2014). Comparison of noise level measurement of road [19] Haider, M., Conter, M., Green, M., Schmidt, B., surfaces in the Czech Republic using the Close Sandberg, U. (2016). Status of the EU-Project Proximity method. Transactions on Transport ROSANNE. Transp. Res. Proc., 14, 2946–2955. Sciences, 7(3), 109–116. [20] ISO 11819-2:2017. Acoustics – Measurement [32] Krivanek, V., Markova, P., & Spicka, L. (2016). of the in-fluence of road surfaces on traffic noise Verification of the acoustic parameters of – Part 2: The close-proximity method. Geneva: Close-Proximity method. Akustika, 26, 13–19. International Organization for Standardization, 2017. [33] Krivanek, V., Pavkova, A., Togel, M., Jedlicka, J. (2016). Cleaning low-noise surfaces as a basic [21] ISO 11819-3:2017. Acoustics – Method for condition for improving pavement’s acoustic measuring the influence of road surfaces on absorption capability. Arab. J. Sci. Eng., 41, traffic noise – Part 3: Reference tyres. Geneva: 425–431. International Organization for Standardization, 2017. [34] Krivanek, V., Markova, P., & Hablovicova, B. (2020). An economic comparison of the [22] ISO/TS 13471-1:2017. Acoustics – ordinary and low-noise wearing course during Temperature influence in tyre/road noise the lifetime of roadway. Akustika, 36, 2–15. measurements – Part 1: Correction for temperature when testing with the CPX method. [35] Kwiatkowski, K., Oslakovic, I. S., Hartmann, Geneva: International Organization for A., ter Maat, H. (2014). Potential impact of Standardization, 2017. climate change on porous asphalt with a focus on winter damage. Paper presented at Transport [23] Jafari, Z., Kolb, B. E., & Mohajerani, M. H. Research Arena 2014, Paris, France. (2018). Chronic traffic noise stredd accelerates brain impaitment and cognitive decline in mice. [36] Larsen, L. E., & Bendtsen, H. (2001). Costs and Exp. Neurol., 308, 1–12. perceived noise reduction of porous asphalt pavements. Internoise (pp. 1859–1864), Hague, [24] Jafari, Z., Kolb, B. E., & Mohajerani, M. H. Netherlands. Red Hook, NY: Curran (2020). Noise exposure accelerates the risk of Associates, Inc. cognitive impairment and Alzheimer’s disease: Adulthood, gestational, and prenatal [37] Licitra, G., Teti, L., & Cerchiai, M. (2014). A mechanistic evidence from animal studie. modified Close Proximity method to evaluate Neurosci. Biobehav. Rev., 117, 110–128. the time trends of road pavements acoustical performances. Appl. Acoust., 76, 169–179. [25] Kephalopoulis, S., Paviotti, M., & Anfosso- Lédée, F. (2012). Common noise assessment [38] Licitra, G., Teti, L., Cerchiai, M., & Bianco, F. methods in Europe (CNOSSOS-EU). (2017). The influence of tyres on the use of the Luxembourg: Publications Office of the CPX method for evaluating the effectiveness of European Union. a noise mitigation action based on low-noise road surfaces. Transp. Res. D Transp. Environ., 55, 217–226. 73 Blanka Hablovičová, Vítězslav Křivánek, Petra Marková YRS 2021 REDUCTION OF NOISE ON ROAD SURFACES (WEARING COURSES) IN THE CZECH 15. -17. September REPUBLIC Portoroz, Slovenia [39] Licitra, G., Moro, A., Teti, L., Del Pizzo, A., [50] Tonin, R., & Szabo, A. (2014). RONDA – CPX Bianco, F. (2019). Modelling of acoustic ageing trailer initial test results. Internoise (pp. 711– of rubberized pavements. Appl. Acoust. 146, 716), Melbourne, Australia. Red Hook, NY: 237–245. Curran Associates, Inc. [40] Miljkovic, M., & Radenberg, M. (2012). Thin [51] Transport Research Centre (2011). noise-reducting asphalt pavements for urban Methodology for measurement using Statistical areas in Germany. Int. J. Pavement Eng., 13(6), Pass-By method and Close Proximity method. 569–578. Brno: Transport Research Centre. [41] Ohrstrom, E., Hadzibajramovic, E., Holmes, [52] Transport Research Centre (2014). M., & Svensson, H. (2006). Effects of road Methodology for measuring and evaluation of traffic noise on sleep: Studies on children and noise on roads. Brno: Transport Research adults. J. Environ. Psychol., 26(2), 116–126. Centre. [42] Okokon, E. O., Yli-Tuomi, T., Turunen, A. W., [53] Transport Research Centre (2017). Long-term Tiittanen, P., Juutilainen, J., & Lanki, T. (2018). noise assessment of road pavements. Brno: Traffic noise, noise annoyance and psychotropic Transport Research Centre. medication use. Environ. Int., 119, 287–294. [54] Vaitkus, A., Vorobjovas, V., Jagniatinskis, A., [43] Peeters, B., Reinink, F., & Van Vliet, W. (2018). Andriejauskas, T., & Fiks, B. (2014). Close proximity (CPX) round robin test 2017. Peculiarity of low noise pavement design under Paper presented at Euronoise 2018, Heraklion, Lithuanian conditions. Balt. J. Road Bridge Crete – Greece. Eng., 9(3), 155–163. [44] PIARC, Working Group II of the Technical [55] Valentin, J., Mondschein, P., Bures, P., & Committee TCE2 Environmental Krivanek, V. (2017). Technical specification Considerations in Road Projects of the World 259 – Asphalt mixtures for wearing courses with Road Associacion (2019). Traffic noise – best reduced noise levels. Prague: Ministry of practice guide. France: PIARC. Transportation. [45] Pirrera, S., De Valck, E., & Cluydts, R. (2014). [56] Van Essen, H., Van Wijngaarden, L., Schroten, Field study on the impact of nocturnal road A., Sutter, D., Bieler, C., Maffii, S., Brambilla, traffic noise on sleep: The importance of in- and M., Fiorello, D., Fermi, F., Parolin, R., & outdoor noise assessment, the bedroom location Beyrouty, K. E. (2019). Handbook on the and nighttime noise disturbances. Sci. Total external costs of transport. Luxembourg: Environ., 500–501, 84–90. Publications Office of the European Union. [46] Sandberg, U. (1999). Low noise road surfaces – [57] Van Leewen, H., Kok, A., & Reubsaet, J. A state-of-the-art review. J. Acoust. Soc. Jpn. (2007). The uncertainty of acoustical (E), 20(1), 1–17. measurements on road surfaces using the CPX- method. Internoise (pp. 4394–4399), Istanbul, [47] Sandberg, U., & Ejsmont, J. A. (2002). Turkey. Red Hook, NY: Curran Associates, Inc. Tyre/road noise reference book. Kisa: Informex. [58] Vazquez, V. F., Teran, F., Luong, J., & Paje, S. [48] Sandberg, U., Buhlmann, E., Conter, M., & E. (2019). Functional performance of stone Mioduszewski, P. (2016). Improving the CPX mastic asphalt pavements in Spain: Acoustic method by specifying reference tyres and assessment. Coatings, 9(2), 123–137. including corrections for rubber hardness and temperature. Internoise (pp. 4913–4923), [59] Vieira, T., & Sandberg, U. (2019). Close Hamburg, Germany. Red Hook, NY: Curran proximity (CPX) round robin test: Comparison Associates, Inc. of results from four different CPX trailers measuring noise properties of 10 Swedish road [49] Stansfeld, S. A., Clark, C., Cameron, R. M., surfaces. Internoise (pp. 7176–7187), Madrid, Alfred, T., Head, J., Haines, M. M., Van Kamp, Spain. Red Hook, NY: Curran Associates, Inc. I., Van Kempen, E., & Lopez-Barrio, I. (2009). Aircraft and road traffic noise exposure and [60] World Health Organization (2018). children’s mental health. J. Environ. Psychol., Environmental Noise Guidelines for the 29(2), 203–207. European Region. Copenhagen: WHO Regional Office for Europe. 74 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION* Knut Johannes L. Hartveit, Researcher Askill H. Halse, Chief Research Economist Nina Hulleberg, Researcher Institute of Transport Economics Gaustadalléen 21, 0349 Oslo, Norway klh@toi.no ABSTRACT Policy makers in urban areas are in continuous search for policies that can improve the competitiveness of public transport vs. private cars. In this paper, we propose a framework for evaluation of smaller interventions and study the construction of a new bus lane, which was built to improve travel time and reliability of city buses in Bergen, Norway. To identify the effects of such an intervention, we use disaggregated operational data on travel time measured at each bus stop. This high data resolution enables us to identify the effects exactly where and when they occur. We use different measures of travel time delay and variability to assess whether there are any problems before the intervention was done, and if these vary in certain areas or time periods. Then, we utilize the fact that the intervention is done in only one of the bus routes’ directions to create a control group. That enables us to conduct difference-in-differences analyses to estimate the causal effects of the intervention. The analyses show no apparent effect from the new bus lane. This is largely due to an installation of new active signal prioritizing technology in a signal-controlled intersection just over a month after the opening of the bus lane. This means that if the bus lane had a longer-term effect, this effect would be lost in the control variable we use to account for this technology. Keywords: Public transport, transport economics, econometrics, travel time reliability 1. BACKGROUND delays and congestion driving, will hence be less appealing to people. In Norway there is a zero-growth goal which states that all growth in person transport in the larger urban Additionally, travel time seems to be the most areas is to be absorbed by public transport, bicycling important aspect of inter-modal competition, and walking (Tønnesen et al. 2019). Therefore, it is followed by fare, access/egress/transfer time, and essential that policy makers and public transport number of interchanges (Fearnley et al. 2018; Lunke authorities makes the public transport as attractive as et al., 2021a; Lunke et al., 2021b). possible for users of private cars. So, in order to achieve the zero-growth goal, it is Time is a scare resource, so if a travel has longer paramount to be able to identify projects that reduce duration, that has a cost for the traveller because that travel time and congestion, while at the same time is time the traveller could have spent doing being economically viable. If the authorities can something else. This cost of travel time is normally choose projects that improve the attractivity of called value of travel time, which depends on many public transport and active transport in an factors, like delays or congestion. Time spent in economical sustainable way, the zero-growth goal is congestion has a higher value of travel time easier achievable. compared to free-flowing traffic (see e.g. Wardman In this paper, we propose a framework for evaluating et al. 2016). A public transport sector with more of smaller interventions using disaggregated * The paper departed from a larger, but more general report; Hartveit et al. (2020). 75 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia operational data for buses on travel time measured 2. DATA AND DESCRIPTIVE at each bus stop. ANALYSIS This high data resolution enables us to identify the The main data source we use is disaggregated effects exactly where and when they occur, operational data on travel time measured at each bus compared to if we were to use total travel time, stop received from the public transport authority where the effects could be hidden due to timetable (PTA). This high-resolution data enables us to slack. It also enables us to illustrate and test which identify the effects exactly where and when they measures of travel time and reliability that should be occur, compared to if we were to use total travel used in these analyses. time, where the effects could be hidden due to The intervention we evaluate was done in Bergen, timetable slack. Norway to improve the travel time and reliability for The data consists of when the buses are scheduled to city buses. We do this to give insights into how one arrive and depart bus stops (time and date), when the can measure, and possibly select, projects for buses actually arrive and depart bus stops, journey upgrades in a public sector with limited funds. ID, stop ID and information about passengers More specifically, we study if the construction of a getting on and off the buses. dedicated bus lane at Haukås, north of Bergen has an To ensure that we capture all seasonal effects, like effect on travel time and reliability for city buses. holidays and vacations, we have chosen to have a The new bus lane goes through an intersection along year of data before the construction period (the the E39 in the northbound direction, and continues before period) and a year of data after the for approximately 350 meter north of the construction period. intersection, illustrated by the black marking in The bus lane was under construction from Figure 1 below. 01.09.2017-30.09.2018 (the construction period). Hence, our dataset spans from 01.09.2016 to Breistein 01.10.2019, with the before period being from 01.09.2016 to 31.08.2017 and the after period being North Myrsæter 01.10.2018 to 01.10.2019. We study two of the bus lines that are influenced by the intervention, called line 36 and line 37. These lines are chosen because they are similar in terms of Haukåsvegen frequency and passenger numbers, but they also have some differences in their routes, which adds to Vikaleitet our understanding of the intervention’s effect. Travelling northwards, line 36 drive along E39 the whole way, while line 37 drive via Vikaleitet, then onto E39, and then drive off E39 towards Breistein (reverse order when driving southwards). ASP intersection In the influenced area, there are four bus stops we New bus lane include in our analysis (Vikaleitet, Vågsbotn, Bus stop northwards Haukåsvegen and Myrsæter), illustrated in Figure 1. Bus stop southwards Vågsbotn We choose these stops because they are the stops Source: Map from OpenStreetMap. that give us the information about travel time and reliability we need for our analysis. Figure 1: Illustration of the influenced area. The road section northwards from the Vågsbotn/ There was no dedicated bus lane before the Vikaleitet stop (south) to the Myrsæter stop (north) construction, so the buses are now driving adjacent is the road section being affected by the new bus to mixed traffic for the 350 meters opposed to earlier lane. The stop Haukåsvegen northwards (blue driving in the same lane as the other traffic. square in Figure 1) is situated along the new bus If there was a congestion challenge or travel time lane. Hence, the road section that is affected by the challenge in the area of analysis, we would expect bus lane can be divided into two smaller sections: the buses to drive faster after the construction of the Vågsbotn/Vikaleitet to Haukåsvegen (southern) and dedicated lane and reduce the travel costs for its Haukåsvegen to Myrsæter (northern). passengers. The northern one is significantly shorter than the southern one, while the southern one also has a signalled intersection, a junction and different routes for the buses. Line 36 drive on the European route 76 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia E39, while line 37 drive on a slower, tighter road and intersections, this would also affect the results in our has to enter E39 through the intersection. analysis if we do not account for it. 2.1. Selection of control group Even though the installation happened at the same time, we would expect different effects from the Choosing or finding a suitable control group is southern and northern ASP. The one south of the bus unfortunately not easy, and this applies here as well. lane is between the two stops on the southern sub- There are multiple aspects and reasons for this. First, section, so if ASP has an effect, it will directly affect there is a geographic challenge, where types of roads the driving time there, and influence our estimates of and infrastructure could be different between the the new bus lane. The one north of the bus lane can control and treatment lines. affect the general traffic flow in the area, so it should There is also a demographic challenge, where be accounted for, even though we do not expect it to different demographic areas could have different have nearly as strong an impact as the southern one. demand for public transport. It could also be Another factor is road tolls. Road tolls, or road use temporary changes along one of the bus lines, like pricing, charges the people driving through certain road maintenance or re-routing, that could interfere areas a charge to pass through, and reduce traffic with the comparability of the bus lines. congestion (Hosford et al. 2021). We entertained the idea of using a similar line as a Three different changes in the road tolls happened in control group, but there were no other lines with the Bergen from 2016 to 2020. Time differentiated road same frequency, number of passengers, tolls were implemented 01.02.2016, demographics and driving conditions. environmentally differentiated road tolls were Therefore, we utilize the fact that the intervention implemented in 01.06.2018 and there were was only done in one of the directions (northwards) established more road toll stations which were and use the other direction (southwards) as the operational from 06.04.2019. If these toll schemes control group. This means that both groups are have the anticipated effect, they will disturb the expected to have the same passenger growth and effect of the measure in our analysis if we do not being affected by other factors, like weather account for it. conditions, in the same way. Since new toll stations or changes in the tolls are At the same time, this means that the control and operational from a specific date, we assume that treatment group could have different driving most of the effects arise close to the date they are conditions depending on time of day, but this is can operational. With this assumption, we do not include be accounted for in the analyses. the time differentiation change in our analyses, as it happened more than eight months before the before 2.2. Other relevant factors period. The other two, on the other hand, are in the It is important to know whether there has been other time period we are analysing, and are therefore interventions or occurrences that happened in the included in the analyses. period we are analysing that could affect driving 2.3. Descriptive analysis time and reliability. Knowing this can enable us to control for these, and hence isolate the effect of the In the descriptive analysis, we use the whole period, intervention we are analysing. meaning the before period (09.2016-09.2017), the construction period (09.2017-10.2018) and the after Therefore, in addition to the operational data, we period (10.2018-11.2019). have collected information and knowledge about overlapping events from the PTA, but also other Utilizing the richness of the data, we calculate the official reports and information channels. variables travel time and driving time. Driving time is the time the bus takes driving between the bus The first important other factor we need to be aware stops. Travel time is the time the bus takes driving of, is the installation of technology for active signal from the first bus stop to the last stop in the affected prioritizing (ASP) of public transport in two signal- area, including driving time and time spent on the controlled intersections just south and north of the bus stops. bus lane (the two green squares in Figure 1). With the new technology, sensors before the intersection This distinction is made due to the fact that the pick up when buses are approaching and ensure that driving time variable could tell us about the the buses get a green light earlier than with the older challenges along the road, while the travel time also traffic adjusted signals. says something about the accessibility of the bus stops in terms of stopping, letting passengers on and The installation happened almost two months after off, starting to drive and get back onto the road with the opening of the bus lane, so if ASP works and much traffic. Thus, we might capture some slack reduce the wait time for buses in signal controlled using the travel time variable. 77 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia The travel time and the driving time for line 36 in reductions are in the summer holidays, which the three periods of analysis in monthly average, normally has less traffic and travel time challenges. separating on buses going north and south, is Looking at the after period, there is a clear reduction illustrated in Figure 2 and Figure 3 below. We see in driving and travel time for the treatment group, that the red (north) and blue (south) lines seem to while the control group looks to be unaffected. This follow the same trend in the before period. indicates a strong effect from the intervention. The dip we see in July in the before period, is the However, it should be noted that the treatment group summer holidays in Norway, so less traffic in the first month in the after period seemingly has a congestion is to be expected. However, the fact that slight reduction in driving and travel time, before the is seems to give such a strong effect on the travel second month has a large reduction. We examine time of buses, could indicate that there are some this further in the regression analyses. challenges. Comparing the before and after period visually, it looks like the lines are flatter, converging and at a Before Construction After lower average in the after period. This could indicate that the bus lane has an effect on the travel time. It also seems to be a general downwards going trend (shorter travel times), so we should include a control for this in the regression analysis. Before Construction After Figure 4: Travel time of line 37 in seconds, monthly average, over the course of the analysis period Before Construction After Figure 2: Travel time of line 36 in seconds, monthly average, over the course of the analysis period Before Construction After Figure 5: Driving time of line 37 in seconds, monthly average, over the course of the analysis period 3. METHODOLOGY In this project, we suggest a framework for an ex ante evaluation of whether an intervention in a certain area could be beneficial and for an ex post evaluation of whether the intervention did have an Figure 3: Driving time of line 36 in seconds, monthly effect. We call this the before analysis and the average, over the course of the analysis period regression analysis, respectively. For line 37, we see the travel and driving time in In the before analysis, we want to figure out whether Figure 4 and Figure 5. Again, the control group there are some travel time or reliability challenges in (south) and treatment group (north) in the before the area, before the bus lane was constructed. We period seems to be following the same trend, but also explore which measures of travel time or maybe slightly less than line 36. reliability are the best suited for these sorts of We see that the travel and driving time of the evaluations. northwards going buses are more volatile. This The regression analysis helps explaining what the could be due to travel time challenges, as the clear patterns we saw in the descriptive analysis are 78 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia caused by. Like with the descriptive analysis, we use before and difference after is then the treatment the whole period (before, construction, after). effect (the red arrow). More specifically, the regression analysis we do is In our analysis, we also have a construction period called Differences-in-Differences (DinD). DinD is a between the before and after period. We control for statistical method used for evaluating the effect of an this in our analysis and compare the before and after intervention, a new law or regulation, or a natural period. event. A classic example of a DinD-analysis is Card In the following, we present the before analysis and and Krueger (1994). They studied how an increase then the results from the regression analysis. in the minimum wage affected employment by utilizing that New Jersey’s minimum wage rose, 4. BEFORE ANALYSIS while neighbouring Pennsylvania had a constant minimum wage. Card and Kruger did the DinD- In this analysis, we study how the different measures analysis by comparing the employment in fast food on travel time and reliability fare and what the restaurants in New Jersey (treatment) with the one in different measures highlight. We also use these Pennsylvania (control) before and after the increase. different measures to study whether it seems to be challenges related to driving time and reliability in These interventions or events makes it possible to the before period. These are presented in the identify a control group and a treatment group, following sub-sections. where the treatment group is the group that was affected by the event (treatment). Observing the two The bus lane was constructed in the northwards groups before and after the treatment, we can facing direction to improve travel time, so if the bus measure the change in outcome of the two groups. lane indeed was constructed in an area with travel time challenges, we would expect to see this in our Furthermore, it is essential that the control group and before analysis. the treatment group are as similar as possible. This is due to a fundamental assumption in this 4.1. Delay relative to the timetable methodology, called the assumption of a common trend. It says that the change in outcome for the A way of studying reliability is to see if the buses control group would be the same as the treatment drive according to the timetable. A challenge we group if it had received the same treatment. If it have here is that the stops according to the timetable holds, we can estimate the causal effect of the are in minutes while we have actual arrival or intervention using DinD. departure times at bus stops down to milliseconds. This means that we must use minutes as our analysis, Outcome and because of that, we do not capture variations in Ti me of treatment seconds due to rounding. When we consider that the average trip length between these stops are between one and five minutes, it is clear that this is not an Di fference before optimal measure to use in our regression analysis where we expect changes in travel time of seconds, not minutes. Trea tment effect However, it does give a general overview of how many buses that are delayed at the bus stops. In Di fference after Figure 7, we show the share of buses being on time (delay of 0-3 minutes) and the share of different levels of being delayed at the bus stops compared to Time the time table. Figure 6: DinD illustration (inspired by Angrist and It shows that most of the departures are recognized Pischke (2008)) as being on time when looking at the before period, however, we do see some delays with around 20 % An illustration of a DinD-analysis is showed in of the buses being delayed three or more minutes. Figure 6. The dotted lines express an average level of driving time for the treatment group and the Comparing the buses going northwards and control group. If the treatment did not have any southwards, we do not see a big difference. effect, we would expect to have the same difference between the treatment group and the control group. If the treatment has an effect, we could get something displayed in the graph, where we see there has been a downwards shift in the average travel time. The difference between the difference 79 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia Northwards Southwards We have thus found that a significant share of the northwards facing buses on line 37 is delayed. 4.2. Variation in driving time Another way to analyse reliability and travel time challenges is to see how the actual driving time of the buses between stops vary. This gives us a more precise way of looking at challenges along the road, 0-3 min delay as discussed in the descriptive analysis. 3-6 min delay 6-10 min delay Figure 9 shows the variation in driving time between Over 10 min delay bus stops for line 36 northwards and southwards in a box plot for the whole day. A box plot is a diagram that illustrates the median driving time (the horizontal line inside the coloured Figure 7: Share of departures that are delayed at box) and how much variation there is in the data. The departure time from the studied bus stops compared whiskers of the boxes show the 90th (top) and 10th to the timetable of line 36 (bottom) percentile, while the top of the box shows Figutre 8 shows the share of departures that are the third quartile and the bottom shows the first delayed at departure time from a certain bus stop quartile. The coloured box hence shows the compared to the timetable of line 37. interquartile range. Like we saw for line 36, we see that a large Looking at the figure, we see that the road section proportion of the buses are departing on time, but between Vågsbotn and Haukåsvegen (blue and there is now a large difference between buses brown box) have the highest driving time. This is the driving northwards and southwards. longest road section, so that is expected. However, we see that the variation in the northwards driving The buses going north from Vikaleitet are almost at buses is larger than the ones driving southwards, the same level of delay as the buses going indication some challenges here. southwards, but when the buses leave the next stop North South (Haukåsvegen), the buses are significantly more delayed, and keep almost the same delay to the next Average stop (Myrsæter). In other words, the buses seem to get delayed when going from Vikaleitet to Haukåsvegen. Looking at the same area for the southwards facing buses, they also have an increase in delay, going from Haukåsvegen to Vikaleitet, but it is smaller. Northwards Southwards 0-3 min delay 3-6 min delay 6-10 min delay Over 10 min delay Figure 9: Variation in driving time (in seconds) between bus stops for line 36 Figure 8: Share of departures that are delayed at Furthermore, we observe that the average value is departure time from the studied bus stops compared to the timetable of line 37 somewhat higher than the median value. This means that there are some observations in the data set with a very long travel time, which draw up the average value of the section. This is not an unusual result, but 80 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia it illustrates that one needs to be aware of the This is illustrated in Figure 9 and Figure 10, and we difference of using median and average values. see that the sub-sections we previously identified with the largest challenges also has the largest Figure 10 shows the variation in driving time difference between P10 and the average driving between stops for line 37. We see that the road time. section Vikaleitet to Haukåsvegen (blue box) in the northward direction that has a large variation in 4.4. Summary of the before analyses driving time, compared to the other. This indicates quite clearly that there are driving time challenges In this chapter, we have focused on using different on this road section. measures to discover whether there are driving time challenges in the before period in the areas that North South would later be affected by the new bus lane. Both delay in relation to the time plan, variation in driving time and driving time in relations to zero driving suggest that there are travel time challenges for line 37 in the northwards facing direction on the first section (Vikaleitet – Haukåsvegen). There are some signs of the same for line 36, but this is to a lesser extent, especially if we consider share of departures delayed compared to the time plan. The second section (Haukåsvegen – Myrsæter) does not have the same clear signs of challenges, for both line 36 and line 37. From our before-analysis it does look like there are clear reasons for trying to improve the travel times of the city buses in this area. 5. REGRESSION ANALYSIS In this chapter, we perform regression analyses to assess if the intervention had an effect on travel time and improved the situation. Like with the descriptive analysis, we use the whole period (before, Figure 10: Variation in driving time (in seconds) construction, after) and define the treatment group between bus stops for line 37 as the buses going northwards and the buses going For both line 36 and line 37, we observe a larger southwards as the control group. variation in driving time on the southern sub-section Our base Differences-in-Differences (DinD) (Vågsbotn/Vikaleitet to Haukåsvegen), compared to regression is the following: the northern sub-section (Haukåsvegen to Myrsæter). Additionally, this variation is larger for 𝐷𝑟𝑖𝑣𝑖𝑛𝑔 𝑡𝑖𝑚𝑒 = 𝛼 + 𝛽1𝑁𝑜𝑟𝑡ℎ + 𝛽2𝐴𝑓𝑡𝑒𝑟 + 𝛽3𝑁𝑜𝑟𝑡ℎ line 37 than for line 36, and the difference between ∗ 𝐴𝑓𝑡𝑒𝑟 + 𝜖 the northern route and southern route is also larger (1) for line 37, which correspond well with what we found when using the measure delay relative to the Where α is the constant, β1 shows the effect of timetable. driving northwards or southwards, β2 shows the effect of it being after the construction or not, β 4.3. Driving time compared to zero- 3 is the DinD estimator which shows the effect of the driving intervention and 𝜖 is the error term. Here, we focus on driving time and zero-driving. By As discussed earlier, there have been other zero-driving, we mean driving time without delays, interventions and changes happening in Bergen at which in our analysis is defined by the driving time the same time. To correct for the possible effects of the 10th percentile. from changes in the road toll structure, we include The advantage of measuring deviations from the indicator variables for new road toll stations ( New zero-driving the fact that we have this data down to toll stations) and environmentally differentiated seconds, and that median or average driving time road tolls ( Environmental differentiation). will include some delays (to different extents) while Another change happening in the period of this zero-driving will be close to or equal to free flow analysis is the installation of new technology in conditions. signal-controlled intersections. This happened 81 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia during the second the month after the opening of the North ∙ ASP variable. Due to line 36 having a bus lane, so it must be accounted for, and is included reduction of travel time from the bus lane and no as an indicator variable ( ASP). effect from the ASP, supports this possibility. Furthermore, knowing the transport system, we The variable “North ∙ Construction” shows whether know that there normally are daily variations, there was a change in the construction period. We do weekly variations and season effects, so this should not get a significant result for line 36, but we get a also be accounted for. We include indicator significant increase in driving and travel time for variables for peak hours (Rush hours) with more line 37, approximately 3.8 seconds for both. traffic. The weekends (Weekend) are normally less Looking at the descriptive graphs in Figure 4 and trafficked and, like illustrated earlier, the summer Figure 5, we can see traces of this. vacation in July (July) is also likely to have an effect. Time dependent variables act as expected; driving Additionally, like we saw in the descriptive analysis, and travel time is lower in July and in weekends, a time trend is likely, so we include that as well as a while there are significant increases in driving and control variable (Time trend). travel time during rush hours. The regression results for line 36 and 37, with If we assume that the bus lane has an immediate driving time and travel time, are displayed in Table effect, and then can use the North ∙ After- 5. The most important numbers in the table is coefficients, line 36 seems to have a reduction in marked with a bold font. driving time, while line 37 has an increase in driving The variable “North ∙ After” shows the estimated time. This seems like an implausible result due to the effect of the intervention. It tells us that there was a fact that the two lines are driving through the same significant reduction in driving time and travel time bus lane, even though they have some differences for line 36 northwards as a consequence of the new before the junction and bus lane. bus lane, 7.3 and 6.8 seconds respectively. On the other hand, if we assume that the effect of the Somewhat surprisingly, we see that the driving time bus lane has a more gradual effect, it is more likely and travel time for line 37 northwards seem to that both of the bus lines has a reduced driving and increase, 3.1 and 3.6 seconds respectively. travel time, but that the ASP-coefficient includes However, because we know that the ASP-some of this effect. technology was introduced just after the opening of Lastly, it could very well be that these two the bus lane, we need to be careful in our interventions are multiplicative, meaning that the interpretation of the North ∙ After-variable. reduction in driving time from one of the The variable “North ∙ ASP” shows whether interventions makes the other intervention more something happened with driving time and travel effective, but it could also be that they reduce each time northwards in the period after the ASP- other’s effect. This is something we have not technology was installed. This is an important studied, but it could be a factor in our results. control of this analysis, as both line 36 and line 37 drive through the southern signal-controlled intersection with the new ASP-technology between the stops Vikaleitet/Vågsbotn and Haukåsvegen. It shows that there was a large and significant reduction in driving and travel time for line 37, 33.8 and 35.4 seconds respectively. No significant effect for line 36 is found. This gives us a strong reason to believe that the ASP-technology helps line 37, that is driving onto E39 from a side road, to get through the intersection faster than before (see Figure 1 for repetition of locality). Line 36, on the other hand, was already driving on E39, and probably did not have to wait much for a green light before the measure, so even if the ASP did increase the efficiency of the intersection, it would not affect the driving time. This means that if we had not controlled for the installation of ASP, we would have overestimated the effect of the bus lane. At the same time, if the effect of the new bus lane is not immediate, some or all of the effect of the bus lane will be caught in the 82 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia Table 1: Regression results The before analysis shows that the selected area Line 36 Line 37 seems like sensible area to try to improve the travel time. It demonstrates that line 37 has travel time DEPENDENT VARIABLE DEPENDENT VARIABLE challenges in the northwards facing direction, INDEPENDENT Driving time Travel time Driving time Travel time especially southern sub-section of the influenced VARIABLE area. Line 36 seems to have some challenges in the North 11.507*** 22.501*** 46.798*** 37.314*** same area, but not as large. (0.694) (0.844) (0.453) (0.537) After period -0.739 -2.687 -3.735*** -5.946*** This sub-section is the part of the area where the (1.475) (1.794) (1.195) (1.417) buses are influenced by the installation of the ASP- technology in the after period. The other section of North ∙ After -7.278*** -6.750*** 3.121*** 3.559** the affected area does not seem to have any (1.466) (1.783) (1.183) (1.403) particular challenges in terms of travel time and Construction 1.007 0.715 -0.786 -1.656** period delays. None of the bus lines seem to have any (0.919) (1.118) (0.699) (0.829) challenges when driving southwards. North ∙ 0.415 0.834 3.825*** 3.766*** The regression analysis shows that the driving and Construction travel time is significantly lower after the opening of (0.877) (1.066) (0.625) (0.742) the new bus lane. However, due to the installation of ASP -2.714*** -2.267* -2.329** -1.574 the ASP-technology that got installed just over a (1.049) (1.276) (0.909) (1.078) month after the opening of the bus lane, we struggle North ∙ ASP -1.756 -2.530 -33.800*** -35.382*** to identify whether the reduction in driving time is (1.398) (1.701) (1.181) (1.401) due to the bus lane, the ASP-technology or a New toll combination. 0.813 1.955** 2.131*** 3.206*** stations (0.684) (0.832) (0.552) (0.655) Combining the effect sizes of the intervention and Environmental the ASP, we get a reduction in driving time and 0.786 3.123*** 2.971*** 4.256*** differentiation travel time for the two bus lines. Line 37 has the (0.754) (0.918) (0.585) (0.694) largest reduction, which is not surprising due to 37 July -7.170*** -7.446*** -8.300*** -10.527*** being the line with the most driving time variance (0.773) (0.940) (0.548) (0.649) and delays in the before period. Time trend 0.002 0.005* 0.005*** 0.007*** 6.1. Reservations and further research (0.002) (0.003) (0.002) (0.002) The data and methods used in this article seem to be Weekend -7.097*** -10.223*** -13.928*** -15.107*** well suited to measure travel time and delays, and to (0.441) (0.536) (0.324) (0.384) identify effects of interventions on short road Rush hours 12.592*** 18.695*** 22.976*** 26.387*** sections. (0.327) (0.398) (0.264) (0.313) What we do not know as much about is how possible Constant 157.161*** 106.264** 86.897*** 68.092** time savings would be realized. Would it result in (37.912) (46.112) (27.773) (32.938) more waiting time at bus stops or slack in the Observations 42,223 42,223 61,040 61,040 timetables? The first does not seem to be the case, R2 0.069 0.113 0.391 0.268 because then the variable travel time would have Adjusted R2 0.069 0.113 0.390 0.268 been more constant while the variable driving time * illustrate 10 % significance level, ** 5 % significance level, *** 1 would have been more varying, however, we cannot % significance level be sure. This is possible to study more but could be 6. SUMMARY demanding due to the nature of the changes in the time plan; they are adjusted more sporadically and is In this paper, we have proposed a framework for not easy to pin to a certain intervention. how to decide where to do an effort in improving travel times and reliability and how to evaluate these Furthermore, we have in this article only studied a efforts or improvements. single intervention, so the external validity of this study could possibly be challenged. We have The framework is utilized on an intervention to focused on developing a framework for evaluation improve the travel time and reliability of city buses and would strongly recommend doing more analyses at Haukås north in Bergen, Norway. of effects and economic profitability. This would be Highly detailed operational data on travel time for important both for getting better knowledge about buses allowed us to use different measures of travel how the public transport system works, and to time and reliability, and to capture small deviations ensure that the best steps are being taken. in driving time. 83 Knut Johannes L. Hartveit, Askill H. Halse, Nina Hulleberg YRS 2021 INTERVENTIONS TO IMPROVE PUBLIC TRANSPORT TRAVEL TIME AND RELIABILITY 15. -17. September IN A NORWEGIAN CITY, AN ECONOMETRIC EVALUATION Portoroz, Slovenia REFERENCES [5] Hosford, K., Firth, C., Brauer, M., & Winters, M. (2021). The effects of road pricing on [1] Angrist, J. D., & Pischke, J. S. (2008). Mostly transportation and health equity: a scoping harmless econometrics: An empiricist's review. Transport Reviews, 1-22. companion. Princeton university press. [6] Lunke, E.B., Fearnley, N., Aarhaug, J., 2021a. [2] Fearnley, N., Currie, G., Flügel, S., Gregersen, Commute trips in Norwegian cities: Data F. A., Killi, M., Toner, J., & Wardman, M. combining trip characteristics and revealed (2018). Competition and substitution between mode choice, Data in Brief, public transport modes. Research in https://doi.org/10.1016/j.dib.2021.107319 Transportation Economics, 69, 51-58. [7] Lunke, E.B., Fearnley, N., Aarhaug, J., 2021b. [3] Card, David & Krueger, Alan B, 1994. Public transport competitiveness vs. the car: "Minimum Wages and Employment: A Case Impact of relative journey time and service Study of the Fast-Food Industry in New Jersey attributes, Research in Transportation and Pennsylvania," American Economic Economics, Review, American Economic Association, vol. https://doi.org/10.1016/j.retrec.2021.101098 84(4), pages 772-793, September. [8] Tønnesen, A., Krogstad, J. R., Christiansen, P., [4] Hartveit, K. J. L., Hulleberg, N., Halse, A. H., & Isaksson, K. (2019). National goals and tools Fearnley, N., & Ingebrigtsen, R. Effekter av to fulfil them: A study of opportunities and framkommelighetstiltak for kollektivtransport. pitfalls in Norwegian metagovernance of urban TØI rapport 1811/2020 mobility. Transport Policy, 81(C), 35-44. [9] Wardman, M., Chintakayala, V. P. K., de Jong, G. 2016. Values of travel time in Europe: Review and meta-analysis. Transportation Research Part A, 94, 93-111. 84 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Doron Hekič Andrej Anžlin, PhD Assist. Prof. Aleš Žnidarič ZAG Slovenian National Building and Civil Engineering Institute Dimičeva 12, Ljubljana, Slovenia doron.hekic@zag.si, andrej.anzlin@zag.si, ales.znidaric@zag.si ABSTRACT Bridge Weigh-in-Motion (B-WIM) systems are installed on the bridges to weigh the passing vehicles in motion without traffic interruption. Besides weighing the freight traffic with equivalent accuracy to the pavement WIM, they can be designed to monitor bridge performance under traffic loading simultaneously. This is characterised by performance indicators (PI), most often influence lines, dynamic amplification factors (DAF) and load distribution factors. If suppose these PIs are adequately extracted from B-WIM measurements, the B-WIM takes the role of a Structural Health Monitoring (SHM) tool, in which the bridge performance monitoring can be correlated by the known axle loads of the crossing vehicles. While the traditional B-WIM systems measure strains in the bridge superstructure, some recent methods use accelerometer measurements. The conventional SHM systems typically include various types of sensors, including accelerometers, strain and temperature gauges, inclinometers, etc. They are generally installed on the most important or critical structures only, while B-WIM systems have the potential also for smaller, less important bridges. This paper aims to present the B-WIM capabilities, which are developing towards SHM and are no more limited to just weigh-in-motion the heavy traffic. In the beginning, the paper presents SHM and B-WIM definitions and reviews the PIs that can be extracted from B-WIM measurements. Later, selected studies where B-WIM is used for fatigue lifetime assessment are presented, and finally, two applications of B-WIM for damage detection reviewed. Keywords: B-WIM, SHM, sensors, bridges, viaducts 2. SHM AND B-WIM DEFINITIONS 1. INTRODUCTION 2.1. SHM definitions The main part of the Structural Health Monitoring (SHM) literature was published in the last 30 years. SHM, in its most basic form, appeared parallel with Over these years, it came to the stage when these structural engineering, although this expression was systems are installed on the majority of the not used at that time and began to appear widely in extraordinary bridges, with respect to their the 1980s. The definition of the SHM nowadays, as dimensions, structural system or strategic position. stated by Boller (2009), is: "SHM is the integration SHM systems are nowadays due to of sensing and possibly also actuation devices to misunderstandings of decision-makers about their allow the loading and damaging conditions of a complexity, installation and maintenance costs still structure to be recorded, analysed, localised, and limited to these special structures. One way to predicted in a way that nondestructive testing (NDT) extend applications to the smaller bridges is to use becomes an integral part of the structure and a Bridge Weigh-in-Motion (B-WIM) systems with material". Wenzel (2009), when focusing on SHM some recently developed features, which are of bridges, describes SHM as "implementation of a extending its primary purpose of weighing towards damage identification strategy to the civil the structural health monitoring of the bridging engineering infrastructure. Damage is defined as structures. changes to these systems' material and/or geometric properties, including changes to the boundary conditions and system connectivity. Damage affects the current or future performance of these systems". 85 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia As stated by Catbas (2009), the following challenges weights, processed from strain signals and weights, are interrelated and have to be carefully addressed: obtained from static weighing (Goble et al., 1976), it  was evident that this method could weigh the fundamental knowledge needs,  passing vehicles. Soon after, Moses (1979) laid the technology needs and foundations for B-WIM general principles that are  socio-organisational challenges for routine still valid in today's B-WIM systems. Considering applications. that the bridge during a vehicle passage oscillates Moreover, these challenges are linked with the main about a static displacement position and linear components of a complete SHM system, shown in behaviour of the bridge under traffic loading, then Figure 1. the resulting bending moment is described by the following equation: Similar to Rytter (1993), Catbas (2009) proposed to analyse SHM data at different levels to generate 𝑁 𝑁 information about the health and performance of 𝑀(𝑡𝑗) = ∑ 𝐴𝑖𝐼(𝑥) = ∑ 𝐴𝑖𝐼 (𝑣𝑖(𝑡𝑗 − 𝑡𝑖)) ; 𝑗 = 1 … 𝑁𝑀 structures. Four levels were considered, which can 𝑖=1 𝑖=1 be briefly described as: (1)  Level 1: Information obtained from the raw data without detailed analysis i.e. “Raw data indicators” where:  Level 2: Usage of different analysis methods 𝑀(𝑡𝑗) is bending moment at a time 𝑡𝑗, and different metrics to detect any damage, deterioration or “considerable change” in 𝐴𝑖 are (unknown) axle loads, measurements over the monitoring period i.e. 𝐼(𝑥) are (known) values of bending moment IL; “Identification stage”.  where 𝑥 is location, calculated from measured axle Level 3: Localisation and quantifications of velocity and axle arrival times, “considerable changes”, damage or deterioration. 𝑣𝑖 are axles' velocities,  Level 4: Prediction of future behaviour. 𝑡𝑖 are arrival times of individual axles, 1. Identification of performance and 𝑁 is the number of axles, health metrics 𝑁𝑀 is the number of times in which bending moments are calculated. 2. Sensors and data acquisition networks Assuming a linear relationship between the moment 𝑀 and measured strains 𝜀𝑔 overall 𝑁𝐺 girders or 3. Data fusion, analysis and sections of a slab, we can write the bending moments information management as: 4. Real time data interfaces for 𝑀 = 𝐶𝐹 ∑𝑁𝐺 𝜀 𝑔 𝑔 (2) manual, semi and/or fully automated data evaluation where the average stiffness of the entire structure is represented by calibration factor 𝐶𝐹, obtained 5. Decision making experimentally using the calibration vehicles with their axle loads measured on the static scales. 6. Metrics for success The goal of the B-WIM analysis is to calculate the unknown axle loads 𝐴𝑖 by minimising the difference between the measured and modelled bending Adapted from:(Catbas., 2009) moments. Figure 1: Main components of a complete health monitoring design and challenges 3. B-WIM PERFORMANCE 2.2. B-WIM definitions INDICATORS (PI) Žnidarič (2017) defines B-WIM as a specific Evaluation of several structural performance method that uses an instrumented bridge or culvert indicators is essential to perform efficient B-WIM to weigh in motion the crossing vehicles. The measurements that provide accurate weighing beginnings of the B-WIM date in the 1970s, when results. Over recent years, the calculation of these Goble et al. (1974) investigated measurements and PIs has been elaborated to be suitable for SHM processed bridge strain histories to analyse bridge purposes. These PIs are influence lines, load fatigue life. After a comparison between the 86 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia distribution factors and dynamic amplification cubic splines. These are defined by the following factors. control points: 3.1. Bridge influence line (IL)  'Support' points with abscissas defined by bridge geometry and zero ordinates, The bridge influence line describes the bridge  'M' point, located typically where the highest response, expressed in strain, deflection, inclination strains are expected, etc., at a selected location under a concentrated unit  'Radius' points, which define the radius d/ 2 of load passing the structure (Figure 2). the rounded cap in the vicinity of the 'M' Influence line was introduced by E. Winkler in 1867 point, as "a graph of a response function of a structure as a  Optional points, with abscissas defined and function of the position of a downward unit load ordinates to be optimised, for fine-tuning the moving across the structure" (Kassimali, 2009). For match between the measured and modelled B-WIM analysis, an influence line is the primary response. structural parameter. It is multiplied by the axle loads, which after the summation of all axles' Such an influence line, described in detail in (Žnidarič & Kalin, 2020), has the advantage of using contribution gives the modelled bending moment caused by the vehicle, as shown by Equation (1). random traffic flow instead of a limited set of preweighed vehicles. Furthermore, this paper also demonstrates a promising potential to detect structural damages based on permanent monitoring of their shape over the structural lifetime. 3.2. Girder distribution and Lane factors Girder distribution factor (GDF), traditionally used in the USA, describes how loading is divided to individual beams, girders or sections of a slab. In bridge design codes, such as AASHTO (2012), values of GDFs are specified for different types of beam-slab bridges. When performing B-WIM Source: (Žnidarič & Kalin, 2020) measurements on beam-slab bridges, GDFs for Figure 2: Using influence line to calculate the bending girders, with B-WIM sensors installed, can be moment of a 3-axle vehicle obtained. This performance indicator is calculated The first B-WIM systems used a theoretical from the measured responses of the bridge caused by influence line, which has for the simply supported individual vehicles to obtain statistically defined superstructure and observed strains at the midspan, contributions of traffic loads from various traffic the well-known triangular shape. Nowadays, lanes on the measured girder. For example, if the influence lines for B-WIM applications are always bridge carries two lanes of traffic and has four calculated from the measured bridge response to instrumented girders, then 8 GDFs are calculated. traffic loading. Two EC projects (WAVE (2001) and While assessing the safety of existing bridges, the COST 323 (2002)) initiated this approach. Influence engineers are looking for hidden reserves in line can be obtained either using passages of the structural performance. By considering the GDFs trucks with known axle loads (matrix method) or from the in-situ measurements rather than the design crossing trucks of unknown axle loads that are part codes, they more realistically assess the structural of the B-WIM process (ZAG method). The matrix behaviour, which allows them to optimise the method (OBrien et al., 2005) performs a linear structural models. In a similar way, GDF under optimisation to find the best fit of the influence line certain lanes can be merged into Lane factors ordinate at each location across the bridge, for which (Enright & OBrien, 2013). A more detailed vehicles with known axle loads are needed. The explanation of how GDFs can be obtained from B- ZAG method (Žnidarič & Kalin, 2020) uses random WIM measurements can be found in (Žnidarič, 2017) and (Žnidarič & Kalin, 2020). vehicles from the ongoing traffic flow and performs a nonlinear optimisation to calculate the shape of the 3.3. Dynamic amplification factor (DAF) influence line. One should note that this procedure is the same as when weighing, except that while The dynamic component of bridge traffic loading is weighing, the influence line is known, whereas commonly taken into account with a Dynamic when calculating it, the influence line is a part of the Amplification Factor (DAF) – the ratio between the unknowns to be calculated. dynamic and static load effects on a bridge. In some cases, a resonance effect can drastically amplify the ZAG method, also implemented in the SiWIM® response of a bridge, as can be seen in Figure 3. system (Žnidarič et al., 2011), is constructed from Therefore, this coefficient is one of the most relevant 87 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia performance indicators when reliability analysis of the desired points (in this case, near fatigue-prone existing structures is considered and when some locations) are obtained. hidden reserves from the traffic load perspective In this study, an orthotropic deck bridge over a canal need to be taken into account. was installed with a B-WIM system that collected the traffic loading data for 38 days. Due to the bridge complexity, the B-WIM sensors were installed only on the simply supported span part of the bridge. 24 sensors were installed near the midspans and 4 sensors (for axle detection) near the abutments. Simultaneously, a set of additional strain gauges were installed on the main span and on the pylon to calculate the correlation between traffic loads and the fatigue of the orthotropic deck. The authors concluded that in the ideal circumstances, the same Source: (Kalin et al., 2021) sensors would serve both to weigh-in-motion the Figure 3: Example of a high dynamic response of a traffic and for finite element (FE) model calibration bridge under traffic loading required for the implementation of virtual Replacing conservative DAF values, which are monitoring. defined in some current provisions (AASHTO, Iterative FE model calibration consisted of 2017), with measured data, can significantly calibration of the total response, where model improve bridge performance assessment. Recently a parameters, such as material properties, were varied new method of calculation of this factor based on B- and final calibration to match the measured strain WIM measurements had been developed records at various locations on the bridge. After that, (Kalin et al., 2021). An automated 2-phase method traffic simulation from 38 days of measurements to that calculates DAF values from all heavy vehicle 15 years was performed using Monte Carlo crossing events was developed. In the first phase, the simulation, and fatigue lifetime estimation for optimum filtering frequency for each event is fatigue-prone locations was done. calculated by minimising the differences between the filtered measured and calculated response. Based In addition to confirmation of conservativism in the on the complete population of calculated optimum Eurocode’s fatigue load model, the authors frequencies, the mean cut-off frequency f concluded that the virtual monitoring concept has 0 is calculated. In the second phase, the complete dataset the potential to optimise maintenance costs due to of events is filtered with the frequency f avoidance of unnecessary repairs or even 0, and the DAF of any event 𝑘 is calculated as: replacements of the bridge structures. Similar study on the Orthotropic Steel Deck (OSD) 𝑚𝑎𝑥|𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒 𝐷𝐴𝐹 𝑘| 𝑘 = (3) bridge was performed by Zhang et al. (2019). To 𝑚𝑎𝑥|𝑓𝑖𝑙𝑡𝑒𝑟𝑒𝑑 𝑟𝑒𝑠𝑝𝑜𝑛𝑠𝑒𝑘(𝑓𝑜)| predict the strains at the locations where sensors are This approach removes the need for an expert who not installed, the authors used the bridge's finite selects the parameters based on personal experience. element model, which was updated based on modal The proposed method is based on data from 15 testing. With such a model, it was possible for every different bridges, which invariably demonstrates point on the bridge to determine the influence that a DAF decreases with increasing bridge traffic surface, which, together with known axle positions loads. and loads, served to calculate the fatigue damages without a sensor installed at the specific location. 4. FATIGUE LIFE ASSESSMENT OF For fatigue evaluation, authors used AASHTO BRIDGES USING B-WIM codes and assumed that annual average daily traffic 4.1. Virtual monitoring to predict the would remain constant (as recorded during the remaining life of steel bridges experimental campaign) over its design life, which was 100 years. It was found that the considered As a part of EC 7th Framework Programme project bridge would not fail in fatigue in the most fatigue- Bridgemon, B-WIM and SHM were for the first time prone measurement point, not even after crossing of coupled to evaluate the fatigue life of a cable-stayed a recorded 148 t vehicle. This (overweight) truck bridge (Hajializadeh et al., 2017). The idea of the was almost 2.5 times heavier than the design truck. virtual monitoring approach is to use a B-WIM The authors concluded that the proposed approach system to weigh the axles of vehicles that pass the of virtual monitoring was feasible for fatigue bridge and then use this information as an input to evaluation of orthotropic steel bridges. the calibrated numerical model, in which strains at 88 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia 4.2. Improved fatigue consumption assessment through structural monitoring In a recent study performed within the Assets4rail project, funded by EU Horizon 2020 Shift2Rail Joint Undertaking (Anžlin et al., 2019; Anžlin et al., 2020), authors used B-WIM together with additional sensors for estimating the remaining fatigue lifetime of a railway bridge. While for fatigue lifetime estimation similar concept of virtual monitoring as Source: (Anžlin et al., 2020) by Hajializadeh et al. (2017) was used, both Figure 5: Evolution of the objective functions Je, Jε response (strains) of the bridge during the train and Je+ε during model calibration passage and response (natural frequencies) during shaker test were used to calibrate the numerical This study confirmed that B-WIM system is suitable model. for weighing in motion the railway traffic. Analogous to the B-WIM for the roadway bridges, For B-WIM purposes, two strain gauges were COST323 (2002) specifications were used to mounted near the midspan of the secondary determine accuracy. Accuracy classes were A(5) for longitudinal beam (stringer), which supports the GVW of carriage A in both directions and the sleepers, and additional two strain gauges at the second carriage in one direction, while for the bottom of the rails for axle detection. Along with opposite direction, the accuracy class for carriage B other sensors, B-WIM sensors were used for FE was B+(7). Since only the GVWs from the static model calibration and estimated fatigue lifetime weighting for carriages A and B but not for C and prediction of the selected fatigue-prone details. In locomotive (these were obtained from the technical this study, the FE model calibration procedure spreadsheets) were known, accuracy classes were consisted of manual calibration and calibration using defined only for the GVW of carriages A and B. One the optimisation solver. Furthermore, the objective should note that in this study axial load of traffic function within the optimisation solver has not just could be measured, but unfortunately no static axial considered the natural frequencies from the modal load data were provided. response (during the shaker test), but the response during the soft load test (SLT) (Figure 4) as well. A It was also shown that for model calibration, both detailed description of the SLT can be found in the modal response and the response from the SLT (SAMARIS D30, 2006). could be taken into account. Furthermore, all sensors installed for B-WIM, except axle detection sensors, were used for model updating. 5. ACCELERATION-BASED B-WIM FOR DAMAGE DETECTION Besides conventional B-WIM systems, which measure relative strains at the superstructure to weigh the traffic, new concepts appeared recently, such as acceleration-based B-WIM. To use a bridge acceleration response in a B-WIM system, the Source: (Anžlin et al., 2020) relationship between bridge accelerations and vehicle weights has to be considered. In this way, the Figure 4: Measured response in the space domain for strains at selected sensors during the SLT B-WIM system can become more suitable for monitoring of structures (OBrien et al. 2020) since, Figure 5 shows the evolution of the objective in the case of strain measurements, the response only functions Je (modal response), Jε (response during changes if the sensor is located in the vicinity of the the soft load test) and Je+ε during model updating. damaged point. This also implies that an acceleration-based SHM requires a smaller number of sensors to identify bridge damage. Acceleration based B-WIM uses an interesting concept (Figure 6): the inferred GVWs should be repeatable, but in case of change of global stiffness, different GVWs can be measured. Suppose an adequate deduction of the impact of temperature on the variation of global stiffness can be performed. In that case, one could use a statistical approach and compare the mean 89 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia inferred GVWs to estimate the health of the For traffic population generation, a Monte Carlo monitored bridge. A recent study based on simulation was performed considering normally numerical simulations and field testing (OBrien et distributed parameters of vehicles (mean and al. 2020) has shown that a hybrid SHM and standard deviation), typical for two-axle trucks. For acceleration based B-WIM gives promising results. damage simulation, authors modelled global damage and local damage separately. Global damage was taken into account as a reduction in stiffness of all elements, and local damage was modelled as a reduction in stiffness at only one particular element. The authors proposed EBWIM indicator as a new tool to monitor structural changes. This tool compares GVW, estimated on the damaged bridge and GVW of the same vehicle, estimated on the undamaged bridge (i.e. at the pavement WIM station near the bridge). They concluded that this indicator:  allows distinguishing between global and local damages, Source: (OBrien et al., 2020)  can roughly estimate the location of damage, Figure 6: Schematic diagram of the methodology of  has both for global and local damage bridge damage detection using the acceleration-based situations greater sensitivity to the damage B-WIM system occurrence than the traditional Level 1 damage-identification technique, which is 6. BRIDGE DAMAGE DETECTION based on tracking frequency changes. USING WIM TECHNOLOGY Cantero & González (2015) proposed a novel bridge 7. CONCLUSIONS damage detection method (Level 1) for short to Comprehensive SHM systems were in the past medium span road bridge using both (pavement- mainly present in academia and on some landscape based) WIM system located near the bridge, and a structures. On the other side, simpler SHM B-WIM system, installed on the bridge. Authors applications could be extended to ordinary bridges showed that the ratio of the estimations of vehicles by using recently developed acceleration-based B- weights by both WIM and B-WIM system is a WIM systems, where among other things, the reliable and robust indicator of structure’s condition automatic evaluation of bridge performance and furthermore that this indicator is more sensitive indicators is performed. to damage than the traditional method based on variation in natural frequencies. To obtain PIs, one does not need (besides B-WIM sensors) to install additional sensors. This approach This study aimed to use pavement-based WIM, is different for the “virtual monitoring”, where installed near the bridge, and B-WIM, installed at additional sensors are usually needed for sufficient the considered bridge, for gross vehicle weight calibration of FEM model. For acceleration-based (GVW) estimation (Figure 7). Both systems give B-WIM, typical strain sensors are replaced by GVW estimation, and comparison between them in acceleration sensors, the measurements of which are time could lead to the identification of the used as an input for the conventional (strain-based) occurrence of bridge damage in time. B-WIM algorithm. For bridge damage detection Using the FE model, the authors simulated a vehicle using WIM technology, besides the B-WIM system traversing a bridge. For the bridge, they have installed on the bridge, an additional pavement WIM considered two types of boundary conditions; system should be installed near the bridge to monitor simply supported and fixed-fixed. A four degree of the change of axial load due to damage of the freedom two-axle system was used to simulate the structure. vehicle. In general, most SHM’s are not capable of adequately detecting bridge damages. But given recent experiences, one can state that acceleration- based B-WIM is more effective in the detection of damages than strain-based one. It was therefore concluded at this point to push the future research Source: (Cantero & González, 2015) activities in the direction towards developing hybrid B-WIM that would, besides strain signals, also Figure 7: WIM-based damage-identification concept consider acceleration signals. However, for more for SHM of bridges complex bridges such as cable-stayed bridges with 90 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia orthotropic steel deck, a larger set of sensors than for [8] Enright, B. & OBrien, E. J. (2013). Monte Carlo B-WIM will always be necessary. Particularly if one simulation of extreme traffic loading on short is interested in observing the total behaviour or FE and medium span bridges. Structure and model calibration for purposes such as fatigue Infrastructure Engineering. 9(12), pp. 1267- 1282. evaluation. [9] Goble, G., Moses, F., & Pavia, A. (1974). Field Based on the presented studies, we can conclude that Measurements and Laboratory Testing of B-WIM systems with PI calculation and some Bridge Components - Final report. additional sensors could offer an alternative to the conventional SHM systems. They in general require [10] Goble, G., Moses, F., & Pavia, A. (1976). Application of a Bridge Measurement System. big initial financial bite and can therefore discourage Transportation Research Record No. 579, bridge owners for the implementation of these Washington DC, 36-47. systems. Therefore, the proposed simpler monitoring system could be more extensively used [11] Hajializadeh, D., OBrien, E., & O’Connor, A. on smaller, often older and deteriorated bridges. (2017). Virtual Structural Health Monitoring and Remaining Life Prediction of Steel Bridges. Today, actions on those bridges are mostly based on Canadian Journal of Civil Engineering, 44. condition assessment and not actual performance https://doi.org/10.1139/cjce-2016-0286 under traffic loads. [12] Kalin, J., Žnidarič, A., Anžlin, A., & Kreslin, M. REFERENCES (2021). Measurements of bridge dynamic amplification factor using bridge weigh-in- [1] AASHTO. (2012). AASHTO LRFD Bridge motion data. Structure and Infrastructure Design Specifications, 6th Edition 2012. The Engineering, 1-13. American Association of State Highway and https://doi.org/10.1080/15732479.2021.188729 Transportation Officials, Inc., Washington, D.C. 1 [2] AASHTO. (2017). AASHTO LRFD Bridge [13] Kassimali, A. (2009). Structural Analysis. Design Specifications, 8th Edition 2017. The Cengage Learning. American Association of State Highway and [14] Moses, F. (1979). Weigh-in-motion system Transportation Officials, Inc., Washington, D.C. using instrumented bridges. Journal of [3] Anžlin, A., Hekič, D., Kalin, J., Ralbovsky, M., Transportation Engineering, 105(3). & Lachinger, S. (2019). Section-specific fatigue [15] OBrien, E., Khan, M. A., McCrum, D. P., & load models: method and application, Shift2Rail Žnidarič, A. (2020). Using Statistical Analysis (Grant Agreement Nr. 826250 "Assets4rail") of an Acceleration-Based Bridge Weigh-In- Deliverable D3.2. Motion System for Damage Detection. Applied [4] Anžlin, A., Hekič, D., Kalin, J., Kosič, M., Sciences, 10(2), 663. Ralbovsky, M., & Lachinger, S. (2020). [16] OBrien, E., Quilligan, M. J., & Karoumi, R. Improved fatigue consumption assessment (2005). Calculating an Influence Line from through structural and on-board monitoring, direct measuremetns. Proceedings of the Shift2Rail (Grant Agreement Nr. 826250 Institution of Civil Engineers, Bridge "Assets4rail") Deliverable D3.3. Engineering 159(1), 31-34. [5] Boller, C. (2009). Structural Health [17] Rytter, A. (1993). Vibrational Based Inspection Monitoring—An Introduction and Definitions. of Civil Engineering Structures. Dept. of In Encyclopedia of Structural Health Building Technology and Structural Monitoring. Engineering, Aalborg University. https://doi.org/https://doi.org/10.1002/9780470 061626.shm204 [18] SAMARIS D30. (2006). Guidance for the optimal assessment of highway structures. ZAG. [6] Cantero, D., & González, A. (2015). Bridge damage detection using weigh-in-motion technology [Article]. Journal of Bridge Engineering, 20(5), Article 04014078. [19] WAVE. (2001). Bridge WIM - Report of Work https://doi.org/10.1061/(ASCE)BE.1943- Package 1.2. University College Dublin & ZAG. 5592.0000674 [20] Weigh-in-Motion of Road Vehicles: Final [7] Catbas, F. N. (2009). 1 - Structural health Report of the COST 323 Action. Jacob, B., monitoring: applications and data analysis. In V. OBrien, E.J., Jehaes, S. (Eds.). (2002). M. Karbhari & F. Ansari (Eds.), Structural [21] Wenzel, H. (2009). Bridge Management and Health Monitoring of Civil Infrastructure Health Monitoring. In Health Monitoring of Systems (pp. 1-39). Woodhead Publishing. Bridges (pp. 7-18). https://doi.org/https://doi.org/10.1533/9781845 https://doi.org/https://doi.org/10.1002/9780470 696825.1 740170.ch2 91 Doron Hekič, Andrej Anžlin, Aleš Žnidarič YRS 2021 15. -17. September A REVIEW OF B-WIM AS A SHM APPLICATIONS ON BRIDGES Portoroz, Slovenia [22] Zhang, L., Zhao, H., OBrien, E. J., & Shao, X. [24] Žnidarič, A., & Kalin, J. (2020). Using bridge (2019). Virtual Monitoring of orthotropic steel weigh-in-motion systems to monitor single-span deck using bridge weigh-in-motion algorithm: bridge influence lines [Article]. Journal of Civil Case study. Structural Health Monitoring, 18(2), Structural Health Monitoring, 10(5), 743-756. 610-620. https://doi.org/10.1007/s13349-020-00407-2 https://doi.org/10.1177/1475921718764081 [25] Žnidarič, A., & Lavrič, I. (2011). SiWIM Data [23] Žnidarič, A. (2017). Influence of number and Post-Processing Manual. quality of weigh-in-moiton data on evaluation of load effects on bridges, Doctoral disertation. University of Ljubljana. 92 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Jens Hellekes, M. Sc. Remote Sensing Technology Institute, German Aerospace Center (DLR) Muenchener St. 20, 82234 Wessling, Germany jens.hellekes@dlr.de ABSTRACT The annual average daily traffic (AADT) is a key indicator for describing transport development and an essential parameter for traffic engineering and transportation planning. Commonly, the AADT is collected by permanent counting stations on selected sections of the major road network and supplemented by temporary counts at multi-annual intervals. Because of the high expenditures for construction and maintenance, dense networks of counting stations are rarely established which leads to uncertainties about the spatial and temporal distribution of vehicle volumes. Remote sensing data can be part of the solution as it can be acquired flexibly and over large areas. This paper utilizes aerial image sequences to capture traffic density as well as vehicle speeds to determine the traffic flow on motorway and federal road segments. The instantaneous observations at different days and times are extrapolated to the AADT using time series from reference counting stations in the study area of Braunschweig (Germany). Discrepancies of up to 11 % between estimated and measured AADT are found at three of five metering sections. For the remaining sections, this paper elaborates why limits of the approach are reached here. The presented method is particularly suitable for motorway sections where no or only insufficient data are available. Keywords: Traffic Monitoring, Vehicle Trajectories, Traffic Flow, AADT, Remote Sensing, Aerial Image Sequences 1. INTRODUCTION federal roads. The remaining detectors cover state and municipal road sections that are of particular Road traffic counts and derived quantities such as importance for transport analysis. The continuous traffic volumes or annual vehicle miles travelled are metering is supplemented by temporary used worldwide to monitor transport development measurements at five-year intervals: according to and to accompany measures from design through schemes, vehicles are counted manually or by assessment to their evaluation after implementation. measuring devices on several days at specific hours. Apart from the collection for statistical purposes These data are then extrapolated to the annual (International Transport Forum, 2019), detectors average daily traffic (AADT) to obtain an estimate of provide essential input data for traffic management the vehicle volume until the next counting period systems (Vortisch, 2006) and for quantifying (Schmidt et al., 2020). Outside the coverage area of externalities such as noise emissions (Bajdek, et al., the stationary detectors, uncertainties arise with 2015). The counts are used by road administrations regard to the growth of the traffic volume and the to schedule maintenance cycles and to develop spatial and temporal distribution (Jiang et al., 2006). construction proposals (Bosserhoff et al., 2006). Both This is particularly problematic if a measure for infrastructural and policy measures are usually which significant effects are expected is implemented evaluated with transport models, which in turn are between the surveys: short-term effects are not often validated or calibrated using traffic counts recorded and medium- to long-term effects can only (Cascetta, 2009). With these models, traffic forecasts be partially attributed to the measure due to can be made with regard to various scenarios interference with other factors. (Ortúzar & Willumsen, 2011). For states that lack detailed count data but still face In Germany, about 2,100 permanent counting current traffic problems, exploring new data sources stations record vehicles on the major road network, is promising. Traffic-related applications are with approximately 80 % located on motorways and frequently discussed in the remote sensing 93 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia community because data can be collected flexibly years with the DLR 3K camera system (Kurz, et al., and cover large areas. Numerous works deal with 2012). The flights took place at an altitude of about road traffic monitoring by detecting motorized 700 m above ground, resulting in a resolution of vehicles on aerial and satellite images or videos approximately 9 cm. A repetition rate of 1 Hz leads (Apeltauer et al., 2015; Azimi et al., 2018; Mishra, to an overlap of two consecutive images of about 2012) and deriving metrics describing the current 80 %. In several campaigns, flight strips were flown traffic situation (Azevedoet al., 2014; Larsen et al., twice within a 15 min interval in order to be able to 2013; Yang et al., 2016). perform change analysis over short periods of time. Combined with precise ground control points and Coifman et al. (2006) have investigated the extent to GNSS/inertial data at the time of recording, the which extrapolation to unobserved conditions from images are georeferenced and projected onto a digital optical remote sensing data is possible. For this, they terrain model. Due to the high spatial resolution in used a drone to record vehicle trajectories over a 2-combination with the capture frequency, it is possible hour period and derived the AADT. Jiang et al. to determine velocities of moving objects very (2006) demonstrated that inferring vehicle volumes precisely: in the nadir view, speed differences of up from single aerial images can improve the AADT to 0.36 km/h can thus be detected. update for motorways based on temporary counts. Kaack et al. (2019) created a processing chain that Two flights are selected for this paper, differing in detects trucks on satellite imagery and uses a date (2019-04-25, 2019-10-30) and time (06:45 to regression method to estimate AADT for road freight 07:30 pm, 09:45 to 10:30 am). The images were not transport. taken with a focus on capturing the major road network (motorways, federal and state roads), but for The main contribution of the paper consists of the mapping urban, suburban and industrial areas of exploration of image sequences for AADT about 40 km². However, due to the building structure, estimation, which are acquired on two different days several such roads are covered in the flight area and and times. Imagery of Braunschweig (Germany) is are selected for analysis if they simultaneously fall used to count the number of vehicles on five major within the validity range of a counting station. In road sections and extract their trajectories. The traffic situation at the acquisition time is extrapolated to AADT estimators using hourly time series for the corresponding year. The variations and the gain in robustness are examined when the estimators are combined. In addition, different traffic situations are considered: while mainly motorways with low or Hansestraße medium traffic density have been studied so far, here, on the one hand, the end of a motorway controlled by traffic lights is analyzed. On the other hand, the AADT is estimated for an urban state road with high Ölper traffic density and low speeds due to a construction Berliner Straße site. For a first assessment of the accuracy of the estimated AADT, comparisons are made with the Lehndorf vehicle counts of the corresponding detectors. The paper is structured as follows: First, the dataset of aerial images leading to the selection of measurement sections to be analyzed is presented. Then, the methods for obtaining trajectory data, deriving traffic flows at the time of observation, and Weststadt extrapolating the AADT are described. In the results section, similarities and – in case of deviations between estimated and reference AADT – probable reasons are deducted. Identified opportunities as well as limitations of the Counting station (latest data) approach are discussed and finally an outlook on Road type motorway permanent (2019) possible improvements is given. federal/state road permanent (2015) temporary (2015) 2. DATA ACQUISITION Flight area range of validity boundaries The German Aerospace Center (DLR) utilizes the city of Braunschweig as a test bed for transportation Figure 1: Overview of the study area in Braunschweig research. Within various projects, parts of the city (Germany), primary road network, location of counting were mapped by airplane over the period of several stations and image sequences for analysis 94 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia Figure 2: Motorway end Ölper, controlled by traffic Figure 3: Congestion due to construction works at lights, 2019-10-30 Berliner Straße, 2019-10-30 total, the road sections under analysis are shown in 3. METHODS 648 images. An overview of the study area is displayed in Figure 1. Cross-sectional counts by stationary induction loops usually deliver aggregated data to traffic With the exception of Berliner Straße (this is a state management systems. The macroscopic quantities road), motorway sections are selected. The Ölper traffic flow 𝑄, occupancy and (in case of double-loop measuring section is a special case, as the motorway detectors) the mean speed 𝑉 for a time period are ends here and the transition to the inner-city road provided. Numerous applications rely on this level of network is controlled by traffic lights (see Figure 2). aggregation, so the method developed in this paper Furthermore, construction work is taking place on should also provide data at this level of granularity. Berliner Straße during the second flyover period, resulting in narrowing to one lane and congestion The movement of individual vehicles expressed by (see Figure 3). These features can be considered as trajectory data can be transformed into macroscopic challenging when it comes to AADT extrapolation. quantities by referring on the movement of the vehicle collective. This can be described by the For the AADT estimation, hourly time series per year hydrodynamic relation: of the counting stations are required; in addition, the measured AADT are used for reconciliation with the 𝑄(𝑥,𝑡) 𝜌(𝑥, 𝑡) = (1) estimated values. Hourly loads are only available for 𝑉 (𝑥,𝑡) the Weststadt and Ölper sections for 2019. The traffic density 𝜌 is defined as the spatial average A permanent counting station was operated in the of the number of vehicles on a given road segment 𝑥 Hansestraße section until 2015 – more recent values at a fixed time 𝑡. The traffic flow 𝑄 can be are not available. For the Lehndorf and Berliner determined from the number of vehicles crossing 𝑥 in Straße sections, only AADT values determined on 𝑡. Since 𝜌 is defined as the spatial mean, the mean the basis of temporary counts can be obtained speed 𝑉 also enters as spatial mean. However, most (Federal Highway Research Institute, 2020). The permanent counting stations can only measure the range of validity per counting section shown in temporal average of speeds, which results in Figure 1 is determined in the preparations for the imprecise estimates of macroscopic parameters national road traffic counts. In the case of motorways, (Treiber & Kesting, 2013). the validity range extends from the on-ramp to the next off-ramp in each direction of travel. As it can be The continuity equation (1) applies only for the entire seen in Figure 3, access ways and intersections may unidirectional road section, not to individual lanes, occur on state and federal roads where vehicle given that lane changes are possible. The volumes may change. In general, it is defined that a conservation property of (1) is violated when measurement section on federal and state roads substantial changes in 𝜌 and 𝑉 happen. This is the extends to the next junction of a road of equal or case, for example, when the number of vehicles higher category. Deviations from this are possible, changes as a result of accesses and egresses, or road for example, if a facility with high attraction rate gradients or regulation significantly alternate the causes a significant change in the traffic volume. mean speed 𝑉. This leads to the condition that the road section under analysis must be microscopically small. At the same time, the section cannot be chosen arbitrarily small, since macroscopic quantities can be derived meaningfully only in the case of several vehicles (Treiber & Kesting, 2013). 95 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia 14 100 12 80 ) 10 ) h h m/ 8 60 m/ k( k ( de 6 d e e 40 e sp 4 sp 20 2 0 0 Berliner Straße, D2 Weststadt, D1 (western) (northern) Figure 5: Boxplots of determined vehicle speeds per trajectory element, for selected direction of travel and observation, 2019-10-30 12 %)( 10 wol 8 f cif 6 fart 4 fo er 2 ahS 0 0 2 4 6 8 10 12 14 16 18 20 22 24 Figure 4: Aerial image crop with highlighted Time (h) measurement section for vehicle counting at reference point in time 𝒕∗ (top) and zoomed-in sample for vehicle Ölper D1 Ölper D2 trajectories, derived over image sequence (bottom) Weststadt D1 Weststadt D2 In addition to these requirements, there is an Source: traffic flows published by Federal Highway important hypothesis for the derivation of the traffic Research Institute (2020) flow 𝑄 and subsequently the extrapolation to the Figure 6: Daily time series for selected counting AADT: the traffic density 𝜌 and the mean speed 𝑉 at stations at 2019-04-25, adopted from 2018-04-25 the time of observation are sufficiently good uncertainties must be taken into account in the estimators for the analyzed time slice. For further trajectory analysis. Figure 4 (top) shows an analysis, this hypothesis is assumed to be fulfilled; exemplary aerial image crop at 𝑡∗ with part of the the hypothesis is evaluated in the end. measurement section drawn in. 3.1. Generating trajectory data from aerial This reference image is also used to determine the image sequences number of vehicles for which trajectories are derived. A vehicle is considered as object of observation if the According to estimation theory, randomized front bumper is located in the measurement section. sampling is required to obtain an unbiased estimator. By evaluating the image sequence around 𝑡∗, the Here, randomization refers to the selection of the spatial positions can be determined over time. It is specific route segment and the time of observation. also possible to include vehicles in the analysis that However, the selection of aerial imagery is limited are occluded by bridges at 𝑡∗, for example, but are and, at the same time, it must be assumed that still in the measurement section. Depending on the vehicles are not homogeneously distributed in the flight direction and the road orientation, it is possible section due to the tendency for pile-ups. Therefore, that vehicles at the edge of the section can only be the randomization is approximated by selecting as seen at 𝑡∗. In this case, the velocity of the nearest reference point in time 𝑡∗ the aerial image of a vehicle was transferred to the corresponding vehicle. sequence which shows the road for the first time with This issue occurred in 9 cases, which represents about the greatest coverage. The condition regarding the 2 % of all vehicles analyzed and is considered low for coverage can be justified as follows: at high speeds this study. The resulting trajectories are exemplary and correspondingly large headways on motorways, shown in Figure 4 (bottom). it may otherwise be possible to evaluate not enough vehicle trajectories. However, due to implications of 3.2. Deriving spatial mean speed 𝑽 (1), the section length must be limited to the Another implication of (1) concerns the measurement homogeneous part. From aerial imagery, a change in of the spatial mean speed 𝑉. This is a temporal the number of lanes is clearly apparent, while instantaneous mean. This can only be inferred as an variations in the speed limits are not visible. These approximation from the aerial image sequences, since 96 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia no telemetry data of the vehicle is transmitted. The 4. RESULTS approximation can be judged as good on theoretical A total of 66 image sequences were evaluated, grounds: the length of sequence ∆𝑡 is due to the frame showing an average of 360 m of the measured rate and overlap between subsequent images small. sections. Overall, 452 vehicles were counted, whose This applies in particular to a flight direction trajectories could be tracked over 5 images on orthogonal to the course of the road section. average. In conjunction with the described Empirically, this is also evident: the coefficient of extrapolation method, the following information are variation of the speeds of the individual vehicle is extracted: AADT estimators for the first and second 0.01 at the median, with a maximum of 0.83. For the observation per day and from these derived estimates vehicle with the highest observed coefficient of by averaging the observation values per day and all variation, the standard deviation of the speed is about days, respectively. Table 1 presents an example for 3 km/h and occurs on 2019-10-30 at the section Weststadt. It can be seen that the AADT estimators Berliner Straße due to congestion. Other relatively vary considerably when extrapolating on the basis of high coefficients of variation can also be attributed to a single observation. At the same time, it can be the traffic jam, as well as to the inflow of the traffic shown: by summarizing individual measurements, light controlled Ölper junction (see Figures 2 and 3). both within a day and across all observations, the Figure 5 shows that deriving speed from trajectory estimate stabilizes and, in particular, approximates data offers two advantages: On the one hand, it the presumed true AADT (see Table 2). Also, the allows the traffic situation to be taken into account. directional reference of the vehicle flows is better Due to the congestion on Berliner Straße, vehicles represented by averaging: according to the values of drive at considerably lower speed than permitted the year 2018, no significant difference in AADT per (50 km/h). A naïve transfer of the speed limit would direction has been observed. otherwise significantly overestimate the traffic flow. There are some aspects to consider when comparing On the other hand, it can be seen from Figure 5 that the estimated with the reference AADT: on the one at the time of image acquisition, the speed limit of hand, the time series for extrapolation and the 80 km/h on Weststadt section is exceeded by many absolute values originate from previous years. vehicles at low traffic density. Trajectory data can be Changes are conceivable, for example, due to traffic used to consider this when estimating traffic flow. volume growth, infrastructural changes or different 3.3. Extrapolating measured data to weather conditions. This tends to have a greater effect AADT when there is a large gap between the reference and observed year. On the other hand, parts of the Using the known number of vehicles per reference values are to be questioned with regard to measurement section as well as the trajectories, their accuracy: for the counting station Hansestraße equation (1) can be used to calculate the traffic it is stated that the time series was estimated. For the volume estimator for the analyzed time slice. The Ölper section, it is reported that the AADT was calculation procedure for estimating AADT per calculated with a limited data basis (Federal Highway section and direction follows the approach commonly Research Institute, 2020). Due to the propagation of used in Germany (Schmidt et al., 2020): in a first step, the Ölper time series for Berliner Straße, this also has the daily traffic volume of the counting day is an effect here. extrapolated by means of hour/day factors. They can be derived by the daily time series exemplarily shown The deviations between the estimated and measured in Figure 6. In the second step, the extrapolation to AADT for the Ölper and Weststadt sections are AADT is done by day/year factors. A differentiation 5 and 10 %, respectively, which is considered a good is made according to trip-purpose groups (working agreement. This demonstrates: special cases like the day, Sunday, holiday). Ölper motorway end are not problematic as such. For the section Berliner Straße there are substantial Unfortunately, there is no cross-sectional data by differences (factor 3 between estimated and permanent counting stations for 2019, so that time measured AADT). This could have several reasons: series from 2018 (and 2015 for Hansestraße) must be first, there are no information on the extent to which used. For the section Lehndorf, a propagation takes the Ölper time series reflects usual traffic patterns on place to the effect that the time series for Weststadt Berliner Straße. It can be assumed that the can be observed in a similar way there. An analogous traverse line approach is chosen for Berliner Straße, which presumably shows the greatest similarity to Ölper. If the same date of the 2015/2018 year is not part of the same trip-purpose group, the closest same day of the week was set. This represents a first approximation that can be refined by further investigation. 97 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia Table 1: Estimated AADT on different aggregation levels for counting station Weststadt Date Estimated AADT for observation no., per direction 1 2 average, per day average, all observations D1 D2 D1 D2 D1 D2 D1 D2 2019-04-25 37,572 46,217 31,279 34,693 34,426 40,455 29,396 37,204 2019-10-30 31,426 29,692 17,307 38,214 24,366 33,953 Table 2: Estimated AADT for all observations in 2019 and reference AADT for 2018 (unless otherwise noted) Counting Direction, AADT type station D1 D2 both estimated reference estimated reference estimated reference Berliner Straße 15,982 - 16,472 - 32,455 10,658* Weststadt 29,396 36,644 37,204 37,066 66,000 73,710 Hansestraße 47,649 33,032* 31,299 33,173* 78,948 66,205* Lehndorf 51,615 - - - - 71,000* Ölper 11,398 12,751 13,118 13,171 24,516 25,922 Source: reference AADT from Federal Highway Research Institute (2020). Note: values marked with * reflect the status as of 2015 construction site and the resulting traffic jam on uncertainty regarding the reference data (not 2019-10-30 caused a spatial and possibly temporal available everywhere or only for earlier years, partly change in route choice, so that the transferability of estimated based on short-term counts, not reflecting the time series is lower. Second, even with the 2019- road works), the methodological assumption made 04-25 data (no narrowing), AADT estimates are before can be justified: traffic density 𝜌 and the mean showing to be too high, so the section characteristics speed 𝑉 are sufficiently good estimators for the could also be contributing to the discrepancies. In the vehicle volume in the analyzed time slice. It also urban area, traffic lights are closely spaced, resulting shows, that traffic patterns are reasonably constant in pile-ups of vehicles due to acceleration and and therefore transferable to subsequent years. deceleration phases in quick succession. Thus, this However, refinements of the extrapolation approach could represent a borderline case in the applicability would be valuable for estimating AADT, especially of equation (1). to avoid the propagation of time series of nearby counting stations and instead to apply a series type The estimated AADT for Hansestraße seems to be based on a classification scheme. significantly to high compared to the measured AADT. This is put into perspective when the increase Limitations of the method can be seen in two aspects: in vehicle volume since the year 2015 is considered: first, a single overflight at a time of day with low For Weststadt, a 7 % AADT increase can be observed traffic volume increases the risk of not encountering between 2015 and 2018. Transferring this to the route section with any vehicle. The AADT would Hansestraße as a proxy for the growth rate, the gap then be wrongly estimated to zero. Secondly, the between estimated and measured AADT is reduced demonstrated method can only be applied to to about 11 %. The AADT estimate for Lehndorf is homogeneous sections. Particularly in the inner-city not very meaningful, because – as can be seen in area, these sections are sometimes short due to access Figure 1 – the range of validity falls only slightly into ways and junctions, so that in combination with pile- the flight area. In addition, only one image sequence ups due to traffic lights strongly varying AADT covers the section, which contains zero and three estimators are determined. Further investigations are vehicles per direction, respectively. Therefore, in one needed to explore the limits of applicability in urban direction the AADT cannot be calculated, while in areas. the opposite direction of travel the sample size is very For a large-scale application of the approach, the small. manual detection of vehicles and the extraction of 5. DISCUSSION AND OUTLOOK their trajectories is not practical. Remote sensing methods can be further developed to automate this The AADT estimators obtained on the basis of single process. Depending on the study area, other remote aerial image sequences are subject to large sensing platforms may be more suitable for collecting fluctuations, which can be attributed to the short the data, so the processing chain can be designed for observation time of less than 10 seconds on average generalizability. per section. Nevertheless, the combination of up to four acquisition dates turns out to be promising given 6. CONCLUSION the stabilization of the calculated AADT. Less than 10 seconds on average – this is how long Considering the good agreement between the motorways and federal state roads in Braunschweig estimated and measured AADT for Weststadt and (Germany) are covered by an image sequence dataset Ölper sections, and in light of numerous factors of taken by an aircraft. By counting the number of 98 Jens Hellekes YRS 2021 IMPROVING THE ANNUAL AVERAGE DAILY ROAD TRAFFIC ESTIMATION BY USING 15. -17. September MULTITEMPORAL HIGH RESOLUTION AERIAL IMAGERY Portoroz, Slovenia vehicles and extracting their trajectories at different [6] Cascetta, E. (2009). Transportation Systems times and dates, traffic flows are determined. Analysis: Models and Applications. Springer. Permanent and temporary counting station data in the [7] Coifman, B. A., McCord, M. R., Mishalani, R. study area are used to extrapolate the point-in-time G., Iswalt, M., & Ji, Y. (2006). Roadway traffic observations to robust estimators for the AADT. At monitoring from an unmanned aerial vehicle. the same time, the detector counts serve as a IEEE Proceedings – Intelligent Transport reference to assess the accuracy of the estimate. Systems, 153(1), 11–20. [8] Federal Highway Research Institute (2020). For two measurement sections, there is a deviation of Automatische Zählstellen 2018. 5 and 10 %, respectively, compared to the metered https://www.bast.de/BASt_2017/DE/Verkehrste AADT of the previous year (data of the same year chnik/Fachthemen/v2- were not available). If the growth in traffic volume is verkehrszaehlung/Aktuell/ considered, a similar variation (11 %) could be zaehl_aktuell_node.html?nn=1819516&cms_str achieved for a third metering section, for which Typ=A&cms_map=0 reference data recorded four years earlier were [9] International Transport Forum. (2019). Glossary provided. A federal road section in the urban area for transport statistics (5th ed.). shows significant deviations, which are attributed to [10] Jiang, Z., McCord, M. R., & Goel, P. K. (2006). both a short evaluation segment between two traffic Improved AADT Estimation by Combining light controlled junctions and a reduction in lanes due Information in Image- and Ground-Based Traffic to a construction site. Data. Journal of Transportation Engineering, 123(7), 523–530. According to these findings, the presented method is [11] Kaack, L. H., Chen, G. H., & Morgan, M. G. suitable for provide timely AADT estimators with (2019). Truck Traffic Monitoring with Satellite focus on motorways. The approach may be of Images. Proceedings of the ACM SIGCAS particular interest to states that do not have a well-Conference on Computing and Sustainable developed network of permanent counting stations Societies, 155–164. supplemented by temporary measurements. Third [12] Kurz, F., Türmer, S., Meynberg, O., Rosenbaum, parties that do not have access to counting data may D., Runge, H., Reinartz, P., & Leitloff, J. (2012). also benefit. Low-cost optical Camera Systems for real-time Mapping Applications. Photogrammetrie – REFERENCES Fernerkundung – Geoinformation, 2012(2), 159– 176. [1] Apeltauer, J., Babinec, A., Herman, D., & [13] Larsen, S. Ø., Salberg, A.-B., & Eikvil, L. (2013). Apeltauer, T. (2015). Automatic vehicle Automatic system for operational traffic trajectory extraction for traffic analysis from monitoring using very-high-resolution satellite aerial video data. International Archives of the imagery. International Journal of Remote Photogrammetry, Remote Sensing and Spatial Sensing 34(13), 4850–4870. Information Sciences, XL-3/W2, 9–15. [14] Mishra, R. K. (2012). Automatic Moving [2] Azevedo, C. L., Cardoso, J. L., Ben-Akiva, M., Vehicle's Information Extraction from One-Pass Costeira, J. P., & Marques, M. (2014). Automatic WorldView-2 Satellite Imagery. International Vehicle Trajectory Extraction by Aerial Remote Archives of the Photogrammetry, Remote Sensing. Procedia – Social and Behavioral Sensing and Spatial Information Sciences, Sciences, 111, 849–858. XXXIX-B7, 323–328. [3] Azimi, S., Vig, E., Kurz, F., & Reinartz, P. [15] Ortúzar, J. d., & Willumsen, L. G. (2011). (2018). Segment-and-Count: Vehicle counting in Modelling Transport (4th ed.). John Wiley & aerial imagery using atrous convolutional neural Sons. networks. International Archives of the Photogrammetry, Remote Sensing and Spatial [16] Schmidt, G., Frenken, T., & Mahmoudi, S. Information Sciences, XLII-1, 19–23. (2020). Straßenverkehrszählung 2015: Methodik der manuellen Zählungen. Fachverlag NW. [4] Bajdek, C., Menge, C., Mazur, R. A., Pate, A., & Schroeder, J. (2015). Recommended Best [17] Treiber, M., & Kesting, A. (2013). Traffic Flow Practices for the Use of the FHWA Traffic Noise Dynamics. Data, Models and Simulation. Model (TNM). US Federal Highway Springer. Administration. [18] Vortisch, P. (2006). Modellunterstützte [5] Bosserhoff, D., Follmann, J., & Klotz, S. (2006). Messwertpropagierung zur Verkehrserhebungen: Grundlagen und Verkehrslageschätzung in Stadtstraßennetzen. Grundbegriffe. In V. Bach, D. Bosserhoff, J. Institute for Transport Studies, University of Follmann, S. Klotz, & W. Reuter (Eds.), Karlsruhe. Handbuch für Verkehrssicherheit und [19] Yang, T., Wang, X., Yao, B., Li, J., Zhang, Y., Verkehrstechnik (pp. 91–112). Hessisches He, Z., & Duan, W. (2016). Small Moving Landesamt für Straßen- und Verkehrswesen. Vehicle Detection in a Satellite Video of an Urban Area. Sensors, 16(9), Article 1528. 99 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia ASSESSING THE PERFORMANCE OF A COLD RECYCLED PAVEMENT WITH FOAMED BITUMEN UNDER ACCELERATED PAVEMENT TESTING (APT) WITH MOBILE LOAD SIMULATOR (MLS30) Mehdi Kalantari, Msc.Eng. Design and Structure of Pavements, German Federal Highway Research Institute (BASt) Bruederstr.53, 51427 Bergisch Gladbach, Germany Kalantari@bast.de ABSTRACT In recent years by increasing the demand on more sustainability in transportation sector, different solutions have been defined and applied on production process of pavement’s materials and also in their construction activities. Among different methods, cold recycling has the potential of using higher rates of recycled material and at the same time, lower production temperature. During the last decade, different laboratory researches and monitoring activities were performed on this technology in Germany. Positive international experiences reported on bitumen stabilized materials (BSM) application, especially with foamed bitumen, gained interest in Germany in recent years. In 2018 a research project on topic cold recycling with foamed bitumen was defined in German Federal Highway Research Institute (BASt) in cooperation with the Wirtgen GmbH. The main goal of the project was to gather more information on the behavior of cold recycled material with foamed bitumen by monitoring its response and performance under accelerated loading. The findings of this project will be combined with previous research results for updating and further development of national guidelines and specifications. A 100-meter test section was built at the demonstration, investigation and reference areal of BASt (duraBASt) with two different pavement types: one conventional type (as reference) and one with cold recycled layer. The sections were loaded with the mobile load simulator MLS30 and the performance was monitored with different methods. This paper presents some results and findings from the non-destructive (FWD and rutting measurements) monitoring and laboratory tests on the extracted cores at different time periods. Keywords: cold recycling, bitumen stabilized material (BSM), foamed bitumen, cold recycled pavement performance, accelerated pavement testing (APT), duraBASt, MLS30 1. INTRODUCTION energy consumption and emission production is noticeably lower than the other recycling methods. Increasing demand of more sustainability in As the mixture is produced cold, there is no need to transport infrastructure sector has affected the heat the RAP (reclaimed asphalt pavement) which, construction and maintenance methods of the is the big issue in aduption of asphalt plants for hot pavements. Nowadays, sustainability plays an recycling process. This makes the production important role in developing, evaluating and process relativly simpler and also eases the selecting between different construction and possibility of using higher rates of RAP. maintenance options. Recycling techniques are one of the well-known sustainable methods which To be able to mix the bitumen with aggregates in decrease the negative environmental effects of the ambient temperature, the viscosity of the bitumen pavement’s construction and rehabilitation. It is should be decreased. At the meantime there are two possible to classify asphalt recycling methods in to methods for that. Suspending the bitumen droplets three big categories of hot, warm and cold methods. in water (bitumen emulsion) or making foamed The main difference between these methods is in the bitumen. By injecting a little amount of water in to production temperature of the mix. Cold recycling the hot bitumen, the water droplets will evaporate methods are performed in ambient temperatures and and lead to foaming of the bitumen (Wirtgen, as there is no need to heat the aggregate mix, the Wirtgen Cold Recycling Technology, 2012). The 100 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia produced foam will stay for some seconds and this The APT program of this project can be divided into is enough to be able to mix it with cold and wet four different steps: aggregates. Beside the bitumen, normally cement or 1. Designing activities: test lane layout, mix hydrate lime is added as the second binder to design of the materials, structural design of increase the moisture resistance of the recycled the test sections, justifying of the standard material and its early life strenght. The compacted loading and monitoring plans. mix gains strength and stiffness over time mainly by 2. Construction activites: preparing the test evaporation of water and hydration of the cement lane, material production, sensors (known as curing in the technical language). installation, laying and control tests. Like other countries, cold recycling has been applied 3. Loading and monitoring: preparations, in Germany, but in a limited amount. The main loading and monitorings during and after reason can be the lack of national-level performance the loading. data on the behavior of the recycled material which, 4. Analysis and evaluation of the data which, has affected the existing national guidelines too. in some cases is parallel with loading. During the last decade, different laboratory Each of the activities will be explained in the coming researches and monitoring activities were performed texts. on this technology in Germany (Wacker, Kalantari, & Diekmann, 2020; Kalantari, 2020). The results of 2.1. Test lane layout these activities and the reported positive international experiences, led in to gaining of more For this project a test lane with 100 m length and 3.5 interest in Germany on this type of cold recycling. m width was selected form the D/U part of the In 2018 a research project on the topic of cold duraBASt. It was decided to construct a recycling with foamed bitumen was defined in conventional pavement section type beside the cold German Federal Highway Research Institute (BASt) recycled pavement, as a reference. Two third of the in cooperation with the Wirtgen GmbH aiming to test lane was planned for the cold recycled pavement gather more data on the behavior of this material. A and the rest one third for the conventional pavement. test lane was planned to be constructed and tested in The section with the foamed bitumen cold recycled the BASt’s demonstration, investigation and base layer will be referred as cold recycled (CR) reference areal named as duraBASt. section and the section with the hot mix asphalt base duraBASt is the outdoor test facility of BASt which will be referred as the reference section (RF) in the is located east of Colonge parallel to A3 motorway rest of this paper. For the CR section, two loading north direction intersection with A4 motorway with areas (one as reserve) and for the RF section, one a total length of approx. 1,000 m and area of 25,000 loading area were planned. Each loading area has a m2. The construction of the test facility took place reference point far enough out of that, which is between 2015 and 2017. There are three main areas instrumented with embedded sensors and measured in duraBASt as the reference area (R areal) which is with nondestructive methods each time when a for the approval of the surface characteristics measurement is performed on the loading areas. The measuring vehicles and then the demonstration and aim of these reference points is to be able to check investigation areas (D/U areal). They consist of nine the difference between the loaded and not loaded lanes, six of them are located in the central area (with pavement and also to control if the devices are different widths from 3.5 to 5.5 m and lengthes up measuring correct. to 100 m). For more information about the reader is referred to duraBASt website and (Wacker & Jansen, 2020) This paper aims to present the APT 1 program phases of this project and some of the results from that. 2. APT PROGRAM AND ITS PHASES Source: BASt, Project data Accelerated pavement testing (APT) is an approach Figure 1: Sketch of the test lane, its pavement sections for simulating the truck loads in a short and and different loading areas compressed time period (Steyn, 2012). BASt’s APT Figure 1 shows the sketch of the test lane with program has been defined and standardized based on different sections, its loading areas and reference the gained experiences over the years of performing points. The loading areas in the cold recycled section that in house (Wacker & Jansen, 2018). are named as FC1 and FC2 with Ref.1 and Ref.2 as the reference points for each. The loading area of the 1 Accelerated Pavement Testing 101 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia reference section is named as FC3 with its reference times and half life of 14 seconds). To increase the point named as Ref.3. moisture resistance of the mix, cement (I-425 N) was seclected as the secondary binding agent. Different types of sensors including pressure cells, strain gauges, temperature and moisture sensors The reclaimed asphalt pavement (RAP) was selected were installed at different depths of the pavement in and stockpiled in an asphalt plant. During the loading areas and reference points of different preprocessing the RAP was sieved in to two sections. Figure 2, shows the type and position of fractions of 0-8 mm and 8-22 mm sizes. The activity different sensors in the loading areas. tests (Wirtgen, 2017) showed that the bitumen in the RAP is still active. To nutralise its effect on the resulted mix, addition of crushed aggregates or crushed dust is recommended (Wirtgen, 2012). The tests with the addition of 15% of crushed sand (0-5 mm) or crushed granular (0-20 mm) led to satisfactory results in case of neutralizing the bitumen activity of the RAP. The grading curve results also showed lack of fines which is necessary for foamed bitumen mixes. Different mix combinations of the two RAP fractions and 0-2 mm sand were produced with different amounts of foamed bitumen and 1% cement. Source: BASt, Project data Figure 2: Schematic picture of loading areas and different sensors installed at different depth of tham. A: for FC1 and FC2 and B: for the FC3 The letters show the sensor’s type and the numbers shows its depth in cm. T stands for temperature Figure 3: The grading curve of parent material mix of sensors, S stands for strain gauges, C stands for the selected mix design. Upper and lower limits are pressure cells and M stands for moisture sensors. from Wirtgen (Wirtgen, 2012) 2.2. Materials and thicknesses of the Based on the indirect tensile strength tests results pavements (ITS wet and dry), the final mix combination of 75% RAP (40% 8-22 mm and 30% 0-8 mm) and 25% The reference section was selected from the existing sand (0-2 mm) was selected as the parent material national design code for pavements in Germany (Figure 3) with 2.2% foamed bitumen, 1% cement RStO 12 (FGSV, RStO 12, 2012) for 1 million and 4.5% water. The amounts are based on dry ESAL (of 10 ton standard axle) bearing capacity. It weight of the parent material mix. consists of 4 cm of wearing coarse and 14 cm of hot mix asphalt base coarse. Stone mastic asphalt with Based on the existing national guidlines for cold nominal aggregate size of 8 mm was selected for the recycled materials (FGSV, 2005; FGSV, 2007), for wearing coarse (SMA 8 S) and asphalt concrete with the same bearing capacity as the reference section, nominal aggregate size of 22 mm was selected for the design should consist of 4 cm of wearing coarse, the base coarse (AC 22 T N 50/70 BAS RA) which 6 cm of hot mix asphaltic base and 18 cm of cold contains around 30% of RAP. recycled layer. According to international literature and structural designs from the author (Kalantari, For the mix design of the foamed bitumen mixture, 2020), it was figured out that the recommended the proposed method for foamed bitumen stabilized design is kind of conservative for the bearing mixes (BSMs) from Wirtgen was applied (Wirtgen, capacity of 1 million ESAL (of 10 ton standard axle) 2017). Nyfoam80® from Nynas company was used and therefore, it was decided to adjust it to 4 cm of to produce the foamed bitumen. Foaming tests were wearing course and 20 cm of cold recycled layer performed at different bitumen temperatures and with foamed bitumen (CRF) as the base layer. foaming water contents with the WLB 10 S foaming equipment from Wirtgen. The bitumen temperature of 160° C with 2% foaming water satisfied the desired foaming parameters (expantion ratio of 13 102 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia 2.3. Production of the recycled mix Recycled mix was produced in the asphalt plant where the RAP was already stockpiled for this project at the time of mix design tests. To produce the mix, a Wirtgen KMA220 mobile plant was utilized. On day before the production, the 3 fractions of parent material were dry mixed together Source: BASt, Project data by the KMA220 based on the mix design formula. At the production day, the prepared mix was fed in Figure 5: Strain gauges installation, laying and to KMA220 and the recycled mix was produced with compaction of the cold recycled layer 2.2% foamed bitumen and 1% cement (Figure 4). To control the construction quality, fresh samples of The amount of additional water was adjusted based recycled mix were taken from both sides of the test on the measured in situ moisture of the mixed section during the laying process at different fractions. The produced mix was loaded directly on positions (30, 40, 50 and 60 m) and were compacted trucks and transported to the construction site. in the laboratory at the same day to produce specimens for ITS tests. After curing (72 hours in 40° C) they were tested in dry (cured state) and wet (24 hours in 25° C) states at 25° C. Beside the ITS controls, the Medium Falling Weight Deflectometer (MFWD) was applied on middle points of the 2 loading areas (FC1 and FC2) and their reference points (Ref.1 and Ref.2) from the day of construction and up to 2 days after that to monitor the evolution of the stiffness at early life of the material. More information on the device and its specifications can be found in (Jansen, 2013). Figure 6 shows the increase of the Evdyn over time at different measured points. It can be seen that the Source: BASt, Project data stiffness development on FC1 point (at 42.40 m) is Figure 4: Production of the cold recycled mix with different from the others. mobile plant (KMA 220) 2.4. Construction of the test sections Construction was started with preparing the existing lane by milling the asphalt and excavation up to -40 cm, compacting, installing the pressure cells, then laying the granular anti frost layer back and compacting to reach the desired depth level for each of the sections. Plate load tests were perfomed at reference points and also at some points in the loading area for quality control. Figure 6: The results of the MFWD measurements The construction started from the cold recycled Table 1 shows the ITS and the TSR (Tensile section (11.09.2019) by laying the recycled layer Strength Ratio) for the positions at which the with asphalt paver and compact it with steel drum samples were taken (each one is the average of 3 and rubber wheel rollers. Before laying the layer, the replicates). strain gauges were installed at planned positions. Moisture sensors were installed in the fresh laid Table 1: The results of the ITS tests on fresh samples layer before the compaction. compacted and tested in laboratory Positions of the samples [m] Parameters 30 40 50 60 ITSdry [kPa] 339 224 252 341 ITSwet [kPa] 303 133 207 276 TSR [%] 89 59 82 81 It can be seen that the results of the 40 m point are weaker than the other points which confirms the results from the MFWD measurements. A relatively lower TSR and lower rate of early stiffness gain, can be the evidence of different amount of cement in this 103 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia section. It can be also the result of a sudden change in aggregate gradation (kind of segregation). These observations opened the topic of homogeneity in production and construction. To investigate the extend of this issue and the effect of curing on that, extra FWD measurements were planned which are explained later under the monitoring plan. Source: BASt, Project data The in-situ density was determined in 10 m intervals Figure 8: Mobile load simulator of BASt (MLS30) with the balloon method. Table 2 shows the It has four test bogies with one wheel on each. They positions and the amounts. The dry density in mix rotate in aclosed frame on guid rails in a vertical-design stage was 2.2 gr/cm3. oval direction. Every time one of them rolles on a Table 2: The results of in-situ density tests (in gr/cm3) linear strip over the loading area. It is possible to Positions apply variable wheel loads between 45 to 75 kN and 20 30 40 50 61 [m] also different types of wheels (single, super single or Density 2.26 2.24 2.25 2.28 2.21 dual). Variable loading speeds of 6.5 up to 22 km/hr (6.1 m/s) are achievable. At its maximum speed, it is possible to get around 6000 passes per hour which, means every 600 ms one pass on the same position (Wacker, Scherkenbach, & Jansen, 2018). It has the ability of 50 cm lateral load wander on each side. For loading of each area, the MLS30 is positioned on the desired place and will rest on its 4 side colums (one each side). MLS30 setup for this project was to use the super single tyre with 50 kN load and the speed of 6000 passes per hour. Based on the gained experiences from the last projects, it was planned to load each section to 3 million passes. To minimize the effect of weather changes, it was decided to load both sections (reference and cold recycling loading) Source: BASt, Project data together by changing the loading in weekly periods Figure 7: Controlling the in-situ density of the between them. From the cold recycling section, the recycled layer FC2 loading area was selected for this purpose. The loading started on Feb. 2020 and till the end of Nov. The asphaltic base layer for the reference section 2020, each loading area was loaded to 2.2 million was laid a week later (17.09.2019). A polymer passes. The loading started again in Feb. 2021 and modified fluxed bitumen was sprayed on the surface will continue till reaching to 3 million passes for of the cold recycled layer at the same day (Viaflex® each. It is also planned to load the FC1 area (the 2nd C60 BP-4-S). The whole surface of the test lane was one from the cold recycling section) after that up to then overlaid with the wearing course the day after. 1 million passes too. During the laying of the asphaltic layers, fresh material was samples for the laboratory mix 2.6. Investigation plan of the project controls. Investigation plan of the project is consisted of two 2.5. Loading of the test sections main parts: field monitoring and tests plus laboratory tests on the extracted cores. Field BASt uses a Mobile Load Simulator (MLS30) since monitoring plan consisted of bearing capacity 2013 for its indoor (test halls) and outdoor measurements with Falling Weight Deflectometer (duraBASt) APT programs. MLS30 is a device (FWD), subsurface scan with a 3D Ground which has been designed to simulate the wheel Penetration Radar (GPR), transverse profile passes of the heavy trucks in a compressed time measurements along the loading area and duration. monitoring with the installed sensors. FWD measurements The standard bearing capacity measurement plan aims to assess the bearing capacity of the pavement and to monitor if it changes with the evolution of 104 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia loading. It consists of 3 main concepts. For each Transverse profile measurements (evenness) measuring point, 3 drops of 50 kN are applied. All A profilometer is used for this measurement which the 3 reference points (named as Ref. 1 to Ref. 3) are has a small wheel rolls over the surface. The measured in all of the FWD measurements. In measurement is performed along the loading area at concept 1 (named as MK1), measurements are 5 different cross lines (Figure 10). performed on three points along the center line of the loading strip in each loading area. Concept 2 (named as MK2) contains the same points from the first concept plus 2 more points each in left and right sides of the centerline points. These points are on the line parallel to the center line and with 25 cm apart to each other. In measuring concept 3 (named as grid measurement), a grid of points are measured around the loading area. Grid lines will have a distance of 50 cm from each other and cover an area of 5 m from Source: BASt, Project data the center of the loading strip to each side (Figure 9). Each concept has different measuring intervals Figure 10: Measuring the transverce profile on a based on the load cycles. cross lane in FC2 (cold recycled section) loading area Three measurements are done at each cross line before the start of the loading to define the average zero measurement profile. Measurements are performed at different loading intervals starting from 50,000 intervals and later to 200,000 loading intervals after the first 200,000 cycles. The measured data are used to calculate the rutting depth and its change over the loading cycles. Source: BASt, Project data The rutting depth is determined based on the Figure 9: FWD grid measuring concept points of the difference between the minimum (Min.) and FC2 loading area maximum (Max.) amounts calculated for each Beside the standard FWD plan, two extra concepts profile measurments. The Min. amount is were defined for this project one after the determined by averaging the loading strip results construction and the other one after the progress of over a cord length and the Max. amount is the loading. The first one is to measure the whole test average of two side points. Figure 11 shows the lane on its center line at 50 cm intervals (named as method concept. The cord length is 20 cm. whole lane concept). The aim was to assess the construction homogeneity considering the ITS and MFWD results after construction. After the first measurement round, it was decided to continue this concept at different time intervals to monitor the process of stiffness gaining of the recycled material over the time through the whole test lane. The second additional concept aimed to look closer to the effect of temperature changes on FWD measurements over the cold recycled section. As the cold recycled material has lower bitumen content Figure 11: Rut depth calculation method plus cement as the secondary binding agent, Ground Penetration Radar measurements therefore its response to the load is less temperature dependent than the asphalt section. In this concept This type of measurement is to get a better view of (named as FWD-Temp.), different points were the thickness of different constructed layers and to selected through the centerline of the CR section and be able to evaluate the construction homogeneity. in different days FWD measurements were Knowing the as built thickness of the layers is a performed on them through the day at different crucial parameter for interpreting the observed temperatures (of the pavement) to be able to assess performance of the pavement sections. It is also its effect. needed for computational models of the pavement sections to calculate the remaining life based on the All the 5 above mentioned measuring concepts were material as built parameters determined from the applied for this project. tests on the extracted cores. 105 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia Sensors monitoring then used for ITS tests at 25° C as the stiffness tests in the mentioned strain range doesn’t affect the ITS As described before in the layout section, different amount (Kalantari, 2020). These cores are planned types of sensors were planned for this project and to be extracted at different time intervals after the were installed at different depth of the tests sections construction and it was tried to perform a round of during the construction process. Strain gauges and the whole lane FWD measurement simuletaneously. pressure cells capture data every 15 minutes for a duration of 30 seconds in each loading period but the The second cores are planned for about a year (or a temperature and moisture sensors gather little more) after the construction (considering that continuously every 15 minutes. the curing is mostly completed). The aim is to obtain the stiffness master curve and the fatigue behavior Laboratory research plan and tests of the recycled mix. The tests are the same as the A comprehensive laboratory investigation was ones for the asphaltic base coarse samples. The only planned parallel to the field monitoring. It has two different is that to be sure the specimens are not main goals: to monitor the evolution of damaged, the stiffness tests are performed at lower stiffness/strength in the cold recycled material over horizontal strain levels than the normal amounts for the time (curing process) and to assess material’s HMA (around 0.03 to 0.04‰). In his PhD, the main characteristics and performance parameters in author has developed a method to produce the both test sections. A coring plan has been prepared general stiffness model of the material by for this purpose to extract cores at different intervals combininig the results of the multi-step stiffness after the construction and to perform stiffness, tests and the stiffness master curve tests. This indirect tensile strength and fatigue tests on them. general model captures both the strain dependency Coring is performed to obtain 150 mm diameter of the stiffness (which comes from its granular part samples. behavior) and also its temperature dependency (which comes from the asphaltic part behavior). The cores taken from the reference section, will be cut to separate the wearing and the base coarses. The Considering the loading areas and the requirements specimens will be used for cyclic indirect tensile of from FWD measurement concepts and the stiffness tests at different temperatures and construction homogenity of the test sections, the frequencies according to German method (named position of the cores were planned. Extra cores will TP Asphalt-StB Teil 26-2018) which is very similar be extracted from the loading areas (in both test to EN12697-26: 2012-06 (annex F). The results will sections) after the end of APT loading with MLS30 be used to obtain the stiffness master curve of these to investigate the possible changes in the material two types. The asphaltic base coarse specimens will parameters. later be used for indirect tensile fatigue tests at 20° C and 10 Hz (also based on the German method 3. RESULTS AND DISCUSSIONS (named TP Asphalt-StB Teil 24-2018) which is As the program is still undergoing all the results are similar to EN 12697-24: 2012-08 (annex E). The not analyzed and assessed for reporting at this results are used to determine their fatigue properties. moment therefore only some of the results from The cores taken from the cold recycled section monitoring and the laboratory plans are presented aiming to monitor the curing, are cut to get here in this section. The MK concepts of the FWD specimens with 60 mm height. The loading and the measurements and the results of the second core sensor places around the speciemens are evened series are not reported here. with gypth and then they are conditioned in the oven 3.1. Transverse profile measurements at 40° C for 72 hours to accelerate the loss of their moisture because of coring and cutting. After that results they are tested under cyclic indirect tensile test (at The amount of rutting was determined for each 20° C and 10 Hz) at different stress levels (up to loading area at different loading cycles based on the 0.1‰ horizontal strain level) to obtain the stiffness results of transverse profile measurments. Figure 12 modulus at different horizontal strain levels. This and Figure 13 show the total rutting amounts at five test is named as multi-step stiffness test; the method different measurement lanes along the loading areas and the way to analyze the results developed by the of FC2 (CR section) and FC3 (RF section). author during his PhD (Kalantari, 2020). It captures the stress/strain dependency of the stiffness which is one of the characteristics of the foamed bitumen recycled/stabilized mixes beside their temperature dependency of the stiffness. Minimum of four specimens are used for this test. Based on the results it is possible to determine the stiffness at any desired strain level and probability level. The samples are 106 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia Table 3: Average weighted temperature (AWT) on the surface of FC2 loading area during the first 1 million loading cycles Cold recycled section (FC2) loading range % of the total AWT [x1000 rolling] loading [°C] All 74.8 21.7 0-200 20.0 20.1 200-253 5.3 27.5 355-600 24.5 17.8 750-800 5.0 24.8 Figure 12: The rutting amount versus loading cycles 800-1000 20.0 25.9 for CR section Table 4: Average weighted temperature (AWT) on the surface of FC3 loading area during the first 1 million loading cycles Reference section (FC3) loading range % of the total AWT [x1000 rolling] loading [°C] All 75.0 23.5 0-200 20.0 16.1 200-400 20.0 22.7 650-800 15.0 25.0 800-1000 20.0 30.5 Figure 13: The rutting amount versus loading cycles for RF section Looking to the tables shows clearly that each jump Comparing the results of the two construction types in the rutting amounts of the reference section (FC3) together shows a slightly higher amount of total is related to a temperature increase. It was decided rutting in FC3 section with HMA base layer. Both to plan the future loadings on the section by sections have no cracks sofar. The maximum considering the temperature forecast of the coming amount of rutting in the cold recycled section (FC3) week. The data of the temperature sensors will be is 4.4 mm after 2.2 million cycles of loading. analysed for the whole loading cycles. Considering the allowable design limit of 10 mm 3.2. FWD measurements results (5% of the layer thickness), the section behavior is assessed as very good. It can be concluded that the Here the results of three different concepts of the performance of the recycled section against the FWD measurements are presented. rutting, is equal to the reference section. Whole lane measurements results For a better comparison between the two The first measurement proved the findings from the construction types and also to be able to define a quality control tests. Figure 14 shows the SCI300 rutting model, it is necessary to consider its parameter (it is the difference between the surface variations between the two sections during the deformations measured at the center of the loading loading periods, As the temperature is an important plate and 300 mm apart from that) in the centerline factor which affects the rutting rate of the of the whole test lane at different measurement bituminous layers. To get a closer look to this issue intervals (50 and 490 days after construction). the data from the temperature sensors have been used. Table 3 and Table 4 show the surface temperatures up to 1 million cycles of loading for each of the loading areas. The sensors were successful to record the surface temperature for about 75% of the 1 million loading cycles. The temperature in each cumulative loading range (loading ranges in the first colum of the tables) was calculated based on a weighted averaging on the temperatures during each of the loading ranges. Figure 14: The results of the whole lane FWD measurements (SCI300 parameter) Looking to the cold recycled section, by comparing the results with the distance covered with each of the mixture transport tracks, shows which truck had the suspicious mix. It is also possible to see that with the changes of the hauling truck, a change has happened 107 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia in the pavement too which can be because of that the precipitation) were combined with the results operator closed the paver hoper and the mix (Figure 16). gradation was changed because of the segregation. The findings opened the question of the effect of laying method on the homogeneity of the pavements with CR layer. Looking to the reference section results, it can be seen that the construction is more homogenous. The results also were used to plan the position of the cores. The measurements were continued at different intervals to monitor the changes of the material over the time. Figure 15 shows the SCI300@ 20° C parameter from different intervals of whole lane FWD measurements on a section of the cold recycled section, where the cores are extracted to check the stiffness evolution over the time. The cores were extracted between 27 to 29 m position. It can be seen that the SCI300@ 20° C parameter Figure 16: Changes of the SCI300@ 20° C over the decreases over the time which is a result of the time at different points in combination with weather increase of the stiffness in cold recycled layer. parameters The results showed that the response of the material is not only temperature dependent but also depending to the wet-dry cycles (or seasonal variations) too. It is then crucial to consider beside the temperature corrections of the FWD measurements, this effect in to account too. Grid measurement cocepet results As mentioned before this concept is used to get a more detailed look on the changes in the sections because of the loading cycles. Till now two sets of measurements were performed on each of the loading areas. First round before the start of the loading and the second round after 1.5 million cycles Figure 15: The evolution of the bearing capacity of loading on each. Figure 17, shows the difference (SCI300@ 20° C) over the time in cold recycled of d0 (temperature corrected) between the two material (the numbers in legends, show the days after measurements for the FC3 loading area and around construction) that. FWD-Temp. concept results The aim of this concept was to assess the effect of temperature changes on the FWD results. 6 different points (containing the weakest, strongest, 2 reference points and the middle of each loading area) were selected and were measured at different days. One of the analysis was to look at the changes of the SCI300 at different temperatures during each daily measurement. After the calculation of SCI300 for each measuring point and each temperature, a linear equation was fitted to the results of each point and used to determine the SCI300 at 20° C for that point Figure 17: The difference of the parameter d0 (temp. at that measurement day (which can be related to corrected) [μm], between 0 and after 1.5 million days after construction). cycles for FC3 Putting all the calculated SCI300@ 20° C together in a It is clear that the loading strip has got some damage time order after construction, showed that even by as the difference is with negative values. The side having the same temperature for them, their values amounts with positive values show the decrease of change over the life of the pavement. To investigate the d0 because of the loading. This is because of the that, the weather data (temperature and side effect of loading on consolidation of the 108 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia granular anti-frost layer and also the subgrade which, leads to deacrease in the whole deformation mould response. The upper and lower left sides show the damaging effect of the heavy corner colums of the MLS30 but as can be seen they didn’t affected the loading strip. The same calculation was done for the results on FC2 (cold recycling) area. Figure 18, shows the results. Source: BASt, Project data Figure 19: An example of the sawed cores and the prepared specimens from them for the tests Figure 20 shows the evolution of the stiffness over the time which was calculated from the results on upper part specimens. They are at 20° C and 10 Hz and determined for 0.05‰ horizontal strain level with 95% level of confidancy. Looking to the trend, clearly shows the curing process of the material over the time which continues even after one year from construction. The trend is also in agreement with the FWD results (Fgure 15). Figure 18: The difference of the parameter d0 (temp. corrected) [μm], between 0 and after 1.5 million cycles for FC2 The same as the other section, the loading strip shows damage and the around points show the decrease in the d0 which in this case is a combination of not only the consolidation of the granular and subgrade but also the cold recycled layer too. As the material is semibound it has partly responses like granular material. The second point is the increase of its stiffness because of curing over the time which is more obvious on not loaded areas. Looking to the Figure 20: The stiffness of the cold recycled material sides it is possible to see the effect of the corner side from the cores at different days after construction columns of the MLS30. As the section was not homogenous before the loading, the resulted damage 4. SUMMARY AND CONCLUSIONS from them is also different in each side but they didn’t affect the loading strip. Cold recycling is a method which got a high amount of interest during the recent years in Germany. It has 3.3. Laboratory tests results the benefits of lower energy consumption and up to 100% use of the high amount of stockpiled RAP in In this section only the results of the stiffness tests asphalt plants. Considering the positive international on the cold recycled specimens are presented as the reports on the behavior of cold recycled mixes with other results are not completely analyzed at the time low amounts of foamed bitumen and cement of writing this paper. As explained in the laboratory (classified as BSM) and relatively low amounts of plan, the cores were taken at different intervals after experience and performance data on these type of the construction from a section which had a material, encouraged BASt to define a research homogen layer of cold recycled material. Figure project in cooperation with Wirtgen GmbH. The 19Figure shows an example of the cores and the project applies the APT concept to evaluate the prepared specimens from them for the tests. performance of a pavement type containing foamed Normally from each core, two specimens from upper bitumen recycled layer. The project started end of and middle part of the recycled layer are prepared 2018 and continued with the preparation tests and for the tests. activities till the construction of the test lane in Sep. 2019. The loading with MLS30 started in Feb. 2020 and is still ongoing. This paper aimed to present the APT program of this project, its different phases from planning to execution, monitoring, analysis methods and a part of the results. 109 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia Different points and lessons learned from this considered in correction method for FWD research in fields of preparation and production of measurements. the mix, construction and quality control, field 6- Monitoring the development of rutting with the monitoring and nondestructive tests and laboratory load cycles, showed a very good performance of the tests to assess the material behavior. Some of them cold recycled pavement section with a considerably can be mentioned as below items: lower than expected rut amounts. The results may 1- Parent material characteristics have a great impact suggest the need of revising the existing national on behavior and performance of the resulted design guide for the cold recycled pavements. recycled mixture. It is recommended to crush the Comparing the results from the reference section, RAP to get a better size distribution and internal shows the importance of temperature on the rate of angle of friction in the compacted aggregate permanent deformation in pavements with thick skeleton. This also increase the possible rate of RAP HMA packages. usage in the mix as the need for virgin aggregate to 7- The results of stiffness tests on the specimens adjust the gradation curve will decrease. It is also from the extracted cores at different lifes of the important to consider the level of the activity of the recycled layer showed the increase of the stiffness old bitumen in the RAP especially when different over time even after one year of construction. They sources of RAP are mixed together in a stockpile. agree with the different FWD results together. 2- In case of the specimen compaction in laboratory, The program will continue with the loading of the using the vibratory hammer showed good test areas of each sections till 3 million cycles with experiences in case of ease of compaction and the the monitoring activities. The stiffness and fatigue production rate of the specimens. It is still needed to tests results on the specimens from the extracted gain more experience with the method. Using the cores will be applied in to a mechanistic-empirical optimum moisture contents (OMC) amounts from software to assess the life of the section with their modified proctor compaction tests as a base for this field behavior. The next phase of the research will compaction method seems not be appropriate. It is deal more in detail with the failure mechanism of recommended to use the vibratory hammer this group of material and to define failure transfer compaction also for the determination of the OMC function for structural design purposes. too. Acknowledgments 3- In case of using the mobile plant machine to produce the cold recycled mix, the experience with The author emphasizes his great appreciation to the this project showed that the method needs a very big project team which is from different sections of short setup time (half a day) and a small compact BASt institute. Their high flexibility in adoption of working place. This advantage can be considered in different plans, their special care on variable tasks of topics like resiliency in pavement construction. It this complicated project and their trust was the key has the possibility of mixing different to the success. fractions/types of the parent materials together and of more control on the production process. Beside REFERENCES the possibility of higher production quality in case of utilizing mobile plants, it is important to notice [1] FGSV. (2005). Merkblat für Kaltrecycling in that the construction and laying of the mix should be situ im Straßenoberbau, MKRC. Cologne, Germany: FGSV. performed with care to avoid segregation and reach a homogenous layer. [2] FGSV. (2007). Merkblatt für die Verwendung von pechhaltigen Straßenausbaustoffen und von 4- The experience in this project showed that Asphaltgranulat in bitumengebundenen Medium Falling Weight Deflectometer (MFWD) Tragschichten durch kaltaufbereitung in tests are fast and reliable method to check the daily Mischanlagen, M VB-K. Cologne: FGSV. work, to evaluate the construction quality and the [3] FGSV. (2012). Richtlinien für die evolution of material strength at its very early stages Standardisierung des Oberbaus von of the life before having the construction tests Verkehrsflächen, RStO 12. 2012. Cologne, results. Still more data should be collected to be able Germany: Forschungsgesellschaft für Straßen- to define acceptance thresholds for the measurement und Verkehrswesen (FGSV). results. [4] Jansen, D. (2013). Dynamic load plate tests: 5- The results of FWD-Temp. concept calibration procedure for test equipments in measurements proved the lower temperature Germany. Ninth International Conference on dependency of the cold recycled material compared the Bearing Capacity of Roads, Railways and to HMA. Beside that they showed that the material’s Airfields, (pp. 229-238). Trondheim, Norway. field response is affected by the seasonal weather changes (wet-dry cycles) which, should be 110 Mehdi Kalantari YRS 2021 ASSESSING THE PERFORMANCE OF A COLDRECYCLED PAVEMENT WITH FOAMED 15. -17. September BITUMEN UNDER ACCELERATED PAVEMENT TESTING WITH MLS30 Portoroz, Slovenia [5] Kalantari, M. (2020). Assessing the effect of [9] Wacker, B., Kalantari, M., & Diekmann, M. binding agents on resulted mechanical and (2020). Cold Recycling in Germany—Current fatigue characteristics of foamed bitumen and Experiences and Future Projects. Raab C. (eds) cement stabilized mixes,Draft manuscript for Proceedings of the 9th International Conference PhD, unpublished. Siegen, Germany: on Maintenance and Rehabilitation of University of Siegen. Pavements—Mairepav9. Lecture Notes in Civil Engineering, vol 76 (pp. 813-823). Springer, [6] Steyn, W. (2012). NCHRP Synthesis 433: Cham. doi:https://doi.org/10.1007/978-3-030- Significant Findings from Full-Scale 48679-2_76 Accelerated Pavement Testing. Washington, D.C.: Transportation Research Board. [10] Wacker, B., Scherkenbach, M., & Jansen, D. (2018). Temperaturuntersuchungen am und im [7] Wacker, B., & Jansen, D. (2018). Toolbox for Mobile Load Simulator MLS30. Bergisch APT program monitoring at BASt and Gladbach: Bundesanstalt für Straßenwesen optimized data handling on new test facility (BASt). duraBASt. TRB Annual Meeting. [11] Wirtgen. (2012). Wirtgen Cold Recycling [8] Wacker, B., & Jansen, D. (2020). Two Years of Technology (1st ed.). Windhagen, Germany: APT Program on the New Test Site duraBASt. Wirtgen. Chabot A., Hornych P., Harvey J., Loria-Salazar L. (eds) Accelerated Pavement Testing to [12] Wirtgen. (2017). BSM Cold Recycling, Transport Infrastructure Innovation. Lecture Laboratory Handbook. Windhagen, Germany: Notes in Civil Engineering, vol 96 (pp. 32-39). Wirtgen GmbH. Springer, Cham. doi:https://doi.org/10.1007/978-3-030-55236- 7_4 111 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia HAZARD PERCEPTION, TRAFFIC SITUATION OVERVIEW AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S DISEASE Kas Kamphuis, Msc SWOV Bezuidenhoutseweg 62, The Hague, Netherlands kas.kamphuis@swov.nl ABSTRACT In the Netherlands patients with early Alzheimer’s disease (AD) are allowed to drive if their fitness to drive is positively evaluated. The assessment consists of a clinical interview and an on-road examination. Previous research concludes that on-road performance by AD patients could be accurately predicted (92.7 percent) within a clinical setting on the bases of a combination of three assessments: 1) a clinical interview, 2) neuropsychological tests and 3) driving in a driving simulator (Piersma et al., 2016). In the current research at SWOV, the Dutch Institute for Road Safety Research, we examine if this method leads to the same prediction with a new group of AD patients. The neuropsychological tests include a series of reaction time tests, a hazard perception test, the Adaptive Tachistoscopic Traffic Perception Test and a driving simulator test drive. It is hypothesized that early AD patients react slower, have more difficulty overseeing traffic situations and perceiving possible hazards while driving. Results confirm the hypothesis in the sense that, compared to healthy controls, AD patients have a slower cognitive/motoric RT to visual/auditory/combined stimuli, need more time to react to traffic situations, have a harder time getting an overview of traffic situations and brake later in response to a driver that suddenly swerves onto the road in a simulated drive. In future research the reduction of hazard perception ability, the ability to get an overview of traffic situations and cognitive/motoric RT at which an AD driver can still safely participate in traffic and also the effectiveness of coping strategies could be investigated. Keywords: Alzheimer’s Disease, Traffic safety, Hazard perception, reaction time, Fitness to drive Introduction 1. INTRODUCTION experiences but are not yet impaired in performing their daily activities [5]. The pre-dementia phase is As people get older age related diseases become more commonly called mild cognitive impairment more frequent. Dementia is a well-known example (MCI). As the MCI progresses to actual dementia of such an age related disease. The most common the memory symptoms worsen (now also affecting type of dementia is Alzheimer’s Disease (AD) (50 – older memories) and linguistic and spatial 75%), followed by vascular dementia (20%), orientation deficits become apparent [5]. Reduced dementia with Lewy bodies (5%) and ability to plan, judge and organize start to interfere frontotemporal lobar dementia (5%) [1]. AD has with ability to complete complex tasks [6]. developed from a rare disease at the beginning of the 20th century to the sixth leading cause of death in An example of a complex task commonly performed the world [2,3]. AD is a progressive disease in which by older adults is driving. Driving is the preferred multiple cognitive domains are affected. In addition mode of travel for older adults and is a meaningful to the memory domain, also the domains of instrumental activity of their daily living as it attention, visuospatial abilities, executive enables them to visit family, friends and make use of functioning, language and praxis are frequently services they require [4,7]. impaired [4]. As AD affects domains that are highly relevant to Alzheimer’s disease includes a pre-dementia and driving skill, fitness to drive may be severely dementia phase. The pre-dementia phase is impaired [8]. Research shows that demented drivers associated with a loss of episodic memory. During forget to fasten their seatbelt, have difficulties with this phase patients often forget recent memories or 112 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia the identification of landmarks and traffic signs and they were informed about their diagnosis. get lost more often [8]. Alternatively, participants with AD could also find information about the research online and directly The degree to which a patient is impaired by AD is contact SWOV to apply for the research. often rated using the clinical dementia rating scale Participation was on a voluntary basis and after (CDR). The CDR is a 5-point scale that is used to assessment positive results could be used in an categorize six cognitive and functional domains: official relicensing procedure. After license renewal memory, orientation, judgment, problem solving the license lists a code referring to the driver’s and community affairs. The resulting scores range condition. In case of a negative assessment the from 0 (no impairment) to 3 (severe impairment). research concluded and the participant was strongly In the Netherlands drivers with very mild or mild advised to stop driving. The participant could then AD (a CDR-score of 0.5 or 1.0) are allowed to drive choose to cease driving immediately or to let their a car if their fitness to drive is positively evaluated. driving ability be assessed a second time directly by Fitness to drive is evaluated with a series of the Dutch driving test association (CBR). Failing the neuropsychological tests and an on road evaluation. CBR assessment would lead to a revoked license Previous research concludes that on-road immediately. Participants also received a monetary performance by AD patients could be accurately compensation for their time and effort. predicted (92.7 percent) within a clinical setting on In total 42 patients with AD were included in the the bases of a combination of three assessments: 1) research. Patients were aged 53 to 90 years (mean = a clinical interview, 2) neuropsychological tests and 75.4 years; SD = 7.4 years) and 29 (70.7%) of the 3) driving in a driving simulator [4]. patients were men. All participants held a valid Currently at SWOV, the Dutch Institute for Road driver’s license for between 35 to 68 years (mean = Safety Research, a follow-up study is performed to 54,7 years; SD = 7.4 years). On average participants examine if this method leads to the same prediction drove between 10.000 – 20.000 kilometers per year, with a new group of AD patients. The due to a missing response one participant was neuropsychological tests include a series of reaction excluded from this average (n =41). time tests, a hazard perception test, the Adaptive 2.2. Control-group participants Tachistoscopic Traffic Perception Test (ATAVT, which gives insight into the ability to get an In total 27 healthy individual participated in the overview in traffic situations) and a driving study. The criteria for inclusion were: 65 years of simulator test drive in which another driver suddenly age or older, no diagnose of a different neurological swerves onto the road from a parking spot. These or psychiatric condition that is detrimental to driving four tests are of particular importance to road safety ability or usage of medication that has a negative as these indicate how quickly a dangerous traffic effect on driving ability. The minimal age for situation is recognized and subsequently is acted inclusion was higher for control participants than for upon. participants with AD to keep the groups comparable. The aim of this study was to determine if the reaction As AD is an age related disease the chance is higher time and hazard perception ability of drivers differed for the control group participants to be younger, this between healthy drivers and drivers with early AD. is somewhat balanced by the higher age limit for It is hypothesized that early AD patients react inclusion. The majority of this group is very healthy slower, have more difficulty getting an overview of and has no age-related driving problems. traffic situations and have more difficulty perceiving Participants were recruited through means of a call possible hazards while driving. on the SWOV site and through word of mouth, meaning participants were asked if they knew 2. MATERIALS AND METHODS someone who would also be interested in This section describes the materials and method participating in our research. Participants then used in the current research. contacted SWOV by e-mail or telephone and were sent more detailed information about the research by 2.1. Participants with AD post. Participation was on a voluntary basis and also the control-participants received a monetary Patients with AD were assessed at hospitals and compensation for their time and effort. The healthy nursing homes. Inclusion criteria were that the participants were aged 65 to 86 years (mean = 71.6 patient: was 50 years of age or older, had a CDR years; SD = 5.4 years) and 20 (74.1%) of the score of 0.5 (very mild AD) or 1 (mild AD), had not participants were male. All participants held a valid been diagnosed with a different neurological or driver’s license for between 40 to 62 years (mean = psychiatric condition or used medication that was 51.6 years; SD = 4.6 years). On average participants detrimental to driving ability and wished to keep drove between 10.000 – 20.000 kilometers per year. driving. If the patient met the inclusion criteria a folder was handed to them by their geriatrician, neurologist, psychiatrist or general practitioner after 113 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia 2.3. Measuring tools higher the overall score is automatically 0.5 or higher. The current research at SWOV is a longitudinal validation study of the research done by Piersma et 2.6. Vienna Test System (VTS) al. (2016). In both studies the participants take part The RT-tasks [11] and the ATAVT [12] are in a series of pen and paper tasks, followed by a administered to the participant using the Vienna Test neuropsychological assessment which consists of System [13]. The VTS is a test system used for tasks that measure cognitive functions that are computerized psychological assignments to measure important to driving and lastly four test drives in a abilities of the participant that are relevant to road driving simulator. safety. It is a helpful tool to assess fitness to drive in The measuring tools that are relevant to the current a participant. paper consist of a driving questionnaire, the clinical dementia rating-scale (CDR-scale), a series of 2.6.1. Reaction time (RT) tasks Visual, reaction time tasks, the Adaptive Tachistoscopic Auditory & Combined. Traffic Perception Test (ATAVT), the Hazard Three RT-tasks are administered: a visual task, an perception task developed at SWOV and a driving auditory task and an inhibition task in which visual simulator ride. I recommend consulting the paper of and auditory cues are combined. In the visual task Piersma et al. (2016) for an overview of all the participant looks at a black circle and has to press measuring tools, including those outside the scope a button the moment the circle changes to yellow. In of the current paper. the auditory task the participant has to press a the 2.4. Driving questionnaire moment an auditory cue is heard. In the combined task the participant should only press the button All participants were requested to fill in a driving when the black circle turns yellow and at the same questionnaire. The driving questionnaire was time an auditory cue is heard. The button should not adapted from the Safe Driving Behavior Measure [9] be pressed if: 1. the circle turns red 2. the circle turns and consists of three parts: a demographical profile yellow without an auditory cue or 3. only an auditory of the participant (7 questions), a driving profile cue is heard. Between cues the participant rest with questions concerning driver experience, his/her finger on a ‘rest button’. The cognitive RT is mileage, the use of medication and knowledge of measured from the moment the cue is presented until traffic laws (23 questions) and questions relating to the moment the index finger moves from the rest the driving skills of the older driver (54 questions). button. The motoric RT indicates how fast the These 54 questions were translated from the Fitness- participant is able to press the button after the correct to-drive-screening measure or FTDS by Classen et stimuli is recognized. It is measured from the time al. (2013). the finger of the participant leaves the rest button till 2.5. Clinical Dementia Rating (CDR) the moment the answer button is pressed. The VTS provides the mean and standard deviation of the The CDR rating is derived from a semi-structured cognitive /motoric RT. interview with the patient and their informant, they are questioned separately from one another. The 2.6.2. ATAVT input of informants is of particular importance because they can provide valuable background The ATAVT (Schuhfried, 2009) is used to measure information about the condition of the patient and the ability of the participant to get an overview in their answers also serve as a means to ‘fact-check’ traffic situations. The participant is presented with a the answers of the patient. The aim of the interview series of photographs that only stay visible for is to assess the degree to which AD impairs the roughly one second. After each photograph the functioning of the patient in six domains: memory, participant has to choose which elements were orientation, judgment, problem solving and present in the shown photograph. The list of options community affairs. Examples of questions are: consists of: pedestrians, cars, (motor)cyclists, traffic ‘Could you tell me what you did last weekend’ signs and traffic lights. The ATAVT is an adaptive (memory), ‘What day is it today? / Where are we test, meaning the number of elements shown in the right now?’ (orientation) to ‘How many 5-cent coins photographs increase or decrease based on the are there in a euro?’ (problem solving). Each domain performance of the participant. A psychometric is scored on a 5-point scale: 0 (not impaired), 0.5 model [14] is used to estimate a general performance (very mildly impaired), 1 (mildy impaired), 2 parameter, that takes into account the difficulty of (moderately impaired) and 3 (severely impaired). the items that participants saw and their performance Using the scores on the six domains an overall CDR on them. The parameter indicates the ability of the score can be calculated [10]. The memory-domain participant to get an overview in traffic situations weighs more heavily compared to the secondary given the age of the participant. A score above 0 domains. For example if memory is scored as 1 or reflects a better ability, below 0 reflects a worse ability in comparison to the participants age group. 114 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia 2.7. Hazard perception test measure is whether or not the participant braked for the car. In the hazard perception test participants were presented with 25 photographs of traffic situations. 2.9. On Road assessment The photographs were taken from the driver’s point Participants took part in an on-road assessment of view and the current travelling speed was during daylight hours and in their own car. The displayed on the mock displayed (see Figure 1). assessment was carried out by approved experts Participants had 8 seconds to decide whether they from the CBR. These experts are trained to evaluate would do nothing, slow down by releasing the gas whether an impaired driver is still fit to drive. The pedal or brake in the given traffic situation. The expert did not know if the participant was part of the participants took the test either with or without time healthy or the AD group. The expert makes use of a pressure. This decision was made by the person that Test Ride Investigating Practical Fitness to drive administered the test to prevent the participant from (TRIP) form consisting of 60 items to evaluate getting upset/stressed as in some cases the fitness to drive in the participant. Each item is scored participant was unable to respond within the time sufficient, doubtful or insufficient. The 60 items limit. The number of correct answers and reaction concern 11 subject: place on the road, following times were measured. Separate statistical outcomes distance, speed, looking behavior, traffic signs, are presented in the results section for participants in taking over and passing, anticipation, the timed condition and all participants together communication with other road users/pedestrians, (timed/non-timed). judgement of traffic situations, operation of the vehicle and overall impression. The expert also gives an overall judgement of pass, doubtful or fail. 3. PROCEDURE SWOV collaborated with several hospitals to recruit participants with AD, patients took part on a voluntary basis. After the patient was informed about their diagnosis and the possible consequences for their driving ability, the research at SWOV was discussed. Patients that met the inclusion criteria and wished to keep driving received a folder by their healthcare professional detailing the specifics of the Figure 1: Example of a trial used in the Hazard research. The patient or informant subsequently Perception test. contacted SWOV to apply for the research. In other 2.8. Driving simulator rides cases the patient gave their healthcare professional permission to share their contact information with At SWOV a fixed base driving simulator was used SWOV. The patient was then contacted by a SWOV to assess various aspects of driving behavior. The researcher shortly after to provide more details about simulator consisted of an open cabin mock up with the research and schedule an appointment. In other a steering wheel, gear box, gas pedal, brake pedal, cases participants or their family looked up clutch and simulated driving sound. Three screens information on the SWOV website and contacted the provided the participant with a 200⁰ view on the road research team directly. Healthy participants were and surroundings. The dashboard and rear view recruited from the general community through mirror were visualized on the middle screen, the car means of a call on the SWOV website detailing the windows and side mirrors were visualized on the specifics of the research and through word of mouth. screens to the left and right of the participant. The All participants provided their written consent to graphical interface was designed and scenario’s take part in the research. were programmed in previous research [4]. As in the original study by Piersma et al. (2016) the After familiarizing themselves with the driving assessment of each participant was divided into two simulator in a practice drive the participants took parts. In the first part the participant is invited to part in a lane tracking ride, an intersection ride and SWOV to conduct clinical interviews, a a merging ride. In this paper we will only consider neuropsychological assessment and driving an event that takes place during the intersection ride, simulator rides. The second part consists of an on see [4] for a detailed description. Near the end of the road assessment. The day of testing and the on road intersections ride the participant is confronted with assessment make up one cycle of the research. Each a car turning onto the road from a parking bay. It was participant goes through a maximum of three cycles. programmed in such a way that the participant had The results on the day of testing and the on road to brake strongly to avoid collision. The outcome assessment together lead to an evaluation. The on road assessment weighs more heavily in this 115 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia evaluation compared to the (neuro)psychological Table 1: Comparison of patients with AD and assessment. If participants are positively evaluated healthy participants according to general the research continues and they are again invited characteristics. after a few months to start the subsequent cycle. A AD Healthy Characteristics participant is only allowed to continue the research (n = 42) (n = 27) Age, mean 75 (7.3) 71.1 (5.4) if the on road assessment is positively evaluated. All Male sex, No. (%) 29 (69%) 20 (74.1%) participants received a monetary compensation for Education, mean of 7 3.38 (1.1) 4.41 (.9) participating in the study. This compensation was stages (SD) either 25, 50 or 100 euro depending on the number CDR-score, No. (%) of cycles the participant took part in during the 0 1 (2.4%) 27 (100%) 0.5 38 (90.5%) 0 (0%) study. Patients with AD that were positively 1 3 (7.1%) 0 (0%) assessed after a cycle could use this outcome in an AD Healthy official relicensing procedure. (n = 41) (n = 26) Driving experience 54.7 (7.4) 51.6 (4.6) 4. STATISTICAL ANALYSIS AND mean (SD), y RESULTS Table 2: Comparison of participants with AD and Analysis were performed using IBM SPSS Statistics healthy participants according to outcome measures. 25. Patients with AD and healthy participants were Reaction time task, time in ms compared using independent sample t-tests for the P-value AD N Healthy N ATAVT and RT-tasks. Significance level alpha was (df) set at .05. In case the data was non non-normally Visual 347.2 (74.7) 40 302.2 (52.8) 27 .009 (65) distributed a Mann-Whitney U test was used instead cognitive Visual 271.5 (84.3) 42 219 (62.9) 27 .007 (67) of the independent sample t-test. Furthermore, a motoric Chi-square test was used to analyze whether the Auditory 299.1 (71.6) 42 264.4 (65.7) 27 .047 (67) groups differed on braking during the event in the cognitive simulator drive. Auditory 235.6 (72.3) 42 193.7 (62.4) 27 .01461 motoric 4.1. Missing values and outliers Combined 602.3 42 503.4 27 .001 (67) cognitive (124.4) (106.9) For two tests results were not distributed normally. Combined 284.7 (85.4) 42 229.9 (74.8) 27 .008 (67) In both cases this was related to outliers due to motoric participants who were not able to understand or Hazard perception task execute the task, which were removed from further Answers 14.5 (4.11) 42 18.22 (2.81) 27 .0006 analysis. This was done to prevent putting the AD correct participants under unnecessary stress. Reaction time, 5.39 (0.76) 30 4.39 (0.73) 27 .000 (55) sec (timed) After removal of these outliers the data was Reaction time, 9.84 (9.31) 42 4.39 (0.73) 27 .003 (67) normally distributed. There was a similar situation sec (all with regards to the ATAVT test analysis. One participants) participant could not complete the test and was ATAVT therefore excluded from the analysis. Furthermore, Performance -1.145 41 0.487 27 .000 (66) several participants were excluded from the parameter (1.179) (0.915) simulator rides analysis due to simulator sickness. Lastly, in the brake RT analysis participants that Simulator rides Braking, 22/5 27 21/0 21 .059 (1)2 failed to brake were excluded from that analysis. yes/no Brake 1.82 (.42) 22 1.27 (.26) 21 .000 (41) 4.2. Results reaction time, sec In this section results from the different measuring tools are presented for drivers with AB and healthy controls. In Table 1 the groups are compared Results indicate that the AD participants had a according to general characteristics. In Table 2 the significantly shorter (motoric) reaction time to groups are compared according to outcome visual cues compared to the reaction time of the measures. participants in the control group. Similar results were found on the auditory cue RT-task for both cognitive- and motoric RT. In the combined task, when presented with combined stimuli and certain 1 As the values were not normally distributed a Mann-Witney U 2 A chi-square test was used. One of the cells had a count of less test was used. This test does not provide degrees of freedom than five which violated an assumption of the Chi-square test. To correct this the Fisher’s Exact Test was performed. 116 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia responses were to be inhibited again a similar result research as well [4]. The results on the driving was apparent for both cognitive and motoric RT. simulator ride showed that there were 5 AD participants out of 27 who did not manage to brake On the hazard perception task the results show that for the car entering the road while this did not AD participants answered fewer questions correctly happen for the control group participants. However, compared to the control group. AD participants also it has to be noted that this result only approached needed significantly more time to provide an answer significance. compared to healthy participants. Reaction time, hazard perception ability and the The two groups also differed on the ability to get an ability to get an overview quickly are instrumental overview in traffic situations. Again results showed to take part in traffic safely as the driver needs to be a significantly worse performance for the AD able to handle complex traffic situations. It is vital participants. that the driver is able to quickly recognize what During the simulator ride an event took place where elements make up the traffic situation, how to a driver suddenly turned into the road from a parking prioritize them and what the appropriate action to bay, forcing the participant to brake heavily to avoid take is (e.g. giving or taking priority). AD drivers collision. In general AD participants failed to brake need more time to get an overview in traffic five times while the healthy participants failed 0 situations and respond to hazards later compared to times. This effect however only approached healthy controls. Once the hazard is recognized the significance. After excluding the participants who AD driver also has a slower motoric RT which had not braked, the driver braking reaction times of further delays the intended action (e.g. braking or those who did brake were compared among the evading). The AD driver is therefore at an increased groups. Results again show that AD participants risk of being involved in a traffic accident. reacted slower to the car turning into the road than the control participants did. AD drivers use coping strategies to mitigate their impairment, for example by driving more slowly to 5. DISCUSSION allow for more time to react or by avoiding complex traffic situations altogether (e.g. only drive outside The present study aimed to determine if reaction of rush hour or only on non-busy roads). Research time, hazard perception and the ability to get an shows mixed results in the effectiveness of these overview in traffic situations differ between healthy coping strategies. It was shown that avoidance drivers and drivers with very mild (CDR = 0.5) or strategies employed by older adults are related to mild (CDR = 1) AD. It was hypothesized that drivers reduced driving skills [17]. This implicates that with AD would react slower, have more difficulty coping strategies might actually worsen ones ability getting an overview in traffic situations and have to take part in traffic safely. On the other hand more difficulty perceiving possible hazards while research shows that older adults can use tactical driving. compensation strategies to reduce crash risk [18]. More research is needed to: 1) determine at what The results are relatively straightforward. Patients reduction of hazard perception ability, the ability to with AD disease reacted significantly slower get an overview and cognitive/motoric RT an AD cognitively than healthy controls whether it is to driver can still safely participate in traffic and 2) visual, auditory or combined choice stimuli. These determine the role that coping strategies play. A findings are in line with previous research [4,15,16]. naturalistic study of AD drivers in which (motoric) Motoric RT, the time to press the button after RT, traffic overview and hazard perception recognizing the correct stimuli, was also slower for outcomes are related to crash or near-crash AD participants in all RT tests compared to healthy involvement could be performed to this end. This controls. This finding is also in line with previous could aid in deciding when it is the right time for the research [4]. It was also apparent from the results AD driver to stop driving. This is of vital importance that AD participants had a slower response to not only with traffic safety in mind but also the hazards compared to healthy participants. The wellbeing of AD patients as driving is a meaningful results on the hazard perception task also showed instrumental activity of daily living that enables that AD participants made more incorrect decisions them to stay socially connected to the community. and for the correct ones, they had a longer reaction time. It should however be taken into account that REFERENCES the participants responded to static images of traffic situations and therefore could not see the traffic [1] Cunningham, E.L., McGuinness, B., Herron, B. situation develop like you normally would be able to & Passmore, A.P. (2015). Dementia. Ulster Med while driving. J, 84(2), 79 – 87. Patients with AD also had more trouble to quickly [2] Cornutiu, G. (2015). The epidemiological scale of Alzheimer’s disease. oversee traffic situations, this is in line with previous Journal of clinical medicine research, 7(9), 657. 117 Kas Kamphuis YRS 2021 HAZARD PERCEPTION AND REACTION TIME IN DRIVERS WITH EARLY ALZHEIMER’S 15. -17. September DISEASE Portoroz, Slovenia [3] Haque, R. U., & Levey, A. I. (2019). [10] Hughes, C.P., n.d.. Clinical Dementia Rating Alzheimer’s disease: A clinical perspective and (CDR) Scale. Retrieved May 12, 2021, from: future nonhuman primate research https://www.mdcalc.com/clinical-dementia- opportunities. Proceedings of the National rating-cdr-scale#evidence Academy of Sciences, 116(52), 26224-26229. [11] Schuhfried, G. & Prieler, J. (2008). RT. [4] Piersma, D., de Waard, D., Fuermaier, A., SCHUHFRIED gmbh, Mödling. Davidse, R., Bredewoud, R., Brouwer, W., & [12] Schuhfried, G. (2009). ATAVT. Tucha, O. (2016). How to investigate fitness to drive of patients with Alzheimer’s dementia. In SCHUHFRIED gmbh, Mödling. International Conference on Traffic and [13] Schufried. (n.d.). The Vienna Test System. Transport Psychology 2016. Retrieved March 23, 2021, from: https://www.schuhfried.com/en/vienna-test- [5] Guarino, A., Favieri, F., Boncompagni, I., system/ Agostini, F., Cantone, M., & Casagrande, M. (2019). Executive functions in Alzheimer [14] Rasch, G. (1993). Probabilistic models for some disease: a systematic review. Frontiers in Aging intelligence and attainment tests. MESA Press, Neuroscience, 10, 437. Chicago, Illinois. [6] Förstl, H., & Kurz, A. (1999). Clinical features [15] Hutchison, K. A., Balota, D. A., & Ducheck, J. of Alzheimer’s disease. European archives of M. (2010). The utility of Stroop task switching psychiatry and clinical neuroscience, 249(6), as a marker for early-stage Alzheimer's disease. 288-290. Psychology and aging, 25(3), 545. [7] Breen, D. A., Breen, D. P., Moore, J. W., Breen, [16] Tse, C. S., Balota, D. A., Yap, M. J., Duchek, J. P. A., & O'Neill, D. (2007). Driving and M., & McCabe, D. P. (2010). Effects of healthy dementia. Bmj, 334(7608), 1365-1369. aging and early stage dementia of the Alzheimer's type on components of response [8] Toepper, M., & Falkenstein, M. (2019). Driving time distributions in three attention tasks. fitness in different forms of dementia: an Neuropsychology, 24(3), 300. update. Journal of the American Geriatrics Society, 67(10), 2186-2192. [17] Schulz, P., Beblo, T., Spannhorst, S., Labudda, K., Wagner, T., Bertke, V., Boedeker, S., [9] Classen, S., Wang, Y., Winter, S. M., Velozo, C. Driessen, M., Kreisel, S.H. & Toepper, M. A., Lanford, D. N., & Bédard, M. (2013). (2019). Avoidance behavior is an independent Concurrent criterion validity of the safe driving indicator of poorer on-road driving skills in behavior measure: A predictor of on-road older adults. The Journals of Gerontology: driving outcomes. American journal of Series B, 75(10) , 2152-2161 . occupational therapy, 67(1), 108-116. [18] De Raedt, R. & Ponjaert‐Kristoffersen, I. (2000). Can strategic and tactical compensation reduce crash risk in older drivers?. Age and ageing, 29(6), 517-521. 118 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Aristi Karagkouni, PhD Candidate Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece arkaragk@econ.duth.gr Maria Sartzetaki, Senior researcher/Adjusted lecturer Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece University Neapolis, Pafos, Cyprus msartze@econ.duth.gr Dimitrios Dimitriou, Professor Associate Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece ddimitri@econ.duth.gr ABSTRACT In a competitive business environment, series of decisions should be connected to data leverage of strategic importance – the data-driven management could mitigate risks and increase awareness, while re-engineering the out-performance production process will be much easier to be implemented. This paper deals with the presentation of the key components for developing a data driven dynamic corporate management to assess uncertainty and unpredicted risks. The business sectors of the analysis focused on transport and supply chain infrastructure operators where the capital-intensive business environment may dramatically affect by out of time decisions. Adopting a System of System (SoS) approach the utility of data driven service in the real business is addressed. By a dedicated literature review the key areas for action are highlighted and the spots for innovation are depicted. The analysis includes the results from a questionary survey, where the needs of transport infrastructure operates are evaluated and the key priorities are presented. Key objective is to present the transport market needs for a data driven management system (platform) upon the development of intelligent services to support decisions for operators and large enterprises managing transportation systems and critical transport infrastructure. Conventional wisdom is to present the key components for an intelligent data-driven risk assessment system for the different group of potential users’ in transport sector, which addressing the stakeholder’s expectations, shareholder’s prospects and management abilities to response. The research outputs will provide essential messages for managers and decision makers towards business intelligence and corporate management performance optimization, especially in sector of transport and supply chain. Keywords: Intelligent services, Research Infrastructure, Prospects and Transportation 1. INTRODUCTION which includes physical networks, terminals and intermodal nodes, information systems, as well as The research needs are demanding both at refuelling and electrical supply networks, is disciplinary level – requiring more and more often necessary for the safe, secure operation of road, rail, the acquisition of diverse complementary data from civil aviation, inland waterways and shipping and is different methods and instruments – and also across crucial to the European Union’s (EU) economic disciplines – as the study of complex phenomena growth and social development. Unlike other demands to jointly analyse data obtained from transport operations, transport infrastructure is Research Infrastructures belonging to different owned by public sector organisations at country or scientific domains [1]. Transport infrastructure, regional level [2]. An issue of concern arises from the 119 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia growing pressure on existing transport systems, 2. BACKGROUND RESEARCH AND combined with regularly underfunded maintenance LITERATURE activity, the growing digital connectivity, infrastructure vulnerabilities to man-made or natural 2.1. Research Infrastructure (RI) disasters and the carbon efficiency of infrastructure objectives reuse. Life-cycle optimisation and integrated Since the business ecosystem in which organizations efficient operation can lead to improvements in the act, operate and regulate is changed predictions, existing and future infrastructure stock. forecasts and assessment of scenarios have always Intelligent infrastructure is the deep embedding of been necessity. Especially, recent years where sensing, computing, and communications capabilities technological innovation and external events such as into traditional urban and rural physical climate change and epidemic are drivers for infrastructures such as roads, buildings, and bridges functional changes in market, management and level for the purpose of increasing efficiency, resiliency, of service. Therefore, in today’s management the and safety. According to [3] “Intelligent wording of the term “business resiliency” is wider Infrastructure” is defined as “the integrated sensing than the typical evaluation of the financial and data analytics with municipal capabilities and performance and it also deals with the internalization services that enable evidence-based operations and of the risks caused of the external factors [5]. The decision making” which is: evaluation of the impact of these factors in business  ecosystem is know the baseline of the risk assessment Descriptive: Provides an accurate and promoting new areas of strategic and business timely characterization of current state, e.g., planning in the frame of operational contingency, water level in a storm drain or traffic product line continuity and financial productivity. congestion.  Prescriptive: Recommends immediate and In transport sector, organization that cannot react near-term actions, e.g., re-routing traffic or quickly to changing conditions and that cannot have dispatching onsite service personnel. plans to mitigate future risks, in other words cannot  Predictive: Anticipate future challenges and foreseen the future with a degree of accuracy are opportunities, based on assessment of the boomed to extinction and are characterized as un- current state, patterns of past activity and resilient [6]. In this outline, key driver of creating available resources and capabilities, e.g., Intellignet Transport Systems bargains with the street-level flooding by incorporating water advancement of precision for estimations around sensors, weather patterns and runoff long-term in medium- and long- term time skyline. capabilities. The complexities of nowadays worldwide commerce  Plan-Full: Guides complex decision making environment and globalization of transport trade have and scenario planning, incorporating boosting the transport division in tall request levels economic data, to inform future investment. pulling in expansive sum of capitals to oblige the growing demand; but on the other create huge sums This paper deals with the analysis of the real of information and an overpowering got to extricate requirements for digital services provided by exact, important and valuable data from these ‘Intelligent Research Infrastructure for Shipping, information [7]. Supply Chain, Transport and Logistics’ (EN.I.R.I.S.S.T.) project in order to support policy Research Infrastructure are essential tools for making, strategic and business planning and decision transport sector ecosystem to support planning, making in transportation sector [4]. EN.I.R.I.S.S.T. is management and decision making to meet the new a unique and pioneering Research Infrastructure that conditions of the future for which the knowledge is aims to fill a significant existing research gap in the imperfect. In addition, the need for business analytics fields of Shipping, Supply Chain and Transport in cuts across all functional lines as well as across all Greece. The research outputs provide results about types of organizations [5]. Scenario planning based the real needs for data analytics, event observation, on accurate data are needed in master and business cost-benefit analysis, market trends and forecasting, planning, finance, marketing, personnel and for a variety of potential users in supply chain production areas covering cross the board of transport business ecosystem. sector from small business units up to large multinational companies and national organizations. RIs must be recognized as long-term strategic investments at all levels, deeply rooted in society, and indispensable both for enabling and developing excellence in their respective scientific domains, and also as key players contributing to competitiveness with a very large perimeter. The long-term benefits of Research Infrastructures to society at large are 120 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia unquestionable irrespective of the size or scientific and Transport in Greece. It combines the collection focus of the RI concerned [8]. and processing of data (with the aim of protecting privacy and copyright), the development of 2.2. Intelligent Infrastructure contribution innovative models and programming techniques, the to Research and Innovation development of useful applications, secure and user- Intelligent infrastructure is the profound implanting friendly, and finally the development of digital of detecting, computing, and communications observatories aimed at support for public and private capabilities into conventional urban and country stakeholders (businesses, public bodies, research physical frameworks such as roads, buildings, and organizations, etc.) [4]. In this way, the vision of bridges for the reason of expanding effectiveness, EN.I.R.I.S.S.T. is to become a center of excellence resiliency, and security. For example, inserting that will promote and support research in its scientific controllers, crossing point schedulers, and sensors fields. The objectives of EN.I.R.I.S.S.T. are: along streets makes modern capabilities to control  Develop an intelligent research and business activity signals and optimize activity stream. platform to support key economic activities and Creating intelligent infrastructure gives the implies to small and medium-sized enterprises active in the supply modern efficiencies in control era and areas of research infrastructure interest. transmission, moved forward versatility to common  Collect process and provide researchers and and human-originated disturbances, and streamlined users with information & tools on national & integration of unused sources vitality sources such as international passenger & freight transport wind [3]. including sea, air, inland and intermodal Intelligent Infrastructure can incrementally transport. increment operational execution and the capacity to  Support stakeholders in original research, direct changes through decision support. These investment plans and policies (academic “loops” or cycles of learning span computerization community, researchers, infrastructure and choice back to the possible generation of operators, private & public companies, policy generalized information. For case, progressed makers) transportation frameworks might incrementally learn  To create a multi-dimensional institution of to oversee diverse designs of activity, at that point economic & research development for Greece, give choice back for proactively overseeing by creating new & enhancing existing networks extraordinary cases (e.g., disaster reaction), at that that will ensure the flow of knowledge & point bolster arranging and prioritization for unused information road/control adjustments, and at long last report The Intelligent Research Infrastructure generalized information that can be connected over EN.I.R.I.S.S.T. is composed of 11 Partners which diverse cities with changing transportation include 8 Educational Institutions and 3 Research capabilities. Centers in Greece. Researchers specialized in One of the chief difficulties in developing integrated shipping, supply chain and transportation, as well as intelligent tools in transport sector is an unexpected software developers, come together and unite their and significant shift of the economic outputs caused expertise. This specialization promotes research on unpredicted events. Among such factors are excellence, ensures an integrated approach to the changes in oil and energy prices, inflation surges, challenges of the above sectors, leading to successful currency rates, mobility restriction measures and implementation of the infrastructure. In addition to border/visa policy affected demand in small or large the major partners, the EN.I.R.I.S.S.T. infrastructure regions and most of the times impact even whole ecosystem is made up of Industry Representatives, economic systems such is Euro-zone in time of Private Companies, Public Enterprises, Policy COVID-19 pandemic. Therefore, the cornerstone of Makers, Independent Researchers and Citizens. the term “intelligence” deals with the use of large size information (megadata) to (a) predict future level of 3. CONCEPTUAL ARCHITECTURE demand or economic output (revenues, prices, etc); During the architectural design process, critical (b) assess risks and business resiliency; and (c) issues related to the development and operation of the provide accurate data to support decisions. Intelligent infrastructure were addressed. It was found that many Infrastructure has the potential to transform daily life of the requirements of the users were conflicting and and civic services across many configurations of that the available resources are limited. A service- municipal systems and services. oriented approach is proposed for infrastructure 2.3. EN.I.R.I.S.S.T. Project Overview development. According to it, each functional unit is implemented as a service, which is integrated into a EN.I.R.I.S.S.T. is a unique and pioneering Research platform that will contain a set of different services Infrastructure that aims to fill a significant existing and communicates in a predetermined way with the research gap in the fields of Shipping, Supply Chain rest of the components. 121 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia A mechanism is required to maintain flexibility in the The EN.I.R.I.S.S.T research infrastructure will design of the architecture, ie to be able to add and implement and provide a set of platforms for remove services without significantly increasing the Shipping, Supply Chain and Transport; (a) the computational cost. It is noted that the development ecomarine platform, (b) the maritime heritage of the services of the various platforms will take place platform, (c) the passenger platform, (d) the in parallel. In order for EN.I.R.I.S.S.T to provide new infrastructure platform, (e) the inland and intermodal opportunities in its areas of activity, in which freight platform and (f) a decision support tool for complex processes with economic, social and shipping and financial markets. Each platform aims technological elements are performed, it is required to make available a set of data and web services (with in the design of its architecture to be taken seriously: as user-friendly interfaces as possible), and data  analysis tools/processes. These services aim to The size of the data that its platforms have provide access to data and through their analysis to to manage.  help both researchers and companies to devise new The number of calculations of its services strategies to become more competitive. In order to be prosedures able to do this, it is necessary to somehow record the  The complexity of the procedures needs of the RI’s potential users and then to form performed by its services. some user groups.  All modern technological and business trends. The presentation of the research infrastructure architecture follows a logical path that starts from the In general, the research infrastructure will use supported use cases and proceeds to its functional multiple data sources and a set of new methods that parts at a conceptual level. The architecture must be process and analyze data from these data sources. It scalable and flexible to modifications. In order to be therefore makes sense for its applications to have successful, the architecture design must cover the high data requirements. This has a decisive influence following aspects: on EN.I.R.I.S.S.T's development and conservation  Easy to use strategy. Based on the above, it appears that  Efficiency EN.I.R.I.S.S.T will integrate a large number of  Security applications, which will provide useful services to  Scalability research centers and companies. A wide variety of  Reliability technologies will be used to implement them, which  Maintenance testifies the inherent complexity of the systems and services involved in data analysis. In fact, their complexity increases as they develop and expand. 3.1. Modeling Framework Development The overriding consideration in choosing an analysis The presentation of the architecture begins with the methodological framework is that the results must use cases supported by the infrastructure system and facilitate the decision making process of the then the consolidation of the necessary elements at organizational managers. The essential requirement the functional level. Based on the usage scenarios, the is that the analysis outputs should be understandable cases of using the system are determined. The and reasonable for the users. Therefore, the results sequences of actions that implement the necessary outputs should be accurate, timely and understood by processes to serve the usage scenarios are identified the users so that to use the analysis outputs help them while at the same time the independent functions that to produce better decisions. are combined to form a use case are provided in an abstract way. These cases of use of the system reveal The recognition that any analysis techniques operate the degree of complexity and the need for on the data generated by historical events and communication and orchestration of the various situations leads to the identification of the following elements that will be integrated into the four step in Infrastructure architectural developing EN.I.R.I.S.S.T infrastructure. The proposed process: architectural approach aims to cover all complex scenarios of use of the RI while providing great flexibility in processes of expansion and change of the system. 122 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia Figure 1: Data inflow and analysis levels for developing RI services 1. Data collection – ( Development of data base) the transport sector, such as transport infrastructure operators, transport enterprises and authorities. Data Consists of data gathering from operators, transport inflows are analyzed by using appropriate modeling enterprises and the economic system related to tools in order to rate the developed services, demand (traffic), socioeconomic characteristics, according to users’ needs and expectations, in terms operational characteristics (capacity) etc. Suggest the of support decision making in the transport sector.In importance of getting the proper data and making order to develop the EN.I.R.I.S.S.T. Research sure that are those are needed. This is often the most Infrastructure, it is necessary to consult with potential challenging step of the entire RI development and the project users in order to gather their requirements and most difficult to monitor since subsequent step can be needs regarding the research infrastructure’s services performed on data whether relevant to the problem at related to shipping, transport and supply chain hand or not. Collection and quality control problems activities. This process is performed in 3 steps, as usually abound whether it comes necessary to obtain shown in Figure 2. pertinent data in an organization. 2. Model building – (Creation of metadata warehouse) Consists of the analysis of the incoming data, using quantitative methods and modelling analysis tools, in Figure 2: Depiction of process to define the RI outputs order to create KPIs and forecasting. Modell building involved fitting the collected data into different In the first step, interviews are conducted with quantitative models delivering a variety of outputs. executive professionals and researchers from a Often a balance must be struck between a variety of organizations related to the above sectors, sophisticated modelling framework that offers in order to define the main pillars of the project’s slightly more accuracy and a simple approach that is deliverables. Then, a questionnaire survey is applied easily understood and gains the support of users. in order to gather and review the potential users’ Obviously, judgment is involved in this selection needs and prospects in shiping, transport and supply process. chain sectors. The third and final stage concerns the rating of the potential users’ needs and their 3. Model extrapolation – (Service outputs and relevance to the services provided by EN.I.R.I.S.S.T. interface) research infrastructure. The main purpose of this Consist of the actual modeling extrapolation that infrastructure is to create an intelligent tool, which occurs once the appropriate data have been collected, will be able to receive and analyze data in real time storage and tested. Often the accuracy of the process in order to support decisions in transport, shipping is checked for recent periods in which actual values and supply chain sectors. For this reason, it is are known. It is essentially the creation of an particularly important that the services offered by the intelligent decision support system (DSS), as shown infrastructure and the platforms that will be created in Figure 1. Monitoring the RI’s potential users’ meet the needs of its potential users. needs and expectations is a key pillar of this process, The ability to manage the development of a such as the developed services are mainly aimed at process is crucial for the value of the outcome. The professionals and researchers in the transport sector. usefulness and utility of the service can be improved RI’s data bases are powered by data providers from 123 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia if management adopts a more realistic attitude. The this paper pointed out that the development of a RI in analysis and services outputs should not be viewed as transport sector must adhere two primary rules: a substitute for prophecy but rather as the best way of  The metadata must be technically correctly identifying and extrapolating establish patterns of and produce outputs accurate enough to relationships to support planning, risk assessment meet the transport system ecosystem; and decision making. Given that, the key questions  raised if the outputs is to be properly conducted, The procedure and its results must be where such questions are: effectively presented (interface) according to the needs of managers and administrators Regarding the RI offered services content: so that the modelling outputs are utilized in  decision making process providing Why the analysis outputs needed?  advantage to transport business resiliency. Who will use the offered services and what are their special requirements? Acknowledgement  What level of detail or aggregation is The paper research outputs is supported by the required and what is the proper time research project “ENIRISST – Intelligent Research horizon?  Infrastructure for Shipping, Supply Chain, Transport What data are available and will the data be and Logistics” which is implemented under the sufficient to generate the needed analysis Action “Reinforcement of the Research and outputs (offered services)? Innovation Infrastructure”, funded by the Operational Programme "Competitiveness, Entrepreneurship and Regarding the RI modelling framework for providing Innovation" (NSRF 2014-2020) and co-financed by metadata: Greece and the European Union (European Regional  How accurate and in what time frame or Development Fund). measurement type to be?  Will the modelling outputs be made in time REFERENCES to support decision making?  [1] European Strategy Forum on Research Does the outputs clearly understand how it Infrastructures (ESFRI) (2018) Roadmap & will be used in the organization? strategy report on research infrastructures, Part 1,  Is the feedback process available to evaluate pp. 8-14. outputs after it is made and to adjust the [2] European Commission, Commission staff calculations accordingly? working document — Towards clean, competitive and connected mobility: the 4. CONCLUSION contribution of transport research and innovation to the mobility package, 2017, SWD(2017) 223, Transport infrastructure, which includes physical Brussels. networks, terminals and intermodal nodes, information systems, as well as refueling and [3] Mynatt E., Clark J., Hager G., Lopresti D., electrical supply networks, is necessary for the safe, Morrisett G., Narhstedt K., Pappas G., Patel S., secure operation of road, rail, civil aviation, inland Rexford J., Wright H., & Zorn B. (2017). A National Research Agenda for Intelligent waterways and shipping and is crucial to the European Union’s (EU) economic Infrastructure, http://cra.org/ccc/resources/ccc- growth and social led-whitepapers development [9]. In order to address current socio- economic challenges within an ever-changing [4] Intelligent Research Infrastructure for Shipping, complex and competitive environment, the transport Supply Chain, Transport and Logistics (EN.I.R.I.S.S.T.) (2021), Available at: sector requires new technological developments. https://www.enirisst.gr This will be achieved through research and innovation (R&I), which allows new quality [5] Dimitriou D., Sartzetaki M. (2020) Assessment standards in relation to the mobility of people and framework to develop and manage regional goods. intermodal transport network, Int. Journal: Research in Transportation Business & Moreover, Research Infrastructures play a key role in Management, DOI: the advancement of knowledge and technology and https://doi.org/10.1016/j.rtbm.2020.100455 provide an important link in the innovation chain. [6] Dimitriou D., Κarlaftis, M., Stathopoulos, A. The purpose of a RI is to provide metadata to support (2006). Public Transport Sampling for Special transport sector business ecosystem (operators, Events: A venue based approach, Annual authorities and organizations) to observe and Transportation Research Board (TRB), monitoring up to day performance, assess internal Washington D.C., USA, Proceedings of and external risks and support decisions in planning, Transportation Research Board management and financing. The analysis presented in 124 YRS 2021 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou 15. -17. September DEVELOPING AN INTELLIGENT SERVICES SYSTEM IN TRANSPORT SECTOR Portoroz, Slovenia [7] Sartzetaki M., Dimitriou.D., Karagkouni A. [9] Gkoumas, K., van Balen, M., Ortega Hortelano, (2019), Optimum allocation of time resources for A., Tsakalidis, A., Grosso, M., Haq, G., Pekár, F. transport operation enterprises, International (2019) Research and innovation in transport Journal of Business Research and Management infrastructure - An assessment based on the (IJBRM), Volume 9, Issue 1, pp.29-33. Transport Research and Innovation Monitoring and Information System (TRIMIS), EUR 29829 [8] European Strategy Forum on Research EN, Publications Office of the European Union, Infrastructures (ESFRI), Long-Term ISBN 978-92-76-09539-2 Sustainability of Research Infrastructures, 2017, Vol. 2, pp. 10-12. 125 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Aristi Karagkouni, PhD Candidate Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece arkaragk@econ.duth.gr Maria Sartzetaki, Senior researcher/Adjusted lecturer Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece University Neapolis, Pafos, Cyprus msartze@econ.duth.gr Dimitrios Dimitriou, Professor Associate Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece ddimitri@econ.duth.gr ABSTRACT Investing in transport infrastructure projects is a major concern in terms of the planning process and strategic analysis in order to achieve business resilience and sustainable development. Decision making implies making choices, specifically in the case of transportation infrastructures related to policy making for budget allocations and choices between alternatives for a new transport infrastructure. Governments and decision makers promote investments in transportation infrastructure projects in order to achieve socioeconomic and sustainability goals. Arguments for significantly boosting investment, in capital intensive transport infrastructures, in order to achieve sustained growth rest on high returns on investment in capital-scarce environments and the pressing deficiencies in these areas. One of the most critical issues for decision makers is to select which public investment projects will be funded. The proposed methodology provides an integrated decision support framework based on a combination of ex-ante methodologies for evaluating the impact of the investments on business ecosystem resilience and sustainability. The main content of this paper deals with a dedicated literature review on the subject, promoting the key challenges for planners, managers, and decision makers in the sector of transport. By a system-of-system approach, the role of sustainability in air transport sector ecosystem is depicted, providing evidence that its linkage with business performance is a major challenge for planners, managers and decision makers towards business resiliency and competition. Keywords: transport infrastructure investment, transport development, climate change, business resiliency, sustainability 1. INTRODUCTION Businesses are under increasing pressure from multiple stakeholders (e.g. shareholders, customers, Transport is a major contributor to economic growth employees, society) to manage their positive and that requires operational productive and efficient negative impacts with clear responsibility and infrastructures and services. The necessary condition, strategic intent. Hence, the transformation to that is, ensuring economic, social and environmental sustainable transport requires a redirection, rather sustainability, also affects the challenges and than any substantial increase, in infrastructure opportunities facing the built environment and, expenditure [2]. The need for new approaches in especially, the functioning of critical organizations transport planning, management and policy and businesses [1]. Global sustainability challenges development, new technology, and the global resolve are shaping the way business operates in the 21st toward achieving the 2030 Agenda for Sustainable century. Development and the Paris Climate Agreement is 126 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia crucial for the operation of modern businesses [3,4]. infrastructure requires a considerable financial Moreover, the analysis of the environmental outlay. The Commission estimates that the total implications impacts is essential for the transport investment needs in this area are about €130 billion industry operation and economic environment. per year, with further significant investment needed Sustainable transport variables are defined as a for maintenance. The successful coordinated development process that incorporates the basic deployment of intelligent transport management principles of sustainable development. These systems, which is currently ongoing, is vital to the processes should comply with the objectives of achievement of a pan-European, co-modal and truly environmental, social impact and economic growth integrated transport system. Moreover, automation, [5]. In this context, the adaptation to Climate Change digitalisation and shared mobility are rapidly implications in critical infrastructures is crucial for expanding trends that have the potential to make ensuring business resiliency and sustainability. transport systems more efficient. However, new technologies and mobility patterns also bring The main aim of this paper is to provide a decision challenges relating to the suitability of the legislative support framework towards investments on business framework, privacy protection, safety, liability and ecosystem resilience and sustainability. The data security. methodology is focused on an extended break down of the effects of climate change to critical transport Passenger and freight transport volumes in the EU infrastructure based on the estimation of conditional have been steadily rising in recent decades, from probabilities and the severity in terms of cost to adapt 5,335 billion passenger-kilometres in 1995 to 6,802 or mitigate climate change impact at airports and the billion in 2016 for passenger transport and from level of criticality in terms of time to implement 2,846 billion tonne-kilometres in 1995 to 3,661 policies and measures to maintain growing air billion in 2016 for freight [8]. The Commission transport demand. The numerical application provide estimates that they will continue to grow, albeit at a results for Athens International Airport, the largest slower pace than in the past. It projects an increase of and busiest airport in Greece. 42% for passenger transport activity and 60% for inland freight between 2010 and 2050. The projected 2. TRANSPORT CONTRIBUTION TO rise, for international maritime transport is still ECONOMIC DEVELOPMENT greater, at 71% over the same period. The transport sector is an important component of the Efficient transport services and infrastructure are economy and a common tool used for development necessary to utilize the economic strengths of all EU because of its intensive use of infrastructures [6,7]. regions, to support the internal market and growth This is even more so in a global economy where and to promote economic, territorial and social economic opportunities have been increasingly cohesion. Given its central role, transport also has related to the mobility of people and freight, close ties to policy areas such as the environment, including information and communication jobs and growth, competition, social policies and technologies. Governments in different nations have digitalisation. The adaptation of infrastructure to new different attitudes and priorities regarding how much mobility patterns, as well as the deployment of new attention they pay to these issues. In a mixed infrastructure for clean, alternative fuels, pose economy, one of the ways of measuring the success additional challenges that necessitate new investment of government’s influence on the economy is by and a rethinking of network and business model comparing how it has affected logistics and transport design. Charging infrastructure is essential to growth. The external dimension of transport leads to facilitate the development of electro- mobility, in opportunities and benefits of economic and social particular the uptake of electric vehicles. In 2013, the influence throughout the economy [1]. EU launched its clean fuels strategy, which is intended to ensure a network of alternative fuel Transport is also a strategic sector of the EU stations (including electric charging points) with economy, with transport services accounting for standardised design and use about 5% of the EU’s gross value added and 5.2% (or around 11 million persons) of all jobs in 2016. It 3. SUSTAINABLE TRANSPORT directly affects the everyday lives of all EU citizens and ensures the flow of goods to consumers from Transport drives development, linking people, more than 11 million EU producers and connecting local communities to the world, building manufacturers [8]. This emphasizes the importance markets and facilitating trade [2]. In turn, sustainable of having a well performed transport system for transport can drive sustainable development. economic european integration. Well thought-out, Sustainable transport is the provision of services and sustainable and fully interconnected transport infrastructure for the mobility of people and goods - networks are a necessary condition for the advancing economic and social development to completion and correct functioning of the European benefit today’s and, mostly, future generations which single market. Developing the EU’s transport will be able to live and produce sustainably - in a 127 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia manner that is safe, affordable, accessible, efficient, term for the companies, the governments and the and resilient, while minimizing carbon and other world at large [7]. Resilience planning is an important emissions and environmental impacts [2]. However, feature of sustainable transport development, all goals can not be reached simultaneously. Critical ensuring that passenger and freight transport is the role of priorities given by the industry regarding networks including infrastructure, service and the achievement of sustainability. For istance, it is operations are able to adapt well to climate change necessary to maintain a balance between the related events and other chronic stresses, such as high measures taken to reduce emissions from air transport unemployment and endemic violence and acute and to serve the ever-increasing demand of the sector. shocks, such as earthquakes and terrorist attacks. The challenges are great, but so are the opportunities, Moreover, resilience planning could lead to more in developed and developing countries alike, for effective adaptation of the transport sector to visionary decisions now and in the coming years in unexpected events which can cause llong-term the realm of transport that will set cities and nations movement disruptions, like the Covid-19 pandemic on a sustainable development path. This path will be which has a dramatic impact, especially on the shaped by ambitious goals, targets and indicators. aviation sector [10]. Progress will need to be monitored and evaluated, with course corrections where necessary. The 2030 4. SUSTAINABLE TRANSPORT AND Agenda for Sustainable Development charts this kind CLIMATE CHANGE of path to sustainable development more generally, Transport accounts for approximately a quarter of all and the guideposts are the 17 Sustainable greenhouse gases (GHG) emissions in the EU. As Development Goals (SDGs). Accomplishing the transport emissions started increasing again since SDGs will rely on advances in sustainable transport. 2014, the sector is becoming one of the main Some SDGs are directly and indirectly connected to challenges to the EU’s overall decarbonisation goals. sustainable transport through targets and indicators, In October 2014, the EU leaders adopted the 2030 as illustrated in the figure below. climate and energy framework, including a target of at least a 40% reduction in greenhouse gas (GHG) emissions by 2030 (relative to 1990). In 2015, the EU and all 28 Member States signed the Paris Agreement. Under this agreement, they were requested to submit long-term plans by 2020 showing the efforts by each country to reduce national emissions and adapt to the impacts of climate change. According to the European Environment Agency, transport emissions rose between 1990 and 2007 and then fell until 2014. In 2015 and 2016 they rose again. This means that the sector has become one of the Source: United Nations Economic Commission for Europe, 2017 main challenges to the EU’s overall decarbonisation [3] goals. Under current policies, account being taken of Figure 1: Linkage between transport sector and the the expected growth in freight and passenger United Nations’ Sustainable Development Goals transport, by 2050 GHG emissions from transport are projected to decrease by 15% relative to 2005. Effective transport planning combines the need for However, emissions would still be 10% higher in short-term deliverables with a long-term strategic 2050 than in 1990, owing to the fast rise in transport view, incorporating the social, economic and emissions during the 1990s. Achievement of the environmental aspects of transport and of emissions reduction targets will require a development more broadly [8]. While consideration fundamental shift towards using less energy, and of both short- and long-term needs is common sense cleaner energy, as well as the more efficient use of for all planning, it is particularly relevant for transport infrastructure. transport because of its multi-faceted nature and the resource-intensive, locked-in quality of many Mega infrastructure projects are typically designed transport infrastructure and systems decisions. The for a long-term horizon, requiring sufficient business case for sustainable development is related infrastructure provisions to be included at the to effective and long-term planning. For example, beginning of the project to ensure its sustainability in investing in green energy, or resilient infrastructure, the long run [12]. With these preventive measures to for instance could be proved efficient and save protect infrastructures, airports are expected to money for companies over the longterm [9]. For remain resilient against future climate changes and private sector companies, and even, as noted above, adverse weather conditions well into the future. The for local and national governments, competition is a linkage between the key pillars which ensures the powerful driving force, and competing to become resilient and sustainable business performance is more sustainable will pay dividends over the long- shown in Figure 2. The linkage between 128 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia environmental awareness and the business ecosystem caused by aircraft noise, local air quality and even the is crucial and the appropriate measures to ensure the location of sensitive habitats in surrounding areas can balance between these two pillars could contribute to give rise to operational and infrastructure constraints. the efficient business function. In this context, the Over the past decade, the issue of climate change has adaptation to climate change implications in critical emerged as another, perhaps even more significant infrastructures is crucial for ensuring business environment threat to the growth of the industry, at resiliency and sustainability. both a local and a global level. This has far reaching implications for the future development of aviation in Europe. While the consequences of aircraft emissions for climate change have received much attention in the past, there has until very recently, been little focus on the effects of climate change itself or regulatory and institutional responses to it, both of which present a challenge to the future growth and development of aviation. Furthermore, regarding air transport, and as temperatures are expected to increase in many territories around the globe, airports will need to increase the length of their runways to allow airplanes to take-off without incidents. Due to the increased temperatures, the air becomes less thick and as a result, there is not enough thrust for the Source:Dimitriou et al., 2020 [13] airplanes to leave the ground, so a larger runway is Figure 2: Depiction of resilient and sustainable required. Increased temperatures affect northern business performance countries as well, in which many airports are 4.1. Effect of climate change on Air constructed on ground that consists of ice or snow. These airports are in many cases just for military Transport Infrastructures purposes or only for cargo airplanes. Nonetheless, Climate change related disasters and extreme weather these airports will have to be relocated, because events are expected to significantly increase the risk thawing will undoubtedly have an effect on the of damages on networks, systems and human assets. runway materials, affecting safety on landing and In view of these anticipated adverse effects, growing takeoffs. However, it must be pointed out that attention is placed on adaptation measures, in the possible relocation of airport infrastructure due to form of preventive actions aiming to minimize thawing inflicted materials depends on the correct induced hazards’ negative impacts and to enhance identification of the endangered areas and acts as cross-sectorial resilience. Transportation, as a key prerequisite. economic sector of today’s society, is no stranger to As climate changes, precipitations will become more this regime. The expected climate-related changes and more frequent and storms will intensify, leading will impact both transportation infrastructures and to oftener flooded runways. Though, more advanced networks of operations, independent of transportation draining systems can be applied, another approach means, [14]. The daily functioning of transport can be adopted. Porous asphalt can reduce flooding systems is sensitive to fluctuations in precipitation, effects, as draining becomes easier. However, the temperature, winds, and visibility (and for coastal mixture will have to be reinforced, because of the cities, rising sea levels with the associated risks of high loads with which airplanes strain the landing flooding and damages). surface. Additionally, hurricanes will also become Aviation growth provides significant social and stronger, resulting in takeoffs that are more difficult economic benefits. Many city regions across Europe, and landing procedures due to strong winds. As a especially those in geographically remote locations, result, windbreakers are inevitable if safety is to are highly reliant upon the air transport sector. remain in a high level. In cases where windbreakers Aviation does, however, have significant are not an option, relocation of the runways must be environmental costs, as is accountable for 3.6% of the undertaken. After a complete research on the total EU28 greenhouse gas emissions and for 13.4% direction and intensity of the occurring winds, of the emissions from transport [15]. The adverse runways must be placed in such a way as to avoid environmental impacts associated with the growth side-winds that may affect the direction of and operations of the industry have the poten-tial to approaching aircrafts [16]. ‘feed back’ and restrict aviation’s ability to respond Airports are often classed as nationally critical to demand and contribute to sustainable infrastructure as they facilitate both mobility and development. At an airport level, the disturbance economic growth. However, due to their fixed 129 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia infrastructure and vulnerability to disruptive weather, operations to become more resilient to the changing they are particularly at risk from the potential climate. consequences of climate change, with impacts such An airport is a business, multimodal transport as sea level rise, higher temperatures and greater interchange, employment node, and essential piece of weather extremes creating both an operational and regional and national infrastructure for the business risk. Therefore, to protect vital communities it serves. As an essential service infrastructure and ensure future service continuity for provider to a wide range of stakeholders and users, airport operations, it is necessary to develop the airport infrastructure and operations must have resilience to such risks [17]. high levels of availability, reliability and resilience There has been broad scientific consensus for several [17]. Vulnerabilities to ongoing services from short- years that climate change will cause impacts such as and long-term projected climate changes must be higher temperatures, sea-level rise and greater identified as part of a responsible business continuity weather extremes [18]. However, there is now plan. growing realization that this will require all sectors of Compiling potential impacts and consequences of society to take action to adapt and develop resilience extreme weather events on all aspects of the airport to such impacts [19]. Due to their fixed infrastructure business and operation can enable airports to and vulnerability to disruptive weather, airports are prioritize and better respond to these risks. This may particularly at risk from the potential consequences be based on the airport’s exposure to changing of climate change, with impacts such as sea level rise, climate conditions, its sensitivity to adverse impact, higher temperatures and greater weather extremes or the adaptive capacity when faced with such creating both an operational and business risk challenges. Those impacts ranked the highest priority [20,21]. Overall, the impacts of the climate key or of prime concern should be addressed first through factors on air transport infrastructure are presented in detailed investigation of mitigation options and Figure 3. assessment of cost-benefits as part of an airport’s resiliency plan. The work on climate-change adaptation and resilience should include operational considerations on safety and security, and legal, environmental, financial and business effects on airport operations. Only comprehensive climate-change risk- management strategies will ensure the continuity of operation, profitability and asset value. Some airports are already witnessing gradual change in the investment environment, whereby investors are keen to evaluate climate-change-related risks and opportunities in accordance with the framework Source: Dimitriou et al., 2019 [5] recommended by the Task Force on Climate-related Figure 3: Impacts of climate key factors on airports Financial Disclosures (TFCFD) [21]. 4.2. Adaptation plan of climate change Another important element of an adaptation plan is the coordination with broader airport stakeholders impacts on airports and surrounding communities. An inclusive, Airports are often classed as nationally critical systematic approach to collect intelligence, assess infrastructure as they facilitate both mobility and risks, and interact proactively with these stakeholders economic growth. Therefore, to protect vital will help mitigate long-term financial, economic and infrastructure and ensure future service continuity for operational impacts. Furthermore, as a networked airport operations, it is necessary to develop infrastructure, disruptions in one airport may have a resilience to such risks [5]. More extreme weather- cascading impact on other airports, the wider and climate-related events are expected as the climate economy, and even national resilience. continues to change. The frequency, intensity, spatial extent, duration and timing of events are expected to 5. METHODOLOGY FRAMEWORK increase while slow-onset incremental changes may 5.1. Criticality Analysis lead to fundamental transformation of the socio- economic system. Many airports may remain The analysis of the environmental implications vulnerable to these events as the risks of flooding, impacts the key parameters of transport industry, flight disruptions and cancellations become more analyzed analytically in previous section. The likely. Airports need to understand the risks and transport industry parameters can be affected initiate adaptation measures for both existing and directly, indirectly or through induced effects. new infrastructure, as well as managing critical Climate variations in frost, snow, fog, wind, rain, sea 130 Aristi Karagkouni, Maria Sartzetaki, Dimitrios Dimitriou YRS 2021 PRIORITIZATION OF THE INVESTMENTS IN TRANSPORT INFRASTRUCTURE 15. -17. September TOWARDS BUSINESS RESILIENCY AND SUSTAINABILITY Portoroz, Slovenia water level and heat are key climate factors impact The criticality number is calculated as follows: each transport system business as usual scenario [22]. Therefore, the assessment of the sensibility of each Cr = P * B * S (1) transport option to these climate factors is essential to evaluate the importance for mitigation actions and Where: Cr = the criticality number used to rank the adaptation changes on one hand; and define strategy identified deviations in sensibility towards sustainable development of transport Subsequently, a sensibility scale for variable Cr is infrastructure [5]. adopted, selected from 1.0 to 5.0, where 1.0 is the In this research, the criticality number used to rank impede and 5.0 is the impossible level of business as the identified deviations in sensibility. It cannot be usual scenario. The aim is to determine the average used as a risk measure and it is a product of three severity rate for each weather factor [5]. The risk rough estimates. Before a criticality analysis can be levels are coloured classified into four different performed guidelines have to be developed on how to levels, as shown in Table 4 [23]. determine P, B and S. There are no generally Table 3: Mishap Index/ Criticality Matrix accepted criteria for criticality applicable to a system. Criticality Table 1: Criticality probability factors and severity Hazard/ (I) (II) (III) (IV) Criticality probability factors and severity Mishap High Critical Marginal Negligible Probability Conditional Severity Category P Probability S 3,5 1 000 mcd×m-2×lx-1) were used, which were temporarily applied. Because the selected road marking tapes have, especially when they are new, relatively high RL values, certain peaks (high RL values) should be visible during the evaluation of the raw data for the purpose of quality assurance. Ideally, single raw data could be verified by using the peak values. In addition, the defined measurement matrix is used for static measurements to quantify Source: Federal Highway Research Institute (BASt) Bergisch differences between static and dynamic devices and Gladbach, Germany to establish their correlation. Figure 7: Contrast – road marking / road surface At this time, the evaluation of the measurement data 3.4. Potential of dynamic measuring related to standards and quality characteristics, such systems as accuracy (systematic errors), precision (random errors), uncertainty and repeatability, has not been The replacement of road markings, for example in the completed due to pandemic-related delays. USA, is usually carried out after a fixed time Nevertheless, the first investigation approach will be schedule depending on the traffic volume and the presented here as an example for a measuring section. type of road markings. This can lead to the renewal Independent of the investigation and evaluation of of road markings that still meet the minimum possible influencing factors, firstly, the measuring requirements [12]. In this case, a complete results of the three dynamic measuring devices were measurement and network-wide assessment of the compared, so that any deviation between each condition of road markings is not only important measuring device can be quantified. Error! against the background of road safety, but can also Reference source not found. Table 3 gives an contribute to the further development of a systematic exemplary summary of the averaged measured values maintenance management. For this purpose, analyses of the night visibility of each single measuring device of the performance of road markings and life cycle in 4 diferrent measuring runs within a measuring studies can be carried out on the basis of network- section. wide night visibility values, in order to be able to carry out a targeted renewal of road markings. The Table 3: Dynamic Measuring – comparison of the efficient use of dynamic measuring systems to record night visibility of a measuring section night visibility could help road operators to optimise Static reference Night visibility [mcd×m- Dynamic costs and guarantee an efficient infrastructure. value 2×lx-1] / measuring run Measuring [mcd×m-2×lx- Furthermore, a targeted use of resources is also device 1 2 3 4 1] desirable from an ecological point of view. 1 345 344 340 343 280 2 276 271 276 278 4. FIELD TEST – INVESTIGATION OF 3 299 299 298 297 DYNAMIC MEASURING DEVICES Source: Federal Highway Research Institute (BASt) Bergisch Gladbach, Germany In order to investigate the influencing factors, The measuring values show that in this case study, determine possible use restrictions and evaluate the the difference between individual measuring devices performance, field tests were carried out on the basis can reach as high as 20 %. Figure 8 visualises the of previously developed measuring scenarios with measurement results. the aim of creating a foundation for the usability of 154 Christian Mathea YRS 2021 DYNAMIC MEASUREMENT SYSTEMS – USE OF FAST DRIVING MEASURING VEHICLES 15. -17. September FOR THE NETWORK-WIDE RECORDING OF NIGHT VISIBILITY OF ROAD MARKINGS Portoroz, Slovenia IN DRY CONDITIONS REFERENCES [1] Statistisches Bundesamt (2018). Unfallentwicklung auf deutschen Straßen. Wiesbaden, Germany [2] Bundesministerium für Verkehr und digitale Infrastruktur (2020). Straßenverkehrssicherheit – Alles tun für #DeinLeben. Berlin, Germany Source: Federal Highway Research Institute (BASt) Bergisch Gladbach, Germany [3] European Committee for Standardisation (2018). The European standard EN 1436: Road marking Figure 8: Dynamic Measuring – comparison of the materials – Road marking performance for road night visibility of a measuring section users and test methods Further detailed evaluations, results and knowledge [4] Forschungsgesellschaft für Straßen- und of the field tests will be taken up in the further Verkehrswesen (2013). Zusätzliche Technische presentation. Vertragsbedingungen und Richtlinien für Markierungen auf Straßen (ZTV M). Köln, 5. CONCLUSION AND AN OUTLOOK Germany [5] Forschungsgesellschaft für Straßen- und The use of dynamic measuring devices to detect the Verkehrswesen (1980, 1993 und 2020) condition of road markings offers significant Richtlinien für die Markierung von Straßen (Teil potential in terms of road safety. Furthermore, 1, Teil 2 und Teil A). Köln, Germany measurement data from dynamic measurement [6] Drewes, John & Meseberg (2015). Handbuch systems can contribute to the systematic maintenance und Kommentar für Markierungen auf Straßen. management and cost savings. Measurement systems Bonn, Germany with a diverse range of technology are available on [7] Forschungsgesellschaft für Straßen- und the market which are specially used to measure the Verkehrswesen (2006). Technische night visibility of road markings. The prerequisite for Lieferbedingungen für Markierungsmaterialien developing potential is that, on the one hand, the (TL M). Köln, Germany night visibility of road markings can be reliably [8] Dr. Meseberg, H. (2009). Prüfzeugnis Nr. 0913- recorded by verified measuring devices, and on the 2009-05 über die Eignung des dynamischen other hand, the measurement data as a basis of road Retroreflektometers ZDR 6020 zur dynamischen marking condition assessment can be correctly Messung des Leuchtdichtekoeffizienten bei Retroreflexion R located and uniformly processed. L von Fahrbahnmarkierungen. Berlin, Germany On the basis of the knowledge acquired in the [9] Dr. Meseberg, H. (2020). Prüfzeugnis Nr. 0913- practical investigations, recommendations for the use 2020-02 über die Eignung des dynamischen of dynamic measuring devices are to be developed in Retroreflektometers RetroTek-D Mobile the further process of research, which, in addition to Pavement Retroreflectometer zur dynamischen defined requirements for the recording, such as the Messung des Leuchtdichtekoeffizienten bei Retroreflexion R length of aggregated individual value measurements L von Fahrbahnmarkierungen. Berlin, Germany (measurement blocks), also contain information about suitable measurement periods and [10] Goubert, L., & Lundkvist, S-O (2011). Draft Report of the first round robin test for mobile measurement intervals. A desirable goal for the reflectometers. Belgium future is also the dynamic detection of other traffic- [11] Lee, T. (2017). Advanced Methods for Mobile related properties of road markings, such as skid Retroreflectivity Measurement of Pavement resistance, for which there are no adequate technical Marking, Final Report. New Hampshire, USA: solutions yet. Creating standardised conditions and Transportation Research Board recommendations for the use of dynamic [12] Lee, T. (2011). Advanced Methods for Mobile retroreflectometers, together with the progress of Retroreflectivity Measurement of Pavement sensor technology and findings from research Marking, Final Report for Highway IDEA projects, could promote potential device Project 146. New Hampshire, USA: manufacturers to develop and provide suitable Transportation Research Board measuring instruments. 155 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND THE DESIGN OF PAVEMENTS (MESAS) DATA Mahdi Rahimi Nahoujy, Dr.-Ing. Federal Highway Research Institute Brüderstraße 53, Bergisch Gladbach, Germany rahimi@bast.de ABSTRACT Since 2018, the German Federal Highway Research Institute (BASt) has been using a Multifunctional assessment tool for the structural evaluation and the design of pavements (MESAS) to assess the structural condition of asphalt pavements on network level. MESAS can measure 1,500 km per week with a measuring point each 10 m, where more than 60 values such as bearing capacity, layer thickness, surface picture etc. are collected. This amounts to approx. one terabyte of raw data per week and can thus be called “Big Data”. This results in the following problem: the data has to be evaluated, and quality controlled, but there is no established methodology yet for the analysis, especially for these huge amounts of data. The objective of this study is to develop a new methodological approach for the analysis of the big data collected by MESAS and thus to generate knowledge from it. Data mining is a set of methods applicable to large and complex databases to discover hidden and noticeable (obvious) patterns in the data that were previously unknown. In this study, K-means clustering is used to model the MESAS data. This method is one of the popular unsupervised learning methods, where there is no label (index) for the dataset given. With the help of clustering, the data can be divided into different categories, e.g. “Excellent” or “Good” “Sufficient” and “Failure” or according to other indicators relevant in the assessment of the structural condition of pavements, providing a basis for further analysis and data usage. Keywords: Multifunctional assessment tool for the structural evaluation and the design of pavements (MESAS), NonDestructive Tests (NDTs), Clustering, Machine learning, K-means 1. INTRODUCTION evaluation methods encompassing a wide range of parameters; that is why different research centers are It is essential to continually inspect and evaluate the not able to employ a single comprehensive pavement of all types of roads, e.g. freeways, evaluation system regarding all types of roads. highways, major roads, minor roads, etc., after they are built. The process of pavement evaluation Since 2018, the German Federal Highway Research follows some principles and goes through several Institute (BASt) has been using a Multifunctional stages; in each stage, collected data is recorded. assessment tool for the structural evaluation and the Regular evaluations at appointed times are design of pavements (MESAS) to assess the fundamental in every method of pavement structural condition of asphalt pavements on maintenance; in fact, the data collected from these network level. MESAS can measure 1,500 km per evaluations and inspections enable a pavement week with a measuring point each 10 m, where more maintenance system to make the required plans for than 60 values such as bearing capacity, layer maintenance and rehabilitation. Inspection, similar thickness, surface picture etc. are collected. This to pavement maintenance, is initially done in amounts to approx. one terabyte of raw data per networks, and after the prioritization of projects, the week and can thus be called “Big Data”. This results relevant information and details are accurately in the following problem: the data has to be provided and recorded. There are various pavement evaluated, and quality controlled, but there is no 156 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA established methodology for the analysis, in 2.2. Bearing capacity measurements with particular concerning slope data. For these huge TSD amounts of data, the usage of methods from the field of data mining, e.g. machine learning is needed. The TSD originally comes from Denmark and was developed to measure deflections at high speed In general, there is a wide range of research available (Rasmussen, 2002). The measurement system with about NDTs and also about machine learning with all associated instruments is installed in a truck (see Falling Weight Deflectometer (FWD) data, (Jansen, Figure 1). In this way, the load-bearing capacity can 2009; Chakar, 2011; Wacker, 2020; Rahimi be measured at a speed of approx. 80 km/h. To Nahoujy, 2020). MESAS is a new device, though, ensure equal general conditions when correcting the and not well researched yet. Only a few studies exist individual Doppler laser data, the measuring beam about the device, focusing mainly on its general must be resistant to bending and torsion and, if functionality and evaluating its usage for pavement possible, have the same temperature (FGSV, 2015). structural testing (Muller et al., 2013; Katicha et al., 2020; Jia et al., 2021). Using machine learning with MESAS data is a fairly new approach with hardly any existing materials, but it is expected that more studies will be published soon. In this study for the first time, an approach of unsupervised learning is used to model the slope Source:(FGSV, 2015) data from MESAS. With the help of K-means, the slope data is divided into different clusters, Figure 1: Principle sketch of the TSD providing basic labels for further analysis and data The measuring system consists of a towing vehicle usage. with a single-axle semi-trailer in which the measuring device is mounted. The core of the 2. BACKGROUND measuring system consists of several sensors 2.1. Non-Destructive Tests (NDTs) (Doppler lasers) arranged at defined intervals on a steel beam, which are aligned approximately Non-Destructive Tests (NDTs) are employed at perpendicular to the road surface. The lasers different levels to evaluate, repair, maintain, and continuously record the speed of the short-term rehabilitate various pavements. In a network, NDTs deformation of the traffic surface pavement in the are used for pavements subdivided into sections with direction of the sensor as a result of the load from similar structural capacity. In a project, they are used the trailer axle. The vertical deformation speed to detect the location and reason for destructions in vector is determined from the vectorial relationship flexible pavements. between the measured deformation speed vector (in NDTs evaluating the structural condition use the direction of the sensor), the horizontal speed of deflection measurement mechanisms. The the road in relation to the measuring vehicle deflection of pavement layers due to loading is a (corresponding to the measured vehicle speed) and crucial index in determining the structural the angle to the vertical of the road surface. The performance of the pavement and the strength of Doppler laser Lref (reference laser), which measures different layers. Currently, there are a variety of in the area in front of the measuring system commercial pavement deflection measurement undisturbed by the load application, is used as a devices available. The most popular examples are reference to compensate for the vertical vehicle the Benkelman Beam and the FWD. dynamics and the torsion of the measuring beam. This means that the calculation does not consider the Recently, NDTs have been further developed with actual angle of the Doppler lasers, but the (constant) the Traffic Speed Deflectometer (TSD). The TSD is angular difference to the reference laser. for each a fast-moving measuring system for recording sensor position i. The so-called slope value (slope of measured values that can be used to describe the the deformation at this sensor position) is derived load-bearing capacity of traffic surface pavements from the corrected vertical speed vector Vv,i and the (FGSV, 2015). Since 2018, the German Federal horizontal speed of the measurement vehicle Vh as Highway Research Institute (BASt) has been using follows (Hildebrand et al., 2000; FGSV, 2015): a MESAS to assess the structural condition of asphalt pavements on network level. MESAS’s 𝑆𝑙𝑜𝑝𝑒𝑖 = 𝑆𝑖 = (𝑉𝑣, 𝑖) / 𝑉ℎ (1) basic function is the TSD high-speed load measurement system, and its additional The principle of calculating the slope value is shown measurement equipment such as transversal in figures 2 and 3. evenness and surface image, longitudinal evenness, ground-penetrating radar (GPR) and front cameras. 157 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA Either one has to use deflections to evaluate the pavement with some error, or try to find labels for slopes. In this study a method of cluster analysis for slope data is presented to determine labels for TSD data. K-means as one of the popular unsupervised learning methods is used to train a model at network level for slope data. 3. METHODOLOGY 3.1. The method of machine learning Finding patterns in existing data is only possible for the human brain. But when the volume of data becomes very large and the time required to perform calculations increases, the need for machine learning is introduced as a science that helps people work with massive amounts data in the least amount of Source:(FGSV, 2015) time. Figure 2: Principle sketch for calculating the slope The core of the science of machine learning is to value in a vehicle-related coordinate system (not to design machines that use the examples given to them scale) and learn from their own experiences. In fact, this science tries to create a machine using algorithms in such a way that it can operate without needing plans or prescribed individual actions as input, but that it can learn from itself. In machine learning, instead of programming everything, the data is given to a general algorithm, and it is this algorithm that builds its own logic based on the data given to it. Machine learning has a variety of methods, including supervised, unsupervised, and reinforcement learning. The algorithms used in machine learning fall into these three categories. 3.2. Unsupervised learning Unsupervised learning is a set of machine learning methods for discovering patterns in data. In Source:(FGSV 433 B 5, 2015) supervised learning, the system tries to learn from Figure 3: Arrangement of the Doppler lasers to the previous examples given to it. In unsupervised determine the slope values (not to scale) learning, the data processed with the unsupervised The result files of a TSD measurement contain about algorithm is non-labeled, which means that the input 60 different values that are indirectly and directly variable (X) is specified without any corresponding related to the road condition or the measurement output variable. In unsupervised learning, itself. Among them, in the case of BASt's TSD, algorithms are used to discover interesting pattern in there are ten different slopes (in distances: -450 the data. Therefore, if the dataset is labeled, it is (mm), -300 (mm), -200 (mm), 110 (mm), 210 (mm), considered a supervised problem, and if the dataset 310 (mm), 460 (mm), 610 (mm), 910 (mm), 1510 is unlabeled, the problem is unsupervised. Figure 4 (mm)) and usually at least nine different deflections shows an example of supervised and unsupervised (in distances: D0 (mm), D200 (mm), D300 (mm), learning. The image on the left is an example of D450 (mm), D600 (mm), D750 (mm), D900 (mm), supervised learning, in which the regression method D1200 (mm), D1500 (mm)). is used to find the best fit between the features. In unsupervised learning as on the right, inputs are 2.3. Analysis of slope values separated based on characteristics and predictions As mentioned, 10 slopes are measured by the TSD are based on which cluster they belong to. in each measurement. Since there does not exist any indicator yet to classify the different slopes, the deflection values have to be calculated from them in order to assess the quality of the pavement. But these calculations have error due to curve fitting methods. 158 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA Lloyd. This version of the clustering algorithm is now standard on most computer software that performs K-means clustering. 3.5. Lloyd algorithm Suppose that the observations (x1, x2, ..., xn), which have a dimension D, must be divided into k clusters. These clusters are identified as S = {S1, S2, ..., Sk}. The cluster members should be selected from observations that minimize the “within-cluster sum of squares” (WCSS) function. Therefore, the objective function in this algorithm is described as follows (Arthur et al.,2006): Figure 4: Supervised and unsupervised learning 𝑘 "Clustering" is the most important method in 𝑎𝑟𝑔𝑆 𝑚𝑖𝑛 ∑ ∑‖𝑋 − 𝜇𝑖 ‖2 unsupervised learning. 𝑖=1×∈𝑆𝑖 𝑘 3.3. Clustering = 𝑎𝑟𝑔𝑆 𝑚𝑖𝑛 ∑|𝑆𝑖| 𝑉𝑎𝑟 𝑆𝑖 𝑖=1 Cluster analysis, or clustering for short, is the process by which a set of objects can be divided into (2) individual groups. Each division is called a cluster. The members of each cluster are very similar in their where: characteristics, while the degree of similarity between the clusters is minimal. In such a case, the µi: the center of clusters Si purpose of clustering is to assign "labels" to the | Si |: the number of members in cluster i. objects, indicating that each object is a member of the cluster. 3.6. Training a K-means model Although most algorithms or clustering methods In this study, the K-means algorithm uses an have the same basis, there are differences in how iterative refinement technique that works in the similarities or distances are measured, as well as the following way (Arthur et al., 2006): choice of labels for the objects in each cluster. The 1. First, K centers (mean of the cluster) are main groups for clustering methods are: partitioning quantified randomly, which represent the methods, hierarchical clustering, density-based clusters: 𝜇0, 𝜇0, … , 𝜇0 clustering, model-based clustering and “fuzzy” 1 2 𝑘 2. Each observation is assigned to the cluster clustering. with the closest mean value. In this work, 3.4. K-means the “squared Euclidean distance” is used as the K-means algorithm: The K-means algorithm is one of the most popular 2 𝑆𝑡 = {𝑥 𝑡 ‖2 ≤ ‖𝑥 𝑡 ‖ ∀𝑗, 1 ≤ partitioning methods used in machine learning, 𝑖 𝑝: ‖𝑥𝑝 − 𝜇𝑖 𝑝 − 𝜇𝑗 especially in the field of unsupervised learning. In 𝑗 ≤ 𝑘} K-means clustering, the optimization of an objective (3) function is used. The clustering results in this 3. the mean values for observations method can be performed using minimization or associated with each center are maximization of the objective function. If the recalculated: criterion is “distance measure” between objects, the 1 𝑚𝑡+1 ∑ 𝑖 = 𝑡 𝑥 𝑡 𝑥 𝑗 objective function is based on minimization. The |𝑆 𝑗∈𝑆 𝑖 | 𝑖 answer to the clustering operation is to find clusters (4) where the distance between objects in each cluster is Then, these two steps (2 and 3) are minimal. Conversely, when the dissimilarity performed alternately several times so that function is used to measure the similarity of objects, the means are stable enough or the sum of the objective function is chosen so that the clustering the variances of the clusters does not results maximize its value in each cluster. change much. 4. Finally, the following mean values of the The K-means clustering method was invented by last phase (in time T) will represent the sociologist and mathematician McQueen in 1965 clusters: 𝜇𝑇 𝑇 𝑇 1 , 𝜇2 , … , 𝜇𝑘 and developed and optimized by other scientists. In 1957, another version of this algorithm, the standard K-means clustering algorithm, was developed by 159 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA 4. RESULTS AND DISCUSSION centers. This model results in three boundaries (labels): 4.1. Results  Label 1: ca. 28.15 µm/m In this study, data measured with the TSD of the  Label 2: ca. 64.10 µm/m Federal Highway Research Institute (BASt) on a  Label 3: ca. 133.10 µm/m highway in North Rhine-Westphalia was used. The measurements were processed every 10 m on more By applying these labels, all data is divided into four than 52 km of highway with 3 strips in both groups (clusters). The data of each cluster have a directions, so that after preprocessing 41,618 data special relationship to each other. Based on the sets for slope 110 (which is the slope value at a existing background of experience in connection distance of 110 mm from the axle load) are available with bearing capacity measurements, the TSD data (Figure 5). (here: slope 110 values) is then categorized as follows:  Excellent: 0 – 28.15 µm/m  Good: 28.15– 64.10 µm/m  Sufficient: 64.10– 133.10 µm/m  Failure: more than 133.10 µm/m The next model is trained for slope 310. In this route there are 40,787 data sets after preprocessing for slope 310 available (Figure 7). The slope 310 data is clustered into four clusters as Figure 8. The center values in this model are positioned as follows: Figure 5: slope 110 data after preprocessing  Center 1: 12.3095985112341 µm/m  Center 2: 33.9514330427490 µm/m First, the slope 110 values are divided into four  Center 3: 62.5570365505194 µm/m clusters by K-means (one-dimensional clusters).  Since the results of clustering could be different in Center 4: 106.351718880242 µm/m each running, some models were trained and their deviations of results were analyzed. The deviation of averages and limits of the models were very low (less than 1%), showing that the clusters are robust. Four clusters are shown in the following Figure 6. Figure 7: slope 310 data after preprocessing Figure 6: Cluster diagram for slope 110 The center values in this model are positioned as follows:  Center 1: 14.3641437039793 µm/m  Center 2: 41.9321957198163 µm/m  Center 3: 86.2550691677927 µm/m  Center 4: 179.975589050958 µm/m Figure 8: Cluster diagram for slope 310 All points in each cluster have the smallest distance This model results in three boundaries (labels): from the respective center, compared to the other  Label 1: ca. 23.12 µm/m 160 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA  Label 2: ca. 48.25 µm/m modeled by using K-means as an unsupervised  Label 3: ca. 84.46 µm/m learning method. TSD slopes (110 and 310) were modeled in one-dimensional clusters. For each Using these labels, all data is clustered into four model, four different clusters were defined with groups. The slope 310 data is then categorized as certain boundaries as labels (excellent, good, follows: sufficient, failure). The deviation of averages and  Excellent: 0 – 23.12 µm/m boundaries of the models were very low at less than  Good: 23.12– 48.25 µm/m 1%.  Sufficient: 48.25– 84.46 µm/m The clusters thus determined can be used to classify  Failure: more than 84.46 µm/m new data sets. That means, for new datasets, the labels function as indicators to categorize the 4.2. Discussion different slope values, enhancing the assessment of In this study two different indicators are presented. the quality of pavement. For slope 110 and for slope 310, both as one- When using purely data-based evaluations and dimensional clusters, the data is divided into four classifications, it is nevertheless recommended to clusters with the labels “Excellent”,“Good”, “Sufficient” and “Failure” check or classify the results using, for example, . mechanical models. Since the results of clustering could be different in each running, some models were trained and their REFERENCES deviations were analyzed. The deviation of averages [1] Arthur, D., & Vassilvitskii, S., 2006. k- and limits of the models were very low (less than means++: The advantages of careful seeding. 1%), showing that the clusters are robust. Stanford. After the clusters have been positioned by labels, [2] Chakar, T. (2011). Methode zur Klassifizierung new measured data can be categorized with these von Tragfähigkeitsmessergebnissen des Falling labels and it can be determined in which cluster they Weight Deflectometers bei are located. Asphaltbefestigungen (Vol. 59). KIT Scientific Publishing. Up to now, there has been no existing indicator for slope data, because road agencies and companies [3] FGSV, 2015. Arbeitspapier- Tragfähigkeit für have been using deflection values, that have to be Verkehrsflächenbefestigungen Teil B 5. Schnell fahrendes Messsystem – Traffic Speed calculated from the raw data, in order to assess the Deflectometer (TSD): Gerätebeschreibung, quality of the pavement. The presented approach of Messdurchführung. FGSV-Verl. Köln, 2015, clustering slope data provides an alternative, that FGSV 433 B 5. does not need the extra-step of calculating deflection values, which may contain errors. Future research [4] Hildebrand, G., Rasmussen, S. & Andrés, R., 2000. Development of a laser-based high speed may focus on determining the quantity of error of the deflectograph. In S. D. Tayabji & E. O. Lukanen calculations of deflections and compare it with the (Eds.), Nondestructive testing of pavements and approach presented in this study. backcalculation of moduli. Proceedings of the At this point, it is explicitly noted that the "Third International Symposium Nondestructive Testing of Pavements and boundaries, categories and labels presented are Backcalculation of Moduli", held Seattle, derived from a purely data-based approach and Washington, on 30 June - 1 July 1999 (ASTM therefore have no general validity and are highly STP, Bd. 1375, S. 457–469). Fredericksberg: dependent on the data basis. American Society for Testing and Materials (ASTM). 5. SUMMARY [5] Jansen, D. (2009). Temperaturkorrektur von mit Non-Destructive Tests (NDTs) are used at different dem Falling-Weight-Deflectometer gemessenen levels to evaluate different pavements in terms of Deflexion auf Asphaltbefestigungen. their condition and to derive maintenance needs. The schriftenreihe des instituts fuer strassenbau und verkehrswesen, (2). Multifunctional assessment tool for the structural evaluation and the design of pavements (MESAS) is [6] Jia, X., Woods, M., Gong, H., Zhu, D., Hu, W., a new device, which can assess the structural & Huang, B. Evaluation of influence of condition of asphalt pavements on network level. pavement data on measurement of deflection on The Traffic Speed Deflectometer (TSD) is the core asphalt surfaced pavements utilizing traffic part of MESAS and records slope values. speed deflection device. Construction and Building Materials, 270, 121842. In this paper, an approach of machine learning in [7] Katicha, S. W., Shrestha, S., Flintsch, G. W., & road construction was presented. TSD data was Diefenderfer, B. K. (2020). Network Level 161 Mahdi Rahimi Nahoujy YRS 2021 THE APPROACH OF K-MEANS CLUSTERING AS A TOOL TO ANALYZE THE 15. -17. September MULTIFUNCTIONAL ASSESSMENT TOOL FOR THE STRUCTURAL EVALUATION AND Portoroz, Slovenia THE DESIGN OF PAVEMENTS (MESAS) DATA Pavement Structural Testing With the Traffic [10] Rasmussen, S., 2002. Development of a high Speed Deflectometer (No. VTRC 21-R4). speed deflectograph (Report 117). Roskilde: Danish Road Institute. [8] Muller, W. B., & Roberts, J. (2013). Revised approach to assessing traffic speed [11] USDOT, F., 2011. Use of nondestructive testing deflectometer data and field validation of in the evaluation of airport pavements. Advisory deflection bowl predictions. International Circular, 150. Journal of Pavement Engineering, 14(4), 388- [12] Wacker, B. (2020). Zeitraffende 402. Belastungsversuche mit integriertem Einsatz [9] Rahimi Nahoujy, M., 2020. An Artificial Neural zerstörungsfreier Messverfahren. Network approach to model and predict asphalt deflections as a complement to experimental measurements by Falling Weight Deflectometer. Doctoral dissertation, Ruhr- Universität Bochum, Germany. 162 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR NETWORKS Tibor Petrov Department of International Research Projects – ERAdiate+, University of Zilina Univerzitna 8215, Zilina, Slovakia tibor.petrov@uniza.sk ABSTRACT Applications based on vehicular communications are becoming an important part of the Intelligent Transportation Systems. They are expected to bring many benefits including increased road safety, reduced environmental impact and decreased socio-economic costs of road transport. Over the years, two dominant communication technologies for vehicular communications have emerged – Dedicated Short-Range Communications (DSRC) and Cellular Vehicle-To-Everything (C-V2X). While many studies investigated the performance of both technologies for specific vehicular applications relying on beaconing, some applications require a network level routing of the communication packets for their operation. In order to assess the routing performance of DSRC and LTE-based C-V2X communication, we prepared a federated traffic-telco simulation study using a grid road network. The vehicles in the simulation run a Privacy Ensuring Emergency Vehicle Approaching Warning application using IEEE WAVE communication standard in the first simulation scenario and LTE-V2V in the second simulation scenario. The unicast performance of both technologies is evaluated in terms of average message end-to-end delay and packet delivery ratio. Keywords: V2V, VANET, IEEE WAVE, LTE-V2V, emergency vehicle, warning, routing, simulation 1. INTRODUCTION or leave the network without any prior allocation of communication resources. Applications based on vehicular communications are becoming an important part of the Intelligent On the other hand, C-V2X based systems exploit the Transportation Systems. They are expected to bring already installed communication infrastructure for many benefits including increased road safety, 4G/LTE, 5G, or other V2X-capable generation of reduced environmental impact and decrease of socio- cellular networks. Communication resources in these economic costs of road transport. While many networks are usually a subject to a centralized communication technologies have been considered network management. C-V2X has been first for intervehicular communications, two of them have introduced in 3rd Generation Partnership Project emerged as dominant over the years - Dedicated (3GPP) Release 14 [4]. 3GPP Release 14 included a Short-Range Communications (DSRC) and Cellular definition of LTE Advanced Pro V2X services, Vehicle-To-Everything (C-V2X). In literature, architectures, radio access network regulations and DSRC communication is often used as an four use case scenarios [5]: overarching term for technologies based on IEEE  Vehicle-to-Vehicle (V2V) communication 802.11 OCB [1] as an Access layer standard of the  Vehicle-to-Infrastructure (V2I) Station Reference Architecture [2,3]. Two major communication families of standards based on IEEE 802.11 OCB  Vehicle-to-Network (V2N) communication include ETSI ITS [2] developed for deployment in  the European Union and IEEE WAVE [3] to be used Vehicle-to-Pedestrian (V2P) communication in North America and other countries. The C-V2X communication can be carried out using Communicating nodes using either of these LTE Uu and PC5 interfaces. LTE Uu interface is used technologies form a Vehicular Ad hoc Network for communication with the cellular base station and (VANET) – a communication network without any the PC5 interface is used for direct communication centralized management where nodes are free to enter between V2X nodes [6,7]. Furthermore, 3GPP 163 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS Release 14 defined two additional C-V2X Modes - 3, technologies come with their own set of strengths and 4 which are used to allocate communication limitations when applied in the vehicular resources to the V2X nodes. While Mode 3 is used to environment. allocate network resources to the V2X nodes with the Bazzi et al. [12] conducted an analytical comparison support of the cellular base station, Mode 4 allows the of IEEE 802.11 OCB and LTE-V2V applied to nodes to allocate and manage their communication vehicle-to-vehicle beaconing. Numerical results resources fully autonomously in a case when the show that LTE-V2V has been able to support the connection to the cellular base station is not available same beacon generation frequency as IEEE 802.11 [8]. OCB but with less dedicated communication With the ever evolving cellular network resources. The authors, however, observed that the infrastructure, the use of C-V2X for vehicular LTE’s advantage quickly reduces with the increasing applications might seem as a no-brainer choice. communication range. However, due to its relative novelty compared to the Cecchini et al. [13] compared the performance of already mature DSRC, some studies advise more both technologies by a simulation study considering cautious approach [9]. Studying the performance of Cooperative Awareness Service in a highway both technologies in the context of a specific scenario. They conclude that LTE-V2V outperforms vehicular application may provide valuable insights IEEE 802.11 OCB in terms of packet reception ratio on their ability to meet the application performance while IEEE 802.11 OCB achieves lower requirements. communication delay with increasing packet length While many studies investigated the performance of and communcaition range. both technologies for specific Cooperative Intelligent Mir et al. [14] conducted another simulation-based Transport System (C-ITS) applications relying on study comparing the performance of both beaconing, some applications require a network level technologies. The results indicate that IEEE 802.11 routing of the communication packets for their OCB performs acceptably in sparse networks with operation. For this reason, it is essential to study the limited vehicle mobility while LTE meets most performance of both technologies in unicast reliability, mobility and scalability application scenarios and determine the conditions under which requirements. On the other hand, the study points out one of the technologies has a performance edge over that satisfying stringent delay requirements by LTE the other. To the best of our knowledge, this issue has is a challenge. not been addressed in the literature yet. While more studies, e.g. [15] tend to support those In order to assess the routing performance of DSRC results as well, overall, there seems to be no and LTE-based C-V2X communication, we consensus in the available literature about the conducted a federated traffic-telco simulation study dominance of either of the communication using a grid road network. The vehicles in the technologies in all target key performance indicators simulation run a Privacy Ensuring Emergency relevant to C-ITS applications. Instead, majority of Vehicle Approaching Warning application [10] using the studies show that one of the technologies tends to IEEE WAVE [3] communication standard in the first overperform the other in certain aspects when simulation scenario and LTE-V2V [4] in the second deployed by a specific C-ITS application. simulation scenario. We evaluate the routing Furthermore, to the best of our knowledge, relevant performance of both technologies in terms of first works consider only broadcasting of the C-ITS message delivery delay, average message end-to-end application messages and the comparison of unicast delay and packet delivery ratio. performance of both technologies is missing in the The rest of this paper is organized as follows. Section available literature. 2 explores the relevant literature and similar efforts undertaken by other authors. Section 3 describes the 3. SIMULATION EXPERIMENT simulation setup, the case study and methodology DESCRIPTION used to obtain the results. In Section 4, simulation The simulation experiments were carried out using results are presented and discussed. The paper is OMNeT++ discrete event network simulator [16] concluded by Section 5. with Veins [17], INET [18] and SimuLTE [19] 2. RELATED WORK simulation frameworks. Simulation of Urban Mobility (SUMO) [20] was used to generate the Some studies, for example by Nguyen et al. [12] traffic flows for the simulation. suggest that LTE provides a significant improvement over DSRC in communication range without As a transport network model, we used a Manhattan suffering in other aspects. The majority of the grid with overall size of 2500x2500 m and 100 m available works, however, suggest that both street length. The topology of the simulation experiment is detailed in Figure 1. 164 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS to-device (D2D) communication. Hence, the vehicles communicate directly but the scheduling and allocation of the network resources used for the transmissions is managed by the eNodeB. Target metrics for evaluation of the communication network performance are:  First message delivery delay – it is a time between creation of the first successfully delivered communication packet at the EV’s communication module and the moment of reception of the same communication packet by the application layer of the destination vehicle’s communication module. The first message delivery delay includes also the time necessary to establish the communication route between the EV and the destination vehicle and is therefore expected to be higher compared to the end-to-end delay of the subsequent communication packets; Figure 1: Topology of the simulated network. The two  Average end-to-end delay – is the average eNodeBs are displayed with their X and Y coordinates time between creation and successful in m. The diagonal route of the vehicles is displayed in reception of all communication packets blue. transmitted between the EV and the The network consists of two LTE eNodeBs. The destination vehicle during the whole overall route length is 4800 m. The intensity of the simulation; vehicle flow is 1200 vehicles/hour. Maximum  Packet delivery ratio – defined as the ratio vehicle speed is set to 50 kph. between packets sent by the EV 𝑁𝑠𝑒𝑛𝑡 and received 𝑁𝑟𝑒𝑐𝑒𝑖𝑣𝑒𝑑 by the destination vehicle We consider three communication scenarios. In the during the whole simulation as described in first communication scenario, the eNodeBs equation: (eNodeB1-B and eNodeB2) are uniformly spaced along the diagonal route of the vehicles, as presented 𝑁 𝑃𝐷𝑅 = 100 𝑠𝑒𝑛𝑡 [%] (1) in the Figure 1 (in black). The distance between the 𝑁𝑟𝑒𝑐𝑒𝑖𝑣𝑒𝑑 sending vehicle and its serving eNodeB1-B 𝑑𝑒𝑁𝐵1is 1200m. In the second considered communication Each destination vehicle transmission has been scenario, the eNodeB1 (eNodeB1-A) is closer to the simulated using three different random seeds for each sending vehicle, as illustrated in the Figure 1 (in of the investigated communication technologies. green). The distance 𝑑 Detailed values of simulation parameters are 𝑒𝑁𝐵1 in this case is 600 m. In the third considered communication scenario, the summarized in Table 1. vehicles communicate directly using DSRC Table 1: Simulation parameters communication. The cellular infrastructure is not DSRC LTE used in this case. Carrier frequency 5900 MHz 2100 MHz Channel bandwidth 10 MHz 10 MHz The performance of the communication network has TX Power 20 dBm eNodeB: 40 dBm been studied in the context of the PEEVA-WS [10] UE: 26 dBm application. This application has been designed to Data rate 6 Mbps 10 Mbps provide warning messages indicating an approaching Packet generation 1 Hz emergency vehicle (EV) while maintaining the frequency privacy of the sensitive EV data. PEEVA-WS is run Communication 100-2800 m distance by an emergency vehicle which appears after the Traffic intensity 1200 vehicles/hour simulation initialization period of 450 seconds. Routing protocol AODV - Immediately after the EV enters the simulation, it starts to transmit communication packets to a selected 4. SIMULATION RESULTS destination vehicle. The distance between the destination vehicle and EV is varied in the range 100- In terms of Packet Delivery Ratio, the results 2800 m. Data packets are exchanged directly between achieved in the three simulated scenarios seem the two vehicles. In the case of LTE technology, the surprising in the context of the available literature. As transmission between the EV and the destination can be seen in the Figure 2, the DSRC technology vehicle is carried out as an eNodeB-assisted device- achieved the highest PDR regardless of the 165 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS communication distance, while LTE in both depends on the network load [21]. Mean end-to-end scenarios experienced a sudden drop in PDR in the delay 𝐿 ̅𝐷𝑆𝑅𝐶 averaged throughout the whole range between 250 and 1400 m. simulation and throughout all destination vehicles is (7 ± 70) 𝑚𝑠. In the case of LTE with 𝑑𝑒𝑁𝐵1 = 600 𝑚, the 𝐿̅600𝑚 𝐿𝑇𝐸 = (68 ± 181) 𝑚𝑠. The worst performing case in terms of end-to-end delay is LTE with 𝑑 1200𝑚 𝑒𝑁𝐵1 = 1200 𝑚, where the 𝐿 ̅𝐿𝑇𝐸 is (2640 ± 5373) 𝑚𝑠. Figure 2: Packet Delivery Ratio. Figure 3 shows the results in terms of the first message delivery delay. The best results were achieved by LTE with 𝑑𝑒𝑁𝐵1 = 600 𝑚. The delay in this case does not differ significantly from the average end-to-end delay. In the case of DSRC Figure 4: Average end-to-end delay. technology, the induced first message delivery delay is higher since it includes the duration of AODV 5. CONCLUSION protocol route discovery. After the communication In this paper, we compared the unicast performance route is established by the AODV protocol, the pf DSRC and LTE-V2V technology in a vehicular subsequent packets are delivered with much lower environment in the context of Privacy Ensuring average end-to-end delay. Emergency Vehicle Approaching Warning application. The simulation study has been carried out using a 25x25 Manhattan grid road network model with 100 m street length. Our results suggest, that in such a scenario, best results in terms of communication delay and packet delivery ratio are achieved by DSRC technology. The LTE-V2V technology has been investigated for two different distances between eNodeB and the transmitting vehicle 𝑑𝑒𝑁𝐵1 – 600 m and 1200 m. In the case of 600 m 𝑑𝑒𝑁𝐵1we observed significantly lower communication delay and higher PDR than in the case of 1200 m 𝑑𝑒𝑁𝐵1. However, in both cases, there is a steep drop of PDR in communication distances between approximately 250 and 1400m. Destination vehicles in those ranges are at the Figure 3: First message delivery delay. boundaries of the transmitting vehicle’s In terms of average end-to-end delay, our communication range or beyond. In such a case, the observations confim the results achieved by [21]. The error rate of the communication increases results presented in Figure 4 demonstrate that once dramatically and the same communication packet is the communication route is established, the DSRC is being retransmitted multiple times until the LTE able to operate with lower end-to-end delay MaxHarqRtx parameter is achieved. After this point, compared to the LTE. One of the contributing factors the direct communication between the vehicles is for that is the fixed delay induced by the LTE impossible. The communication is then switched resource selection algorithm which cannot be from D2D mode to infrastructure-assisted mode, avoided. On the contraty, the resource access delay in where the data packets between the transmitting DSRC is given by the CSMA-CA parameters and 166 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS vehicle and the destination vehicle are relayed by the [6] Qualcomm Technologies. (2019). Introduction to serving eNodeB. Cellular V2X. Available online: https://www.qualcomm.com/media/documents/f The simulation results further indicate that the DSRC iles/introduction-to-c-v2x.pdf (accessed on 8 communication in the presented scenario is very June 2020). reliable. This is caused by the fact that the traffic [7] Wang, J., Shao, Y., Ge, Y., & Yu, R. (2019). A volume in the simulation is high enough for the survey of vehicle to everything (v2x) testing. routing protocol to find the optimal route to the Sensors 2019, 19, 334. destination in each case. Moreover, the communication route is rather stable and since there [8] Molina-Masegosa, R., & Gozalvez, J. (2017). System Level Evaluation of LTE-V2V Mode 4 are no significant connection break-ups, the PDR Communications and Its Distributed Scheduling. stays very high. 2017 IEEE 85th Vehicular Technology The variability of the results in the available literature Conference (VTC Spring) (pp. 1-5). Sydney, indicate that the communciation performance of both NSW, Australia. doi: 10.1109/VTCSpring.2017.8108463. technologies is likely very scenario specific. Therefore, an investigation of the communication [9] Yu, T., Zhang, S., Cao, S., & Xu, S. (2018). performance under different traffic scenarios and Performance Evaluation for LTE-V based traffic volumes could be beneficial. Vehicle-to-Vehicle Platooning Communication. 2018 24th Asia-Pacific Conference on Acknowledgment Communications (APCC) (pp. 15-20). Ningbo, China. doi: 10.1109/APCC.2018.8633551. This work was supported by the Slovak Research and Development Agency under the contract no. SK-IL- [10] Petrov, T., Pocta, P., Roman, J., Buzna, Ľ., & RD-18-005 ICT and smart cars for efficient Dado, M. (2020). A Feasibility Study of Privacy Ensuring Emergency Vehicle Approaching emergency response and traffic management Warning System. Appl. Sci. 2020, 10, 298. (SENECA). https://doi.org/10.3390/app10010298 This publication was realized with support of [11] Nguyen, T.V, Shailesh, P., Sudhir, B., Kapil, G., Operational Program Integrated Infrastructure 2014- Jiang, L., Wu, Z., Malladi, D., & Li, J. (2017). A 2020 of the project: „Value of travel time from an comparison of cellular vehicle-to-everything and individual traveller’s perspective“ (ITMS2014+ dedicated short range communication. 2017 313011T241), co-financed by the European Regional IEEE Vehicular Networking Conference (VNC) Development Fund. (pp. 101-108). Turin, Italy. doi: 10.1109/VNC.2017.8275618. REFERENCES [12] Bazzi, A., Masini, B.M., Zanella, A., & Thibault, I. (2016). Beaconing from connected vehicles: [1] IEEE Std 802.11p-2010. (2010). IEEE Standard IEEE 802.11p vs. LTE-V2V. 2016 IEEE 27th for Information technology Local and Annual International Symposium on Personal, metropolitan area networks Specific Indoor, and Mobile Radio Communications requirements Part 11: Wireless LAN Medium (PIMRC) (pp.1-6). Valencia, Spain. doi: Access Control (MAC) and Physical Layer 10.1109/PIMRC.2016.7794598. (PHY) Specifications Amendment 6: Wireless Access in Vehicular Environments. [13] Cecchini, G., Bazzi, A., Masini, B.M., & Zanella, A. (2017). Performance comparison between [2] ETSI EN 302 665 V1.1.1. (2010). Intelligent IEEE 802.11p and LTE-V2V in-coverage and Transport Systems (ITS); Communications out-of-coverage for cooperative awareness. 2017 Architecture. IEEE Vehicular Networking Conference (VNC) [3] IEEE Guide for Wireless Access in Vehicular (pp.109-114). Turin, Italy. doi: Environments (WAVE) – Architecture. (2014). 10.1109/VNC.2017.8275637. In IEEE Std 1609.0-2013, pp.1-78, doi: [14] Mir, Z.H., & Filali, F. (2014). LTE and IEEE 10.1109/IEEESTD.2014.6755433. 802.11p for vehicular networking: a performance [4] Evolved Universal Terrestrial Radio Access (E- evaluation. J Wireless Com Network 2014, 89. UTRA) and Evolved Universal Terrestrial Radio https://doi.org/10.1186/1687-1499-2014-89 Access Network (E-UTRAN); Overall [15] Vukadinovic, V., Bakowski, K., Marsch, P., description; Stage 2. (2016). In 3GPP TS 36.300. Garcia, I., Xu, H., Sybis, M., Sroka, P., [5] Kiela, K., Barzdenas, V., Jurgo, M., Macaitis, V., Wesolowski, K., Lister, D., & Thibault, I. (2018). Rafanavicius, J., Vasjanov, A., Kladovscikov, L., 3GPP C-V2X and IEEE 802.11p for Vehicle-to- & Navickas, R. (2020). Review of V2X–IoT Vehicle communications in highway platooning Standards and Frameworks for ITS Applications. scenarios. Ad Hoc Networks. 74. Appl. Sci. 2020, 10, 4314. 10.1016/j.adhoc.2018.03.004. https://doi.org/10.3390/app10124314 [16] OMNeT++ Discrete Event Simulator, https://omnetpp.org 167 Tibor Petrov YRS 2021 A ROUTING PERFORMANCE COMPARISON OF LTE-V2V AND IEEE WAVE FOR 15. -17. September EMERGENCY VEHICLE APPROACHING WARNING APPLICATION IN VEHICULAR Portoroz, Slovenia NETWORKS [17] Sommer, C., German, R., & Dressler, F. (2011). [20] Alvarez Lopez, P., Behrisch, M., Bieker-Walz, Bidirectionally Coupled Network and Road L., Erdmann, J., Flötteröd, Y., Hilbrich, R., Traffic Simulation for Improved IVC Analysis. Lücken, L., Rummel, J., Wagner, P., & Wießner, IEEE Transactions on Mobile Computing E. (2018). Microscopic Traffic Simulation using (TMC), vol. 10 (1), 3-15. SUMO. 2019 IEEE Intelligent Transportation Systems Conference (ITSC) (pp. 2575-2582). [18] INET Framework, https://inet.omnetpp.org Maui, USA. [19] Virdis, A., Stea, G., & Nardini, G. (2016). [21] Mannoni, V., Berg, V., Sesia, S., & Perraud, E. Simulating LTE/LTE-Advanced Networks with (2019). A Comparison of the V2X SimuLTE. In M.S. Obaidat, J. Kacprzyk, T. Communication Systems: ITS-G5 and C-V2X. Ören, J. Filipe, (Eds.), Simulation and Modeling 2019 IEEE 89th Vehicular Technology Methodologies, Technologies and Applications Conference (VTC2019-Spring) (pp. 1-5). Kuala (Volume 402 of the series Advances in Lumpur, Malaysia. doi: Intelligent Systems and Computing, pp. 83-105). 10.1109/VTCSpring.2019.8746562. Springer, doi: 10.1007/978-3-319-26470-7_5 168 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY OF BŘECLAV Eva Pitlová, Ing. PhD. Lucie Vyskočilová, Ing. CDV - Transport Research Centre Líšeňská 33a 636 00, Brno, Czech Republic eva.pitlova@cdv.cz, lucie.vyskocilova@cdv.cz ABSTRACT The red signal violation may lead to traffic conflicts that often result in death or serious injuries at signalized intersections. Therefore, the flashing green was designed to work as a psychological measure to eliminate this violation. It works on the principle of dilemma zone reduction as it simplifies drivers’ decision to smoothly stop or safely drive through the intersection. As side effects of the flashing green introduction, the increase of premature stops and rear-end collisions were monitored in recent research. The aim of this study is to demonstrate the flashing green effect on the intersection safety and capacity. For this purpose, a pilot study of 6 consequent intersections in Břeclav was carried out. Drivers' behavior surveys were performed in two ongoing steps. During traffic conflict observation, video from a drone situated above the intersection was recorded. The results show decrease in vehicles’ speed at time of yellow onset. On the contrary, a slight increase in conflicts at the time of flashing green onset was observed. The survey shows that drivers stop earlier before yellow onset which leads to a 1,8s reduction of effective green time. Interviews with local drivers about the signal change show peopleś positive attitudes towards the flashing green introduction (84%). Furthermore, results show that a quarter of respondents do not know how to correctly respond on flashing green signal, so a nationwide campaign would be needed before flashing greenś widespread introduction. Keywords: intersection, signal plan, flashing green, driver’s behavior, survey. 1. INTRODUCTION Therefore, the intermittent green signal, also known as flashing green, was designed to work as a Traffic signal change from green to yellow informs psychological measure to eliminate the red and the drivers about the end of free flow. According to yellow signal violation at signalized intersections. It that they can adapt their speed and therefore stop at works on the principle of reducing the dilemma zone intersection. Still, there exist many drivers that pass and thus makes it easier for drivers to decide whether through the intersection during red signal because the they can smoothly stop or safely drive through the length of yellow signal is insufficient to safely stop at intersection. Its introduction is also recommended by the intersection. In a recent CDV - Transport traffic psychologists, who rely on research and Research Centre survey (Ambros, 2013) in Czech experiences in many countries (Austria, 1960). Republic the stop obligation during the intersection However, drivers’ behavior among countries may red signal is violated in 15% of the cycles and the vary, and therefore, what is good for one country may obligation to stop during the yellow signal is violated not be useful for another. The aim of this study is to in up to 30%. As results show, most of the drivers demonstrate the effect of the flashing green on the violated the yellow signal on purpose. They approach safety and capacity of signalized intersections in the the intersection in such a great distance when the Czech Republic. yellow signal onsets that they can safely stop but instead of it they decide to take the risk and get 1.1. Flashing green experience through the intersection unsafely. Such traffic For the first time in Europe the flashing green was violations occur in more than 1/2 of the traffic signal applied at signal controllers in Austria, in 1983. After cycles, in which traffic accidents occur. The driver’s successful implementation, many countries followed failure to comply with the red or yellow signal stop their example such as Switzerland, Germany, obligation exposes the drivers themselves as well as Estonia, Latvia, Lithuania, and others (Köll, 2002). the other road users to the risk of an accident. Extensive research in Germany and Switzerland 169 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia (Factor R., 2012) have shown that, in addition to its lights or equip it with a counting device, which is positive impact, the flashing green can also cause the expensive especially for a nationwide application. In driverś earlier stops at intersection which may lead addition, currently most of the traffic lights operate to a capacity reduction and rear – end collisions. This under the dynamic traffic management, to accurately is due to the driverś sudden and premature stop as a react on the actual traffic flow and, for these signal reaction to the signal change into the flashing green. plans, the countdown traffic lights cannot be used. The research results show different numbers of International studies (Köll, 2002) (Novotný T., 2013) driverś earlier stops throughout the countries. In (Tang, 2015) (Wu, 2014) confirm that flashing green Austria, the increase rises to 80%, in Switzerland has a positive psychological effect on the driver, 35% and in Germany 28%. which reflects in intersection safety enhancement. Later in 2009, there was a study of flashing green However, it has been found that in addition, flashing impacts at 9 intersections and 3 crossings in green may result in capacity decrease or in a slight Bratislava, capital of Slovakia (Novotný T., 2013). accident increase in the specific cases of rear-end After its completion, the introduction of flashing collisions (Koll, 2004). Therefore, it is necessary to green at all intersections continued for three more verify these possible negative effects on a selected years with recommendation to apply the flashing sample of intersections before its widespread green at all intersections in the city. After the three implementation. years, the city decided to set it aside, although there 1.2. The law behind its application was no negative effect monitored on traffic flow fluency and the intersections safety was increased. It The flashing green introduction preceded the turned out that at several intersections in Bratislava legislation change. This change has been already there are obsolete traffic light controllers and for the made in neighboring countries such as Austria, flashing green introduction they would have to be Slovakia, and Poland (Austria, 1960) (Slovak replaced with new ones, which would lead to high Republic, 2009) (Poland, 2002). There is a difference financial demands for the city. Due to the in the signal meaning and legislation among the disintegration of the intersection’s coordination, the countries, specifically in the definition of the yellow city even decided to cancel the flashing green at the signal. While in Austria there is hard restriction existing intersections. passing on the yellow signal, in Slovakia, as in the In Poland, extensive survey of driver’s opinion about Czech Republic, the law allows drivers to pass innovations in traffic light signals shows that only through the intersection when the yellow signal 12% were in favor of flashing green, 29% of onsets and the car approaching the intersection is in countdown signals, and 10% of those asked preferred such a small distance that it is impossible to safely a combination of both (Łopuszyński, 2018). stop at intersection. In Austria, the flashing green According to the interview results we can say that means end of the free driving through phase, and drivers in Poland are quite positive about the signal although its effect on the driver behavior has been modification of traffic lights, but rather than flashing demonstrated, still it is not considered in the capacity green they prefer traffic lights with information about calculations of traffic lights (Factor R., 2012). the remaining signal duration, which is currently the 2. DATA AND METHODS safest, but more expensive option and has its limitations. The flashing green impacts were determined based on the driver behavior pilot study at selected Outside Europe, the flashing green exists mostly in intersections. The survey was performed in two steps. the US, Mexico, Turkey, Israel, Russia, China, India, While there was a drone recording current and Cambodia. The most significant research took intersection from above, the conflicts on selected place in America and China (Wu, 2014) (Tang, approaches were concurrently monitored. In 2015). The results show that the flashing green has a addition, a questionnaire was created to observe the positive psychological effect on the driver, which information about driver’s knowledge, manner, and reflects in the intersection safety. However, currently attitudes towards the intermittent green signal at the the traffic lights with signaling countdown are being traffic lights. introduced much more instead of the flashing green. The countdown signal provides drivers with 2.1. Case study in Břeclav information about the duration of the current phase when they approach an intersection and so it is easier Based on the research abroad and with regards to for them to adjust their behavior. In addition, the red technological requirements, 6 intersections in signal countdown significantly reduces the vehicleś Břeclav, that are part of the 1st class road number 55, delay time at the start of green phase, thus increasing were selected for monitoring driver behavior. The the overall capacity of the intersection. To introduce city is located near the Austrian and Slovak borders the countdown device at an intersection, it is and therefore it is considered that the drivers are necessary to completely replace the current traffic aware of the flashing green operation. The signal plan 170 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia was modified from 1st September to 29th October The drivers' reactions were monitored in three phases 2019 at all six intersections shown in Fig. 2, after and compared to each other: obtaining all the necessary permits. The test  F1 before the flashing green introduction, operation of the modified signal plan was allowed  only for 2 months, and therefore the driver's behavior F2 shortly after the flashing green was monitored in two steps - by analyzing the image introduction (first week after introduction),  from camera recordings taken from drones and traffic F3 about a month after the flashing green conflict observation on the site. introduction (6th - 9th week after the introduction). Since the drone can only fly at a specific height above buildings and roads, most of the intersections were recorded from an angle, and therefore some approaches or whole intersections were hidden behind the trees or buildings. Thus, two out of six intersections were excluded from the survey (Fig. 2). Nevertheless, recording was performed at 4 intersections (Table 1) during the saturated traffic flow and in good weather conditions (without raining and strong wind). The monitored approaches of monitored intersections in Břeclav. Figure 1: Location of Břeclav in the Czech republic Table 1: The monitored approaches of monitored and neighboring countriies intersections in Břeclav 2.2. Data collection Road crossing Int. Approach 1 Approach 2 the I/55 To provide complex data analysis of traffic flow, K1 Sovadinova I/55 – Vienna Sovadinova there was a drone recording the intersection from K2 17.listopadu I/55 – Brno II/425 above, while the potential conflicts were monitored. In addition, during the investigation the questionnaire K3 Pod Zámkem I/55 – Brno I/55 - Vienna interview of local drivers was performed. K4 U Nemocnice I/55 – Brno U Nemocnice Figure 2: Map of monitored intersections with flashing green modification in Břeclav 171 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia passing through predefined profiles. On the 2.3. The survey approaches, profiles in 10m distances were formed For traffic accident comparison before and after the starting from the stop line, continuing up to 50m flashing green introduction, the long-term driver (Fig. 3). Based on this information, the average behavior monitoring would be the appropriate option. speeds of each recorded vehicle passing between the However, the test run of the modified signal plan was individual profiles were calculated. allowed only for 2 months, which is a very short The vehicle passages were time synchronized with period for this type of comparison and the results of the signal plans for each traffic movement. The three this method would be insufficient. Therefore, the phases of the survey were compared to each other and traffic conflict monitoring methods were used to evaluated due to: monitor driver behavior. At the time of the drone recording, the traffic conflicts at selected approaches  vehicle speeds that were affected or were were also monitored according to the methodology not by signal change, measured in individual (Ambros, 2013). At each intersection, monitoring zones (between profiles – 10 m), was performed at two approaches at a time. Of all  speed during the yellow signal and possible conflicts, only rear-end types of conflicts in acceleration immediately after the yellow 1–3 gravity were monitored. The conflicts of zero signal onset, severity (violation of red signal) were not reported.  speed on the STOP line for vehicles passing Monitoring focused especially on the signal change an intersection at a yellow or red signal, from green to flashing green.  the proportion of cycles in which at least one vehicle passed the intersection on a red or yellow signal, relative to the total number of cycles. To analyze the knowledge, manner, and attitudes of drivers, local interviewing was carried out using two 0 m combined questionnaire surveys. The first survey took place in Břeclav where data were collected 10 m directly close to the investigated intersections and the second survey was conducted as a classical representative survey. Field data collection took 20 m place near the investigated intersections after about a month of its signal plan modification, while drivers 30 m were asked about their knowledge of the flashing green and the change in their behavior. The sample consisted of a random selection of drivers. The national data collection was carried out using a 40 m statistical survey conducted based on quota sampling. The aim of the combination of these two methods was to verify the influence of flashing green on drivers' behavior and knowledge. 50 m 3. RESULTS Figure 3: Example of profiles at an intersection K4 According to atypical values obtained from conflict The drone recording option was chosen, after monitoring, the individual F1, F2, F3 phases were surrounding inspection that there was not any suitable compared with each other based on the rate of building or structure for static camera fixation to conflicts (number of conflicts on selected approach record necessary area of whole intersection with its related to the traffic volume, expressed as a approaches. The drone was able to record percentage). Special attention was given to the results approximately 18 minutes video with loaded battery comparison of phase 1 and phase 3. (depending on weather and wind) and after this time The results of the relative conflict rate in each phase it needed a battery change. The battery changes were for investigated intersections and its approaches are repeated as many times as necessary to achieve the shown in Table 2 colored for easy comparison (green desired intersection recording time. The drone - lowest value, yellow - mean value, red - highest always recorded the view from the same point and value). It is obvious that at some intersections after angle even after the battery change (based on GPS the flashing green introduction, the gradual coordinates). For each intersection, image analysis development of the increase in conflicts was was performed to obtain text files with time data monitored for each phase. This means that even a (accurate to thousands of seconds) of the vehicles month after the signal modification, the drivers were 172 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia not used to it and their uncertainty has risen. Due to increase at the K1 intersection comparing phase F1 the short testing period (2 months) the further and phase F2 but again in phase F3 it decreases. development in driver behavior cannot be predicted. According to the vehicle speeds, after the flashing green introduction at all investigated intersections, Table 2: Conflicts monitoring evaluation the speeds decreased at the time of the yellow signal beginning. This means that the drivers drove at a Intersection Monitored conflicts Relative confliction [%] lower speed when the signal changed to yellow and so they were more prepared to stop at the red signal. approach That leaded to a reduction in the speed of vehicles F1 F2 F3 F1 F2 F3 that passed the intersection on yellow or red signal at 1 9 10 2 0,413 0,465 0,082 all intersections (Table 3), which reduces the gravity K1 2 0 0 0 0,000 0,000 0,000 of the consequences in case of a vehicle collision. 1 0 2 2 0,000 0,101 0,136 Based on the fact that more than one vehicle can pass K3 on the red or yellow signal in one cycle, the relative 2 0 0 0 0,000 0,000 0,000 number of cycles that were violated (at least one 1 1 5 10 0,038 0,233 0,386 K4 vehicle passed on the red or yellow signal) were also 2 0 4 7 0,000 0,123 0,224 evaluated. To better illustrate that, Table 3 1 1 6 7 0,043 0,235 0,269 demonstrates the violated cycles. At two K5 2 1 4 1 0,165 1,361 0,208 intersections (K2, K3) the percentage of red signal violated cycles keep constant in all phases (around 5%), while at the remaining two intersections (K1, The image analysis from drone recording shows that K4) the percentage of cycles decrease comparing the relative number of vehicles passing the phase F1 with third phase F3. intersection at the red or yellow signal decreased at all intersection approaches comparing the first phase The values for each intersection in the three phases F1 with the phase F3 (at the K5 intersection was the (F1, F2, F3) are colored for easy comparison (green decrease even from 35% to 4%). There was just slight – positive improvement, yellow – middle stage, red – negative deterioration). 173 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia 3.1. Flashing green impacts What was demonstrated in this paper is that, due to phase F1 phase F2 phase F3 the flashing green at all intersections, there was an 40 increase in the number of vehicles that reacted on signal change approximately in 30-10 m before the STOP line and stopped. This distance is a critical dangerous S value, as the necessary car stopping clearance in 30 R zone BE driving speed of 50 km/h (13.9 m/s) is around 35 m. At the time of the yellow signal onset, the drivers UMN were therefore able to better evaluate whether they 20 ICT would safely pass through the intersection or not. On LFN the other hand, sudden deceleration of vehicle speed CO that is approaching the intersection can cause rear- end type of conflicts. Furthermore, surveys showed 10 that about a third to half of drivers stop prematurely in front of the intersection during the flashing green. This, together with the fact that drivers reduce their 0 speed, could negatively affect the capacity of 0 1.000 2.000 3.000 4.000 intersection. TRAFFIC FLOW From the results of conflict monitoring there was a Figure 4: Conflicts gravity evaluation according to the slight increase in the monitored type of conflicts after current intersection traffic volume the introduction of the flashing green. These were mainly the low severity conflicts (numbered as 1), According to the Czech guidelines for the which corresponds to the expected normal maneuvers intersection capacity calculation called TP 188 of the front vehicle and the subsequent reaction of the (Bartoš, 2018) the effective green is considered following vehicle. In Table 2 the relation between depending on the duration of the green signal, and its monitored conflicts and the current traffic flow as a minimum value is the length of the green signal itself relative confliction for each intersection divided in (z '= z at values z ≥ 11 s). It is commonly expressed three phases is determined. Despite a slight increase as follows (Bartoš, 2018): in rear-end type of conflicts in the second and third z ‘ = z − r + ž (1) phases (F2, F3), all intersection approaches still present the safe state according to Figure 4, so the where: introduction of flashing green did not significantly z ‘- effective green signal time [s], affect the intersections safety. In case of vehicle speed, after the flashing green introduction, there was z - green signal time[s], a decrease in speed at all evaluated intersections at r - vehicle start-up time (usually 1 s is considered) [s], the time of the yellow signal beginning. The drivers drove at a lower speed when the signal changed to ž - the yellow signal length (approx. 2 s) [s]. yellow and prepared to stop at the red signal. At the In the case of the flashing green implementation, it is same time, at all intersections there was a reduction necessary to consider its impact on the effective in the vehicle speed that passed the intersection on a green. It will be necessary to consider this effect of yellow or red signal, which reduces the gravity of the the vehicles that stop at the flashing green in the collision consequences. calculation. According to the results, a third to half of drivers stop prematurely at the stop line during the flashing green. The aim of this study is to evaluate the weight of the impact on effective green at investigated intersections. Based on the obtained data, the real time of the effective green was analyzed. Due to the dynamic controlled signal plans at the investigated intersections, the green length varies and so every cycle monitored in the second and third phase (F2, F3) was counted including the flashing green. The real time of effective green was counted based on the time difference between the first and last vehicle that passed in one cycle (regardless of whether the last vehicle passed the intersection on the green, yellow, or red signal). Only cycles with saturated flow were selected for analysis. 174 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia This means that the green time has been fully utilized respondents (419 people) do not remember. The high without large vehicle distances. In the following percentage of people who do not have experience Figure 5 the relation between real effective green with the flashing green means that a nationwide time and the green length of all analyzed cycles is campaign needed to be held before it widespread shown. introduction in Czech Republic. In addition to a nationwide survey, a local driverś experiences with phase F1 phase F2 phase F3 TP 188 65 flashing green were investigated. The people were y = 0,9723x - 0,2495 interviewed directly around the investigated R² = 0,7558 intersections and with the help of the city itself by using their web portal. The survey was held about a 60 [s] month after the introduction of flashing green testing. ne The sample had the character of a driver’s random er y = 0,9219x + 2,7965 selection who experienced the flashing green testing g e R² = 0,7821 55 in Břeclav city. tivc ffee eth 50 f o e timl a 45 y = 0,9392x + 1,2379 Re R² = 0,7754 40 40 45 50 55 60 65 Green signal lenght [s] Figure 5: Efective green comparison for each phase and with the TP regulations (Bartoš, 2018) The value of effective green determined according to Figure 6: National survey results about flashing green the valid regulation TP 188 (Bartoš, 2018) is marked expierence with dotted red line. From the plotted curves the influence of the flashing green on the effective green The results show that when the flashing green occurs, real time value was determined. Throughout the up to 39% of respondents answered that they would analyzed range, the effective green real times were rather slow down and 8% would stop completely. lower than the considered value. Furthermore, we can 15% of people had the opposite reaction and rather say that the values of effective green real times differ accelerated when flashing green occurs and 16% of in whole analyzed range in constant value of 1,8 s people said that they would react differently. 22% of comparing the first (F1) with the third phase (F3). respondents said that the flashing green has no effect Comparing the development of effective green real on their driving behavior. According to the safety times monitored in the third phase (F3) with the value side, 78% of respondents evaluated it positively. The calculated according to the TP188 (Bartoš, 2018), the 8% of respondents think that flashing green has a value is not constant in the monitored range, but the negative effect. This also results in a people’s average value is reduction of about 3,1 s. These positive attitude towards its nationwide introduction reductions in the effective green real time as a in the Czech Republic presented by 84% of consequence of flashing green introduction are quite respondents. significant and it would be appropriate to be considered in regulations capacity calculations (Bartoš, 2018). To accurately determine the impact of flashing green on capacity, it will be necessary to perform long-term monitoring of the light-controlled intersections with flashing green implementation, with respect to the effective green time monitoring. 3.2. Driverś interview The national survey shows that less than a quarter of respondents (24%, 601 people) have already encountered flashing green at traffic lights, while more than half of the respondents (60%, 1508 people) have not yet encountered this signal. About 16% of 175 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia Figure 7: National survey results about flashing green attitude In Figure 7 the complex results of the survey in terms mistakenly marked meaning of it which proof the of the driver’s opinion about flashing green impact at previous studies about their lack of knowledge. As the fluency, safety, and comfort are presented. The for their reaction, 39% of asked said they would largest proportions of positive answers were given by rather slow down when flashing green occurs, which a group of drivers with B +. On the contrary, the most corresponds to 41% of the premature stopped negative answers were given by a group of small vehicles, monitored one month after the flashing drivers under 24 years of age. green introduction. For example, in Poland the interview results show people’s positive attitude 4. DISCUSION towards flashing green but it also shows that they Drivers’ reaction on signal change may vary prefer more the countdown devices so they can see the remaining time of red, which could improve the throughout the countries. According to that, most of capacity as the green begins or with the green signal the countries carried out their own studies about when they approach intersection, they feel better as flashing green before its widespread introduction. As they know the remaining time which affect the safety. side effects, the premature stops of vehicles at However, the countdown device as traffic light intersections resulting in slight increase in rear-end cannot be used at the intersections with dynamic type of conflicts when the flashing green onset were signal plans which are nowadays common in most reported. cities. It is also more expensive if we compare it In Austria, the side effect on traffic flow fluency the nationwide introduction with the flashing green 3% capacity reduction was determined. In this pilot introduction expenses. In this case Flashing green is the negative effect was proven in 1,8 s reduction of more appropriate. effective green measured one month after the flashing green introduction and comparing it to the calculated 5. CONCLUSION effective green (Bartoš, 2018) the reduction is even The survey shows an increase in the rear-end type of 3,1 s. conflicts after the flashing green introduction in In addition, an increase in rear-end type of conflicts almost all monitored intersections approaches (5 out was observed too but the number and gravity of of 8) . This means that the drivers had a higher monitored conflicts still preserve the intersections in confidence in their behavior when the flashing green safe side according to (Ambros, 2013) as it was begin. These were mainly the conflicts of 1st severity established as common maneuver of the consequence level which correspond to the expected maneuvers of vehicle reacting to sudden stop of previous one. The the previous vehicle and the subsequent reactions of reduction in speed has also been demonstrated and the next vehicle. Despite a slight increase in conflicts due to that the number of red lights violated cycles in second and third phases (F2, F3), the intersections decrease. are still in a safe side. The evaluation of speeds after the flashing green introduction shows, that there was Most of the surveys beyond Europe about flashing a decrease in speeds at all evaluated intersections at green, point out that even the dilemma zone has the time of the yellow signal beginning. This means decreased, but the drivers incorrectly react on it and that, after the introduction, the drivers drove at a so it is necessary to enlighten people before its lower speed when the signal changed to yellow and widespread introduction. For this purpose, as a part of this study the local’s knowledges, reactions and therefore were better prepared to stop at the red signal. That leads to reduction in the speed of opinion about flashing green was carried out. Their vehicles that passed the intersection on a yellow or reaction was positive about its introduction but red signal at all intersections, which reduces the mostly the younger group of drivers (18-24 years) gravity of the consequences in case of a vehicle 176 Eva Pitlová, Lucie Vyskočilová YRS 2021 THE IMPACTS OF FLASHING GREEN AT SIGNALIZED INTERSECTIONS – CASE STUDY 15. -17. September OF BŘECLAV Portoroz, Slovenia collision. Also, the results show that there was a which permitted the survey to carry out the necessary decrease in number of the red signal violation cycles observations. one month after the flashing green introduction at all intersection. Furthermore, the survey showed that REFERENCES about a third to half of drivers stop prematurely at [1] Ambros, J. et al., (2013). Methodology for flashing green. This could negatively affect the Monitoring and Evaluating Traffic Conflicts, signal-controlled intersection capacity. According to Transport Research Center, v. V. I., Czech this fact, the real and calculated effective green was republic: Brno compared for each intersection monitored in the three phases. Based on the monitored data the 3 seconds [2] Austria (1960). Federal Ministry for Digital and reduction of effective green was determined Economic Affairs: Act Concerning the Granting of Asylum 1960. In BGBl. Nr. 159/1960 zuletzt compared with the calculated values according to TP geändert durch BGBl. I Nr. 116/2010; § 38. 188 (Bartoš, 2018) and about 2s reduction was Bedeutung der Lichtzeichen. Vienna, Austria: determined comparing the first and third phase (F1 – Bedeutung der Lichtzeichen. before flashing green introduction, F3- one month after introduction). Nevertheless, these are the data [3] Bartoš, L. et al. (2018). TP 188 Technical regulation for capacity calculaton of the from the research at 4 intersections and the effect of intersection and roads profiles. Plzeň: Koura flashing green was monitored only for the permitted Publishing. two-months duration of signal plan modification. Due to the type of intersections surveyed for this [4] Factor R., Prashker J.N., Mahalel D.(2012). The study, it would be appropriate to carry out an flashing green light paradox. Transportation Research Part F: Traffic Psychology and additional survey at intersections with a higher Behaviour.Volume 15, (pp. 279-288). permitted speed (due to the longer braking distance of vehicles) and less congested traffic. Long-term [5] Köll, H., Axhausen, K., Bader, M. (2002). monitoring is necessary to better quantify the Decision-making behavior at traffic lights with capacity reduction, together with a focus on selected and without flashing green as an announcement of the transition time yellow. characteristics of the traffic flow distinguishing Straßenverkehrstechnik, pp. 339–345. vehicles categories. To accurately determine the flashing green impact on capacity, it will be [6] Koll, H., Bader, M., Axhausen, K.W. (2004). necessary to perform long-term monitoring of Driver behavior during flashing green before signalized intersections with flashing green amber: a comparative study. Accident Analysis introduction, focusing on monitoring the effective and Prevention, pp. 273-280. green time. The observations need to be during the [7] Łopuszyński, M., Pendzich, I. (2018). Need to fully saturation of the monitored approaches. use advance notice for urban public transport drivers. Scientific Journal of Silesian University The nationwide survey showed that a quarter of of Technology, pp. 81-90. respondents have already experienced the flashing green. The high percentage of people who did not [8] Novotný T. (2013, august 6). The flashing green disappears from Bratislava, some experts do not have such experience means that a nationwide agree. Retrieved from Auto SME: campaign will be necessary before its introduction https://auto.sme.sk/c/6887860/blikajuca-zelena- and integration into the legislation of the Czech sa-z-bratislavy-vytrati-niektori-odbornici- Republic. The local survey focused on people who nesuhlasia.html had to respond directly to the flashing green as a [9] Poland (2002). Regulation of the Minister of driver. According to the local survey results, up to Infrastructure and Internal Affairs and 39% of respondents lowered their speed and 8% Administration.:Signs and Signals lights. Poland: stopped completely when the flashing green Ministers of Home Affairs and Administration. occurred. Even 78% of respondents evaluated the flashing green impact positively in terms of safety. [10] Slovak Republic (2009). Ministry of the Interior Regulation: Road Act 9/2009. In M. o. This also results in a positive attitude towards its Regulation. Bratislava: Ministry of the Interior nationwide introduction in the Czech Republic Regulation. presented by 84% of respondents. [11] Tang, K., Xu, Y., Wang, P., Wang, F. (2015). Acknowledgments Impacts of Flashing Green on Dilemma Zone Behavior at High-Speed Intersections: Empirical The data presented in this paper were obtained in Study in China. Journal of Transportation study financed by the Ministry of Transport based on Engineering, pp. 1-13. Decision No. 2 / RVO / 2018 File no. 5 / 2018-710- VV / 1 as the institutional support for the long-term [12] Wu, Y. (2014). Comparative Analysis of Different Dilemma Zone Countermeasures at conceptual development of a research organization and implemented with the support of the Břeclav Signalized Intersections based on Cellular city, Automaton Model. Orlando, Florida: Dissertation 177 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia EFFECTIVENESS OF ADVANCED EMERGENCY BRAKING SYSTEMS IN TRUCKS: AN ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Leon Straßgütl, M.Sc. Federal Highway Research Institute (BASt) Brüderstraße 53, 51427 Bergisch Gladbach, Germany strassguetl@bast.de ABSTRACT Advanced emergency braking systems (AEBS) are advanced driver-assistance systems that can prevent or mitigate rear-end collisions. Since 1st November 2015, these systems are mandatory for newly registered heavy goods vehicles (HGV) and buses above eight tones (vehicles of categories N2, N3 and M3) according to EU-Regulation No. 347/2012. This paper aims at evaluating the effects associated with the Regulation in Germany. It addresses two research questions: (1) How did the number of rear-end collisions of HGV and buses develop after the Regulation was introduced? (2) Did the severity of rear-end crashes change after the implementation of the Regulation? To examine the impact of the Regulation, the analysis used individual data of police records on crashes as well as data on technical vehicle information from the German Federal Motor Transport Authority (KBA). At first, relevant vehicles and crashes were identified in the data. Subsequently, treatment and control groups were defined that were compared before and after the Regulation came into effect. The treatment group were rear-end crashes of new vehicles since AEBS mostly affect these crashes. Other crashes and crashes of older vehicles served as control groups. Odds ratios and chi square tests were used to estimate the impact associated with the Regulation. The results indicate a substantial decrease of rear-end collisions on German motorways associated with the Regulation. Moreover, the results further suggest a reduction in the severity of crashes. Keywords: Advanced emergency braking systems (AEBS), Active safety systems, Heavy goods vehicles (HGV), Road safety, Crash analysis. 1. INTRODUCTION Regulation No. 347/2012 (European Commission 2012a). The effect associated with the Regulation Crashes involving heavy goods vehicles (HGV) was evaluated in this paper. have particularly severe consequences. This is especially true for rear-end crashes. Approximately Several studies have estimated the potential effect two thirds of rear-end crashes of HGV occur on AEBS have on the number of rear-end crashes. For motorways. This crashes account for more than 90% Germany, for instance, Kuehn et al. (2011) of all fatalities in rear-end crashes with HGV estimated the potential of autonomous emergency (Panwinkler 2018). braking systems in trucks on the basis of crashes with personal injury or property damage above Advanced emergency braking systems (AEBS) offer 15,000 € in 2004 to 2006. They inferred that these the potential to prevent or mitigate rear-end crashes. systems could have had an effect in 27% of rear-end These systems detect collision risks using camera, crashes on all roads if they only detected moving radar and/or laser. The driver is first alerted about targets and 52% if they additionally detected the possible collision. If the driver does not react by standing targets. braking or steering, the system autonomously initiates an emergency braking. For HGV and buses, In an in-depth analysis of crashes of HGV with these systems are mandatory as regulated in EU- personal injury in Brandenburg in 2016, Trabert et 178 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia al. (2018) estimated that 21 out of 25 crashes (84%) studies. Jermakian (2012) estimated that 37% of all could have been prevented or mitigated with an ideal rear-end crashes of large trucks between 2004-2008 AEBS. An ideal AEBS was presumed to detect an could have been prevented in the US if the vehicle obstacle if it was visible and in the same lane when had had a forward collision warning system. In 83% the gap between HGV and obstacle was at least 150 of the crashes that were assessed as preventable the m. The analysis included all road types although the driver did not brake. For non-fatal injury and fatal majority of crashes occurred on motorways. rear-end crashes respectively, the author estimated a decline by 29% and 17%. Petersen et al. (2020) analysed rear-end collisions with severe personal injury by HGV on motorways Woodrooffe et al. (2013) forecasted the effect of an in Lower Saxony and concluded that 23 out of 40 autonomous emergency braking systems (collision- rear-end collisions of HGV without AEBS could mitigation braking and forward-collision warning) have been prevented if the vehicle had had AEBS. that detects stationary and moving targets to 28-40% This corresponds to 58%. The high effectiveness of for the US. these systems has also been found in international Figure 1: Timeline. These studies estimated the potential benefit of effect on HGV crashes associated with the EU- AEBS. They all assumed that the system was Regulation. activated at all times and that the driver always The article thus addresses two research questions: reacted appropriately. Petersen et al. (2020) and Seiniger et al. (2020), however, pointed out that How did the number of rear-end collisions of HGV drivers could instinctively react to an imminent and buses develop after the Regulation was collision by braking or steering and thus introduced? inadvertently overriding the system. Did the severity of rear-end crashes change after the implementation of the Regulation? To factor in potential shortcomings in the interaction between AEBS and driver, an ex-post analysis of These questions were analysed using individual data real-world crashes is necessary. For the US, Teoh of police records from 2010 to 2018. The remainder (2020) analysed crashes of large trucks on of the article is organised as follows: Section 2 motorways between 2017 and 2019 and found that illustrates the details of the Regulation. Section 3 the crash rate of rear-end collisions of trucks describes the data and method used to analyse the equipped with automatic emergency braking effects associated with the Regulation. Section 4 systems had been reduced by 41%. The analysis of presents the results. In section 5, the findings and Petersen et al. (2020) also suggests that HGV with limitations are discussed. Section 6 concludes. AEBS were disproportionately less involved in rear- end crashes on motorways in Lower Saxony in 2. BACKGROUND Germany than those vehicles without AEBS. In the European Union, AEBS are mandatory for An investigation of the ex-post effectiveness of new HGV and buses with a few exceptions. The AEBS in Germany is missing so far. This analysis is details are regulated in EU-Regulation No. 347/2012 supposed to close this gap. This paper examines the (European Commission 2012a).1 The Regulation 1 Identical regulations were issued at UNECE level (UNECE 2013). 179 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia came into effect at two approval levels each first for Table 1: EU-Regulation No. 347/2012. new types of vehicles and two years later for new Approval Type of Categories Effective as vehicles. The first approval level became effective level vehicles of vehicles* of for new types of vehicles on 1st November 2013 and 1 New types M3, N3, N2 01/11/2013 of vehicles > 8 t for new vehicles on 1st November 2015. The second New M3, N3, N2 01/11/2015 approval level came into force for new types of vehicles > 8 t vehicles on 1st November 2016 and for new vehicles 2 New types M2, M3, 01/11/2016 on 1st November 2018. There is no obligation to of vehicles N2, N3 retrofit already registered (older) vehicles. New M2, M3, 01/11/2018 vehicles N2, N3 At the first approval level, AEBS were obligatory * Except vehicles > 3 axles, off-road vehicles, certain special purpose vehicles, semi-trailer trucks > 8 t, buses in which the for vehicles of the following categories as defined in carriage of standing passengers is possible. Directive 2007/46/EC (European Commission Source: European Commission (2012). 2007): AEBS detect collision risks using camera, radar  Category M3: Vehicles designed and and/or laser. If the system detects a collision risk, it constructed for the carriage of passengers, alerts the driver. The system issues a second warning comprising more than eight seats in addition if the driver does not react to the first warning by to the driver’s seat, and having a maximum braking or switching lanes. During this warning permissible gross vehicle weight exceeding 5 phase, the system reduces the speed of the vehicle tonnes. only slightly. If the driver does not respond to the  Category N3: Vehicles designed and alerts, the system initiates the emergency braking constructed for the carriage of goods and phase. Towards a stationary target, the system had to having a maximum permissible gross vehicle reduce its speed by 10 km/h at the first approval weight exceeding 12 tonnes. level and by 20 km/h at the second approval level.2  Category N2 (AEBS obligatory only if If the target is moving, the collision must be exceeding 8 tonnes): Vehicles designed and prevented regardless of the approval level. constructed for the carriage of goods and To avoid confusion due to false warnings, the driver having a maximum permissible gross vehicle can manually deactivate the system. The driver can weight exceeding 3.5 tonnes but not furthermore override the system during an exceeding 12 tonnes. emergency brake by accelerating for instance. At the second approval level, AEBS became additionally mandatory for vehicles of these 3. DATA AND METHODS categories: Individual German police records from 2010 to 2018  Category M2: Vehicles designed and was used for this analysis. Additionally, technical constructed for the carriage of passengers, vehicle information was used that the German comprising more than eight seats in addition Federal Motor Transport Authority (KBA) provided. to the driver’s seat, and having a maximum 3.1. Study object permissible gross vehicle weight not exceeding 5 tonnes. As illustrated in section 2, the Regulation became  Category N2 ≤ 8 tonnes. effective at two approval levels first for new types of vehicles and two years later for new vehicles. This Exempted from the Regulation are vehicles with study examines the effect of approval level 1 for new more than three axles, off-road vehicles, certain vehicles. The reason for that is twofold. First, there special purpose vehicles, semi-trailer trucks with a is only information available on the year in which maximum permissible gross vehicle weight not the vehicle was originally registered. The date of the exceeding 8 tonnes and buses in which the carriage vehicles’ type approval cannot be derived from the of standing passengers is possible. The data. Second, since approval level 2 did not come specifications of the Regulation are summarised in into force for new vehicles until the end of 2018, Table 1. there was simply not yet (sufficient) data to analyse this effect. As AEBS became obligatory by the end of 2015 for new vehicles, the analysis period began in 2016 and ended in 2018 – the last year for which data was available. The number of crashes in this period was compared to a reference period from 2010 to 2012. 2 For vehicle of categories M2 and N2 < 8 t, a speed reduction of 10 km/h was sufficient at the second approval level. 180 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia AEBS for new types of vehicles were only situations on motorways are generally less complex. introduced by the end of 2013. Thus, between 2010 It is thus easier to identify the situations in which and 2012, there was no obligation for HGV or buses AEBS (would have) had an effect on the outcome. to be equipped with AEBS and the number of Second, AEBS were originally designed for crashes was not influenced by the measure. Both, motorways (Seiniger et al. 2020). Field tests showed analysis and reference periods represent the that false warnings of AEBS on motorways could examination period. only be observed near motorway construction sites (ibid.). Thus, drivers might be less inclined to Figure 1 shows the timeline of the Regulation’s deactivate the systems while driving on motorways introduction and the periods under examination. compared to other road types. Third, most rear-end 3.2. Vehicles included in the study collisions occur on motorways in Germany. According to Panwinkler (2018), around two thirds At approval level 1, AEBS were – with some of all rear-end collisions were located on exceptions – mandatory for vehicles of categories motorways. Rear-end collisions on motorways are M3, N3, N2 > 8 t. Information on the category of typically also more severe than those on other road vehicles involved in a crash is not included in the types. data. Thus, the category was derived using the variables “category of road user” and “maximum The analysis considered all crashes with casualties permissible gross vehicle weight”. To avoid (killed or injured) as well as crashes with property confounding, the categories were narrowly defined. damage only and with at least one vehicle that is not Meaning, the analysis only included those vehicles roadworthy. that most likely were subject to the Regulation. The Furthermore, the analysis included only those variable “category of road user” can take on 37 crashes in which the party mainly responsible for the different values. Fourteen values refer to HGV. Of crash was an HGV or bus. In rear-end collisions, these, nine values were included in the analysis. police record the vehicle that crashes in the vehicle These vehicles most likely comply with the in front of it as the mainly responsible party. AEBS definition of the Regulation. The other values in this vehicle could have helped prevent or mitigate indicate for instance off-road vehicles, special the impact. purpose vehicles or vehicles with a maximum permissible gross vehicle weight lower than 8 t. 3.4. Definition of treatment and control Hereinafter, the term “HGV” refers only to those groups vehicles that have been included in the study. The analysis is conducted using one treatment and The analysis includes only one value for buses from all in all three control groups. These were the five values that refer to this kind of vehicle. All distinguished by two criteria: vehicle age and crash other values also contain vehicles that carry standing scenario. Rear-end collisions of new HGV and buses passengers and were thus excluded to get a narrowly formed the treatment group (Treatment). Rear-end defined group. Henceforth, the term “bus” thus collisions by older HGV and buses served as a first indicates only those vehicles included in the study. control group (Control I). Other crashes of new HGV and buses acted as a second control group The technical vehicle data contains information on (Control II). Other crashes of older HGV and buses maximum permissible gross vehicle weight. were the third control group (Control III). The latter However, this data is only available for a subset of group was only needed to calculate the overall effect crashes. In 2018, for example, this data was missing of the measure. for 20.5% of all HGV involved in a crash on motorways. It is mainly missing if the vehicle was The definition of the groups is summarized in Table registered abroad. In 2018, this applied to 15.5% of 2. Below, the criteria that have been used to define HGV involved in a crash. Data on maximum the groups are described in more detail. permissible gross vehicle weight is also not Table 2: Definition of treatment and control groups. available if the driver fled the scene of the crash or if the police officer recorded the wrong licence Crash scenario Rear-end number. The study excluded crashes with missing Other crash collision technical vehicle information. This ensured that the f New vehicle Treatment Control (I) analysis only took vehicles into account which were e o icle group group covered by the Regulation. Additionally, all vehicles gA eh Older vehicle Control (II) Control (III) v with more than three axles were excluded from the group group analysis. A vehicle was defined as new if it was first 3.3. Crashes included in the study registered during the period under examination. Thus, in the reference period, the registration was The analysis was limited to crashes on motorways. between 2010 and 2012. Equivalently, the This had several reasons. First and foremost, crash 181 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia registration of a new vehicle was between 2016 and to identify rear-end collisions. The type of accident 2018 in the analysis period. Consequently, in the describes the situation which leads to the crash. The first year of each period, only vehicles registered in kind of accident specifies the course of the crash, the same year were considered. For example, in the e.g. the collision or involuntary leaving the road. analysis period in 2016, only vehicles registered in Rear-end collisions were defined as the combination 2016 were defined as new. In the third (and last) year of driving accidents or accidents moving along in of each period, vehicles registered in the same year carriageways (type of accident) and collisions with and in the two preceding years were defined as new. a vehicle moving ahead or waiting (kind of In the analysis period in 2018, for instance, those accident). This definition was based on the vehicles registered between 2016 and 2018 meet the classification in Panwinkler (2018). criteria of the definition (Figure 2). There are two reasons for this differentiation. First, the definition Crashes that were not considered as rear-end had to be constructed in a way that all vehicles in the collisions were defined as other crashes – albeit with treatment group must be equipped with AEBS. some exceptions. A lane departure warning system Second, sufficient data for the analysis was required. became mandatory in HGV and buses at the same time as approval level 1 took effect (European A vehicle was perceived as older if it was registered Commission 2012b). To avoid confounding, lane six years before a new vehicle. In the first year of the departure crashes were excluded from the analysis. period under examination, only vehicles registered These were defined as the combination of driving in one year satisfy the criteria. For instance, in the accidents or accidents moving along in carriageways analysis period in 2016, these were vehicles (type of accident) and collisions with vehicles registered in 2010. In the third year, the definition moving parallel in the same direction or leaving the covers vehicles six to eight years old – thus, vehicles carriageway to the right or left (kind of accident). registered between 2010 to 2012 in the analysis Additionally, collisions with (voluntary) stopping period in 2018. This definition prevented vehicles (e.g. for parking) and collisions with confounding due to differently defined age groups. oncoming vehicles were exempted from the Otherwise, crash years would have been weighted analysis. These crashes are special cases when differently in the age groups and external factors like located on motorways and could have potentially cold or heat waves could have influenced the results. confounded the results. The same applies to single There is no flag for rear-end crashes in the accident vehicle crashes that were also excluded from the data. Therefore, the combination of the variables analysis. “type of accident” and “kind of accident” were used Figure 2: Definition of new and older vehicles. 182 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia The hypothesis was that the Regulation was 4.1. Changes of the level of rear-end associated with an effect on the number of crashes collisions of HGV and buses on motorways. Odds ratios (OR) and chi square tests were computed to test this The treatment group developed more positively than hypothesis. To estimate the overall effect of the the control groups. The number of rear-end system in AEBS-relevant crash situations, changes collisions of new HGV and buses decreased by - between reference period and analysis period in both 13.8% between 2010-2012 and 2016-2018 variables – crash scenario and age of vehicle – were (Treatment). In contrast, the number of rear-end considered. This corresponds to the interaction term. collisions of older HGV and buses increased by It is the same as the ratio of the OR of rear-end about 13.1% (Control I). The odds ratio is collisions in the reference and analysis period and significantly lower than 1 (OR = 0.762, 95% CI = the OR of other crashes in both periods. In this way, 0.591, 0.982) (upper half of Table 3). The chi square the model allows for changes in the composition of test statistic is 4.402 (p = 0.036). These results crash scenarios (e.g. shifts in situations that could suggest that the number of rear-end collisions of new lead to rear-end collisions but no such variations for HGV and buses developed significantly better than other crashes) as well as changes in the composition those of older HGV and buses. The hypothesis that regarding vehicle age (e.g. different developments in there is no difference in the development between the kilometres driven on motorways by older treatment group and control group it can thus be vehicles). rejected. AEBS might affect more severe and less severe The number of other crashes of new HGV and buses crashes disproportionally. The analysis was thus grew by 12.6% between the reference and the repeated for different levels of crash severity analysis period (Control II). The odds ratio between separately to test the hypothesis if the Regulation treatment group and control group II is again lower was associated with changes in the severity of rear- than 1 (OR = 0.766, 95% CI = 0.595, 0.988) (lower end collisions. half of Table 3). The chi square test statistic is 4.235 (p = 0.040). These findings suggest that the number 4. RESULTS of rear-end collisions by new HGV and buses dropped significantly more than the number of other This section presents the results. In sub-section 4.1, crashes. The hypothesis that the development of the the changes in the number of rear-end collisions treatment group is the same as of control group I can associated with the Regulation were analysed. In therefore be rejected. sub-section 0, it was examined if the severity of rear-end collisions changed after the implementation The ratio of the OR of rear-end collisions and of of the Regulation. other crashes is 0.630 (95% CI = 0.420, 0.945). This corresponds to an overall effect associated with the Regulation of -37.0% on the number of rear-end collisions on motorways. Table 3: Results. Odds ratio (95% CI) Crash Period Vehicle age Group Trend Treatment vs. scenario Reference Analysis Control Rear-end New Treatment 311 268 -13.8% Older Control I 191 216 13.1% 0.762 (0.591, 0.982) Other New Control II 191 215 12.6% 0.766 (0.595, 0.988) Older Control III 129 120 -7.0% - Interaction effect 0.630 (0.420, 0.945) Effect of measure -37.0% BASt-U2l-6/2021 183 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia Table 4: Crash severity. Odds ratio (95% CI) Period Crash Vehicle Crash severity Group Trend Treatment vs. scenario age Reference Analysis Control Severe injury Rear-end New Treatment 99 66 -33.3% (fatal/seriously injured) Older Control I 50 74 48.0% 0.450 (0.289, 0.724) Other New Control II 27 33 22.2% 0.545 (0.300, 0.990) Slight injury Rear-end New Treatment 162 169 4.3% Older Control I 114 127 11.4% 0.936 (0.672, 1.305) Other New Control II 72 64 -11.1% 1.174 (0.787, 1.750) Material damage only Rear-end New Treatment 50 33 -34.0% Older Control I 27 15 -44.4% 1.188 (0.551, 2.564) Other New Control II 92 118 28.3% 0.515 (0.307, 0.863) BASt-U2l-6/2021 4.2. Changes in the severity of rear-end with property damage only after AEBS became collisions mandatory can thus be rejected. For vehicle age, the null hypothesis cannot be rejected. In addition to the effect on the total number of crashes, the analysis also investigated the effect on The declining numbers of crashes in AEBS the severity of crashes. The crash’s severity level is situations could thus be primarily attributed to the assigned based on the most severe consequence – decreasing number of crashes with severe injury. thus, a crash with slightly and seriously injured These results suggest that the Regulation is persons is counted as a serious injury crash. Due to associated not only with a decline of the overall the otherwise small number of cases, fatal crashes number of crashes but also with a reduced severity and serious injury crashes were combined into one of crashes. However, especially for crashes with category (severe injury crashes). For crashes with property damage only, the numbers are rather low. property damage only, solely those crashes were These results should therefore be interpreted with considered in which at least one vehicle was not caution. roadworthy. 5. DISCUSSION The number of crashes in the treatment group with severe injury decreased considerable after the The level of crashes declined after the Regulation Regulation came into effect compared to before came into effect for those vehicles for which AEBS (-33.3%). In control group I and II, on the other became mandatory. This reduction is significant hand, the number increased substantially (48.0% and compared to control groups. The analysis further 22.2%). The odds ratio between treatment and suggests that not only the number of crashes control group I and treatment and control group II, declined but also the severity of crashes decreased respectively, were both significantly below 1 for the vehicles affected. implying that the number of crashes had been To check whether the groups are substantially reduced after the Regulation was implemented different, the frequency distribution across selected (Table 4). The null hypothesis that there was no variables were examined. There were no structural change in the number of crashes with severe injury differences detected between new and older vehicles associated with the Regulation can be rejected. with regard to age of driver, time of crash, crash For crashes with slight injury, the results are less severity and proportion of crashes in close proximity clear. In the treatment group as well as in control to construction sites. Rear-end collisions and other group I, the number of crashes grew slightly. In crashes also did not differ substantially with regard control group II, in contrast, the number of crashes to age of driver and proportion of crashes in close sank slightly. No odds ratio is significantly different proximity to construction sites. As seen in section from 1. The null hypothesis cannot be rejected. 4.2., rear-end crashes were more likely severe than other crashes. Rear-end collisions were also more The number of crashes with property damage only prevalent during weekdays than other crashes. declined drastically in the treatment group and in However, the great majority of HGV and bus control group I (-34.0% and -44.0%). In control crashes occurred during weekdays – regardless of group II, the number rose substantially (28.3%). The crash scenario they made up more than 90% of odds ratio of treatment vs. control group II is crashes. Even if crashes on weekends were excluded significantly lower than 1 while the odds ratio of from the analysis, the number of crashes had been treatment vs. control group I does not deviate reduced significantly more in the treatment than in significantly from 1. For crash scenarios, the null the control group. hypothesis of no change in the number of crashes 184 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia The kilometres travelled on motorways by HGV rather uncommon on motorways apart from registered in Germany were approximately constant construction sites. Additionally, in a survey of HGV in the period under consideration (BAG 2019). The drivers in Germany conducted by Trabert et al. vehicle fleet of HGV increased substantially (2018), the majority of respondents (92%) stated that between 2012 and 2018 with a higher growth rate for they deactivated the system rarely or never. older vehicles (+29%) than for new ones (+10%) However, this result is based on a sample of only 85 (KBA 2012, 2018). However, it is not clear whether drivers. The German Federal Ministry of Transport this translates into a higher proportion of kilometres and Digital Infrastructure plans to ban the possibility travelled for older vehicles. Unfortunately, of deactivating the system in the future BMVI kilometres travelled differentiated by age of vehicle (2020). are not collected on a regular basis. A study During the period under examination, lane departure conducted in 2014 analysed this aspect (Bäumer et warning systems became mandatory. To avoid al. 2017). The authors did not distinguish between confounding, the analysis excluded lane departure road types however. They found that vehicle crashes. There were no other improvements kilometres of new HGV were 1.5 times higher than regarding safety systems in HGV that could have those of older HGV. However, differences in the had an effect on the number of crashes. development of the vehicle fleet only affect the comparison concerning vehicle age not those Overall, the results suggest that the Regulation was regarding crash scenario. Since both variables were associated with a positive effect on the number of combined when calculating the overall effect, rear-end crashes on motorways. Future research is changes in kilometres travelled would have had no needed on the percentage of AEBS equipment in the effect on this value. entire vehicle fleet of HGV and buses. Furthermore, the effectiveness of AEBS on other road types is The analysis period and the treatment group were unknown – although for this kind of analysis, more constructed in a way so that 100% of the vehicles in information on the frequency of system deactivation this group and in this period were equipped with is needed. A comparison with the effectiveness in AEBS. In all other constellations between period other European countries can also be useful. and group, AEBS were not obligatory. However, even before the Regulation, manufacturers equipped 6. CONCLUSION vehicles with AEBS. Thus, in all groups in the reference period as well as in the control groups in In this paper, the effect associated with EU- the analysis period, equipment with AEBS was Regulation No. 347/2012 was analysed. This higher than 0%, but below 100%. Unfortunately, no Regulation made advanced emergency braking data is available on the percentage of HGV and systems (AEBS) mandatory for heavy good vehicles buses not affected by the Regulation that were (HGV) and buses. It came into effect at two approval nevertheless equipped with AEBS. Thus, the effect levels each first for new types of vehicles and two estimated in this study was not the effect that came years later for new vehicles. with equipping the vehicle fleet with AEBS. Rather, This study examined the effect associated with the it was the effect on the number of crashes associated first approval level for new vehicles which was with the introduction of the Regulation and thus the effective since November 2015. The paper increased equipment of the fleet. addressed two research question: (1) How did the Additionally, requirements of approval level 1 were number of rear-end collisions of HGV and buses rather moderate compared to today’s state of develop after the Regulation was introduced? (2) technology. Some vehicles in the treatment group Did the severity of rear-end crashes change after the were already equipped with more efficient AEBS. implementation of the Regulation? The effect estimated in the study is thus at least The analysis used individual police records of HGV partly associated with these more efficient systems. and buses between 2010 and 2018 and additional This should be kept in mind when interpreting the technical vehicle information by the German Federal results. The proportion of vehicles already equipped Motor Transport Authority (KBA). Rear-end with AEBS of approval level 2 or more is unknown. crashes of new vehicles formed the treatment group. Furthermore, the data provides no information if the The number of crashes in this group was compared system was activated during the crash. If a to three control groups: (I) rear-end collisions by substantial number of drivers would have older HGV and buses; (II) other crashes of new deactivated the system, the true effect of the HGV and buses; and (III) other crashes of older Regulation had been larger than the effect estimated HGV and buses. These groups were observed in two in this study. However, in their study, Seiniger et al. periods: a reference period before AEBS became (2020), examined the relevance of false warnings. mandatory in HGV and buses and an analysis period They tested the systems in real traffic over a distance after the introduction of approval level 1. The of 1,300 km and found that false warnings were analysis was limited to crashes on motorways 185 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia because crash situations on motorways are less [5] European Commission (2012a): Commission complex. Due to missing technical vehicle Regulation (EU) No 347/2012 of 16 April 2012 information on foreign vehicles, the analysis was implementing Regulation (EC) No 661/2009 of further limited to vehicles registered in Germany. the European Parliament and of the Council with respect to type-approval requirements for Odds ratios and chi square tests were used to certain categories of motor vehicles with regard estimate the impact associated with the Regulation. to advanced emergency braking systems. The number of crashes in the treatment group [6] European Commission (2012b): Commission developed more positively than in the control Regulation (EU) No 351/2012 of 23 April 2012 groups. The overall effect associated with the implementing Regulation (EC) No 661/2009 of measure was estimated to -37.0%. This effect is the European Parliament and of the Council as significant. Additionally, the results suggest that the regards type-approval requirements for the severity of crashes declined disproportionately in the installation of lane departure warning systems in treatment group. motor vehicles. [7] Jermakian, Jessica S. (2012): Crash avoidance Since the treatment group was narrowly defined (all potential of four large truck technologies. In: vehicles had to be equipped with AEBS, only rear- Accident Analysis & Prevention 49, S. 338–346. end collisions), the magnitude of the effect seems DOI: 10.1016/j.aap.2010.10.033. plausible. It should be mentioned, however, that the effect associated with approval level 1 of the [8] KBA (2012): Fahrzeugzulassungen (FZ). Bestand an Nutzfahrzeugen und Regulation is likely to be slightly overestimated in Kraftfahrzeuganhängern nach technischen the analysis. Requirements of this level were Daten (Größenklassen, Motorisierung, relatively low compared to today’s state of Fahrzeugklassen und Aufbauarten). FZ 25. technology. The treatment group thus most likely Flensburg. also included vehicles that were already equipped [9] KBA (2018): Fahrzeugzulassungen (FZ). with more efficient systems. Nevertheless, the Bestand an Nutzfahrzeugen, Kraftfahrzeugen findings imply that there was a substantial decline in insgesamt und Kraftfahrzeuganhängern nach the number of crashes associated with the technischen Daten (Größenklassen, Regulation which corresponded to an accelerated Motorisierung, Fahrzeugklassen und AEBS equipment of the vehicle fleet. AEBS can Aufbauarten). FZ 25. Flensburg. thus contribute to road safety. [10] Kuehn, Matthias; Hummel, Thomas; Bende, Jenoe (2011): Advanced driver assistance REFERENCES systems for trucks - benefit estimation from real-life accidents. In: Proceedings of the 22nd [1] BAG (2019): Mautstatistik. Jahrestabellen International Technical Conference on 2018. Köln. Enhanced Safety of Vehicles. [2] Bäumer, Marcus; Hautzinger, Heinz; Pfeiffer, [11] Panwinkler, Tobias (2018): Unfallgeschehen Manfred; Stock, Wilfried; Lenz, Barbara; schwerer Güterkraftfahrzeuge. Bremen: Kuhnimhof, Tobias; Köhler, Katja (2017): Fachverlag NW (Berichte der Bundesanstalt für Fahrleistungserhebung 2014 - Straßenwesen M, Mensch und Sicherheit, Heft Inländerfahrleistung. Bremen: Fachverlag NW M 277). (Berichte der Bundesanstalt für Straßenwesen Verkehrstechnik). [12] Petersen, Erwin; Scholze, Carmen; Falke, Christoph (2020): Notbremsassistenzsysteme [3] BMVI (2020): We will prohibit switching off im Lkw. Eine Analyse niedersächsischer advanced emergency braking systems for HGVs Autobahnunfälle in den Jahren 2015 bis 2019 to improve road safety. Online verfügbar unter und der Einfluss aktueller Systeme (Advanced https://www.bmvi.de/SharedDocs/EN/Articles/ Emergency Braking Systems for Trucks – an StV/Roadtraffic/emergency-braking- analysis of truck accidents on Lower Saxonys systems.html, zuletzt geprüft am 15.02.2021. motorways from 2015 to 2019 and the influence [4] European Commission (2007): Directive of actual AEBS) (4), S. 245–256. 2007/46/EC of the European Parliament and of [13] Seiniger, Patrick; Heinl, Felix; Bühne, Jan- the Council of 5 September 2007 establishing a André; Gail, Jost (2020): Lkw- framework for the approval of motor vehicles Notbremsassistenzsysteme. Bremen: and their trailers, and of systems, components Fachverlag NW (Berichte der Bundesanstalt für and separate technical units intended for such Straßenwesen F, Fahrzeugtechnik, Heft F 133). vehicles (Framework Directive). [14] Teoh, Eric R. (2020): Effectiveness of front crash prevention systems in reducing large truck crash rates. 186 Leon Straßgütl YRS 2021 EFFECTIVENESS OF ADVANCED EMERGENCY BREAKING SYSTEMS IN TRUCKS: AN 15. -17. September ANALYSIS OF REAR-END COLLISIONS ON MOTORWAYS IN GERMANY Portoroz, Slovenia [15] Trabert, Tobias; Elio, Iryna; Müller, Gerd; [16] UNECE (2013): Regulation No 131 of the Malczyk, Axel (2018): In-depth Analyse Economic Commission for Europe of the United schwerer Unfälle mit schweren Lkw: Nations (UN/ECE) — Uniform provisions Unfallforschung der Versicherer (Nr. 54). concerning the approval of motor vehicles with Online verfügbar unter regard to the Advanced Emergency Braking https://digital.zlb.de/viewer/metadata/3409258 Systems (AEBS). 5/1/-/. [17] Woodrooffe, John; Blower, Daniel; Flannagan, Carol A. C.; Bogard, Scott E.; Bao, Shan (2013): Effectiveness of a Current Commercial Vehicle Forward Collision Avoidance and Mitigation Systems. In: SAE Technical Papers 9. DOI: 10.4271/2013-01-2394. 187 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Tamás Strommer, junior scientific researcher KTI Institute for Transport Sciences Than Károly Street. 3-5., Budapest, 1119 Hungary strommer.tamas@kti.hu ABSTRACT Subsidizing public transport companies is a worldwide used policy due to the scale economies and low specific travel costs of these kind of transport services. Frequent use of subsidies, however, does not mean uniformity, as there are many methods of subsidizing transport companies. Recent studies has shown that the amount of subsidy applied varies greatly depending on the attributes of the city, service provider and transport mode. The purpose of the present study is to reconsider a model that optimizes fares, and modify the model’s computation method by using new parameters that simplifies the management of crowding costs. The study presents the theoretical background of the topic, the important modifications made in the model, and also demonstrates usability of the model by examining the pricing of the Hungarian interurban transport system. The results recalculated with the modified model showed that crowding cost is a critical component of optimal pricing, as even at a moderate congestion level travel costs are increased by 30–50%. This means, that the optimal fares aimed to reduce the costs of crowding are higher than they were previously set. In case of Hungarian interurban transport, a subsidy rate of around 90% should be applied, with the exception of peak period bus transport, for which the calculated subsidy rate is 74%. Keywords: aggregate analysis, pricing policy, vehicle-size optimization, transit fares optimization, public transport regulations 1. INTRODUCTION the accurate definition of subsidy—a more precise and easily usable formulae of the usual rate—is still The extreme increase in the number, size and missing. The appropriate rate of subsidies and the population of cities and megalopolises raises serious suitable financing method of public transport is questions about their livelihood and sustainability. important, because of these services’ attractiveness These questions can only be answered through and reliability—since both feature is strongly prudent and watchful urban planning, by reaching the dependent on financial resources, and they are some sufficient level of human flexibility (e.g. in housing, of the key aspects of the transportation process. (The mobility), and improving the quality and efficiency substantially unprofitable operation at a social level of public and social services. The transport system can also bring optimal welfare conditions, therefore can be able to compensate the shortcomings of urban subsidies are essential to the operation of design by providing high quality mobility services. transportation companies.) However, based on the The smooth operation of the transport system is a results of recent studies, significant (proportional) cornerstone for the well-functioning cities and differences can be observed between subsidies megalopolises. Since the recognition of this idea, granted to companies providing services in similar greater attention has been focused on researching circumstances (see for example the results of Boss optimal planning and operational strategies for tasks and Rosenschon 2008; Doll and van Essen 2008; of transport planning and management. Based on the Parry and Small 2009; Tscharaktschiew and Hirte commonly used microeconomical approach, travel 2011). Although it is hard to draw far-reaching pricing is a well-established and frequently used conclusions because of the variance between the method to influence traveller decisions. compared systems’ and countries’ transport policy, economic situation, and the limits of subsidization, Our study examines the optimal subsidy of public the topic of subsidies given to transportation transport operators. Although the daily operation of operators functioning in similar circumstances is still these companies are often supported by public funds, worth examining. 188 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia The basis of our research was the computational 2.1. Scale economies of scheduled model presented by Parry and Small (2009), which transport services investigates the (ideal) pricing of travel. The model can be used to determine the subsidy rate and the In the case of public transport systems operating on a expected operational-economic characteristics frequent basis and providing scheduled services, the belonging to it under optimal conditions. The aim of phenomenon of scale economies prevails (Mohring our study is to extend this model with parameters 1972, 1976; Turvey and Mohring 1975). The describing crowding in public transport, and also the additional operation costs of passenger services at application of the model for a new context, the higher passenger numbers, under idealized Hungarian interurban public transport system. The conditions, are balanced by the benefits of passenger mode and purpose of adding new parameters to the time savings through the reduction of the average model is to make the model able to handle both new waiting time, i.e. the net marginal social cost of travel and already-built-in parameters in the same is lower than the marginal cost of operation. In other environment, and thus be able to account for more words, the marginal cost of using a service that detailed and accurate results of the effects of operates under scale economies is always lower than crowding. (Nowadays crowding is studied from the average cost, and therefore the optimal price, many aspects from engineering and economics to which is equal to the marginal cost, leads to a loss psychology and ergonomics). that must be compensated with subsidy. The paper is structured as follows: first the paper Regarding the optimization of the social costs, the deals with the optimal pricing of public transport— sum of operating and user costs, by the means of the most influential factors of the calculation, and the service frequency, we obtain simple relation that the basics of the model. The following sections introduce optimal frequency—number of vehicles departing the needs that call for the modifications of the model, per unit of time—increases proportionally with the the use of crowding parameters and the further use of square root of the number of passengers, assuming a optimized parameters. Then the paper presents the proportional relationship of vehicle amount and basic data for the calculation of interurban traffic in operating costs, and that passengers arrive to stops at Hungary and also the application of the model for the random (Mohring 1972). examined region. Finally we summarize all the In addition, in public transport, due to fixed operating results and conclusions that can be drawn from the costs and waiting times, scale economies can also be results of the model and we suggest some further observed in operating and user costs. Since both lead improvements. to subsidies, Mohring’s model requires that the service should be subsidized even if the economies of 2. OPTIMAL PRICING OF PUBLIC scale in operation do not materialize, as the scale TRANSPORT economy still applies to user costs. The benefits of public transport and the positive 2.2. Marginal cost pricing effects (e.g. lower local emissions, moderate fuel consumption and land use) are most dominant in Where user cost is a non-negligible part of social networks with significant traffic and near cost, as in the case of scheduled public transport bottlenecks. Even when utilizing the benefits of services, the marginal cost based pricing should public transport, it is advisable to strive for an include two types of costs at the optimum, the service optimal operational situation and strategy, where the provider’s and the passengers’ costs (Jansson 1979; benefits of the daily use are not outweighed by the Vickrey 1980). This duality can be observed both in inherent negative phenomenon. All systems should the amount of vehicles on the road, proportional with avoid the state, when: user costs and the size of the service area, and the size or capacity of the vehicles, proportional with the  the presence of public transport vehicles cause costs of the service provider. a major congestion in the road traffic of the city (or even cause significant delays of the The uniqueness of the transport market is that the public transport service itself), consumer (passenger) is also involved in the process  the level of crowding on public transport of producing the product, since the locomotion only can be performed by the passenger’s time spent. It is vehicles grow so high, that it causes a major decrease of utility, or when precisely because of passenger involvement that the  right approach is to escape the notion that the only the contamination of vehicles and the decrease of passenger safety cannot be avoided (see the costs which are relevant to optimization are those of results of Perone 2002 on passenger the transport operator. The time-costs of the preferences and mode choices). passengers must be included too, and fares must be equated with marginal social costs. (Turvey and Mohring 1975) The significant change brought by the proliferation of information and communication 189 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia technologies (ICTs) also has to be highlighted, as it offices, etc., or censuses, surveys have already done has made multitasking a realistic option during travel these researches—which also simplifies the (though learning, working, various communication collection of data and thus the investigation and forms etc., see Keserű et al. 2015; Keserű and modelling, too. Macharis 2018; Munkácsy, Keserű, and Siska n.d.). One of the model’s most important methodological This change made travel time almost as valuable as feature is that all parameters are derived to vehicle- other activities—in an appropriate travel miles and the calculations are defined by these kind environment—the value of passengers’ travel time of parameters, too. The key components of the model savings could drop significantly (International can be calculated using the following parameters: Transport Forum 2018).  user benefits (consumption of numeraire good 2.3. Overview of the model of Parry and and the utility of travelling); Small  vehicle occupancy, service frequency, One of the significant works of the recent years on crowding and travel time; the field of transport economics, subsidization and  external pollution and accident costs; pricing policy was the paper of Parry and Small  household budget constraint, the balance of (2009). The paper highlights the importance of the income and expenditure; question whether the subsidy of public transport  ways in which companies adapt the companies needs to be reduced (or even abolished), constraints of the service. or increasing subsidy level brings the system closer to the optimal operating conditions. The model calculates values of costs and benefits and then aggregates these results using passenger-miles The authors stated that despite the differences “travelled” in the system—passenger-miles per travel between the subsidy rates of transport companies, mode and time periods. Similarly, in-vehicle travel there is no generally accepted, practicable calculation time (𝑇), waiting time (𝑊), access time (𝐴), and method that can be used to determine the ideal crowding measure (𝐶) for the extra cost of vehicle- financial strategy for a given transport operator, i.e. crowding can be calculated as the product of specific the rate of subsidy. In most cases, only complex, and values multiplied by the number of passenger-miles therefore location-specific simulation models are travelled during the time period under investigation, able to provide data on traffic flow. Since pricing calculated for each transport modes. A combination policies are often based on these results of modelling of these factors can be used to produce the it is difficult to determine whether public transport generalized (non-money) cost of travel, which is pricing fits well with a city or region’s transport proportional to the time spent traveling: 𝛤 = system. 𝛤(𝑇, 𝑊, 𝐴, 𝐶). The 𝛤 function establishes the The aim of the authors was to create an analytical relationship between the costs received in time model that is simple enough to use without highly dimension. specific data, i.e. it requires data that are typically The way to determine utility is to sum all the available at all transport providers, statistical officest components in cost dimensions, which can help etc., and at the same time the model can manage the balancing the benefits and losses. According to the most important parameters describing the transport model, the user preferences and the benefits of their system. It was also targeted that the calculation activities (𝑈) can be modeled as the difference method could be applicable to transport systems of between the value of the utility function—based on significantly different size, structure, operation the consumption of numeraire good (𝑋), the sub method, and vehicle occupancy, as well as to cities utility from passenger miles travelled (𝑀), and the and countries with different levels of development. generalized (non-money) cost of travel—and factor 𝑍, which takes the magnitude of pollution 2.3.1. Model structure externalities and accidents costs into account. In The model was built using aggregated data that addition to the utility function, the balance of the incorporate peak and off-peak results of all transport representative household must be determined. This modes. The advantage of this model structure is that approach assumes that all households spend the tax- it can replace network models significant in size and deductible portion of their income on traveling and computational supplies, their need for detailed consuming numeraire good. calibration, the necessarily integrated decision The indirect utility function of households is methods of traffic models etc.. It is important to note calculated as the value of the maximized utility that aggregate data are sufficient for this level of function within the budget constraints, see the first analysis, and it is neither necessary nor advisable to line of Equation (1), but an indirect utility function of subdivide the data, as in many cases aggregated a user (passenger) can also be used to measure social results can be obtained from local transport benefits by replacing the transit agency’s budget companies, road transport organizations, statistical constraint. Then, the amount of tax (𝑇𝐴𝑋) expressed 190 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia as a function of operating costs and ticket revenue can lower peak-period rail travel fees on bus and off- be inserted in the second expression of Equation (1). peak-period rail travel characteristics (via scale In which the sum of operating costs (𝑂𝐶) can also be economies, appearance of new passenger, crowding divided into fixed and variable (service-dependent on vehicles etc.). and independent) costs. 2.3.3. Results of previous work 𝑈 ̃ = 𝑢̃({𝑝𝑖𝑗, 𝑡𝑖𝑗, 𝑤𝑖𝑗, 𝑎𝑖𝑗, 𝑐𝑖𝑗}, 𝑇𝐴𝑋) − 𝑍 = 𝑢[𝑋, 𝑀, Γ(𝑇, W, A, C)] + Parry and Small examined the transport systems of max { } = 𝑋,𝑀,Γ +𝜆 ∙ (𝐼 − 𝑇𝐴𝑋 − 𝑋 − ∑ 𝑝𝑖𝑗𝑀𝑖𝑗 𝑖𝑗 ) three cities having drastically different transportation 𝑢[𝑋, 𝑀, Γ(𝑇, W, A, C)] + systems (London, Washington DC, and Los max { } 𝑋,𝑀,Γ +𝜆 ∙ (𝐼 + ∑ 𝜏𝑖𝐻𝑀𝑖𝐻 𝑖 − ∑ 𝑂𝐶𝑖𝑗 𝑖𝑗≠𝑖𝐻 − 𝑋 − ∑ 𝑝𝑖𝐻𝑀𝑖𝐻 𝑖 ) Angeles). The subsidy rates for each city were (1) arbitrarily modified by time periods and transport modes, because a one time intervention in the model As it can be seen from the presented model structure, would have caused unrealistic change in service one can analyse the journeys by transport mode and levels, as well as users’ expected decisions and mode travel time period. This makes easier to examine the journeys’ choices. components in detail, but because of the aggregate level an optimal strategy can also be The paper’s results show that in most cases determined based upon the same data. significant rates of subsidies are recommended for transport agencies. In 11 out of 12 cases, the optimal 2.3.2. Formulas derived from the model value of subsidy is more than two-thirds of the operating cost, and in more than half of the cases it The net marginal social cost of each trip is obtained reaches 90%, while in one case the model suggested by totally differentiating the indirect utility function a minor reduction. The model proves that new of Equation (1) with – 𝑝. The extreme is where passengers appearing would have a negative impact marginal reduction or increase of fares no longer on utility due to the difference between average and increases or decreases social welfare. For example, marginal cost. The results of the model also show that one can determine the extremes of the indirect utility “revenues” are primarily come from scale economies function for peak-period rail travels by differentiating and decreasing externalities. Equation (1) with – 𝑝𝑃𝑅. The result is Equation (2), which shows how the value of utility changes for The model predicts a major increase in passenger each component as a result of a single reduction in miles, more than 50% increase in off-peak periods. the travel fees of peak-period rail transport. This result is perfectly rational, since it is particularly practical to facilitate the best use of capital (i.e. fleet 𝑀𝑊𝑃𝑅 ≡ −(𝑀𝐶𝑃𝑅 𝑃𝑅) 𝑠𝑢𝑝𝑝𝑙𝑦 − 𝑝𝑃𝑅)(−𝑀𝑃𝑅 + of vehicles, stations and infrastructure) in less +(𝑀𝐵𝑃𝑅 𝑃𝑅 𝑃𝑅) 𝑖𝑗 𝑖𝑗 𝑠𝑐𝑎𝑙𝑒 − 𝑀𝐶𝑜𝑐𝑐)(−𝑀𝑃𝑅 + ∑𝑖𝑗=𝑃𝑅,𝑖𝐻(𝑀𝐶 𝑀 ) + 𝑒𝑥𝑡 𝑃𝑅 frequented periods. With significantly lower off-peak + ∑𝑖𝑗=𝑂𝑅,𝑃𝐵,𝑂𝐵(𝑀𝐶𝑖𝑗 + 𝑀𝐶𝑖𝑗 + 𝑀𝐶𝑖𝑗 − 𝑀𝐵𝑖𝑗 − 𝑠𝑢𝑝𝑝𝑙𝑦 𝑒𝑥𝑡 𝑜𝑐𝑐 𝑠𝑐𝑎𝑙𝑒 period transit fares and due to greater spatial and 𝑝𝑖𝑗) 𝑀𝑖𝑗 temporal coverage, on account of the Mohring effect, 𝑃𝑅 more passengers will choose public transport. (2) Based on the examples of the examined cities, it can In Equation (2) the first component (𝑀𝐶𝑃𝑅 𝑠𝑢𝑝𝑝𝑙𝑦) be concluded that in such cases public transport reduces social welfare, since the greater the should be given a decisive role, and the use of public difference between the cost of transporting a transport should be significantly supported. passenger and the price of a ticket, greater losses the However, it is worth emphasizing that this is a transport agency accumulates. The effect of the secondary optimum. Not only in terms of pricing, as transport system’s scale economies impacts through this model ignores the expected impact of other the difference between the marginal benefit of scale actions taken (e.g. infrastructure or vehicle economies (𝑀𝐵𝑃𝑅 𝑠𝑐𝑎𝑙𝑒) and the marginal cost of development). The primary optimum could be congestion (𝑀𝐶𝑖𝑗 𝑜𝑐𝑐). When the Mohring effect achieved through appropriate pricing of car use, but applies, a marginal reduction of fares has a positive this kind of pricing is deliberately omitted by the outcome, since the operator increases the service study. frequency, thereby reduces the waiting time of all the other passengers. In contrast, crowding has an 3. EXAMINATION OF THE opposite effect. With the marginal reduction of fares CROWDING ELASTICITY the number of passengers will increase, and in Determining the values of the parameters built in the parallel the increase of crowding will cause losses of model, Parry and Small (2009) made a suspicious well-being and social benefits. The remaining finding that the proportion of congestion losses are components of Equation (2) expresses the benefits of 𝑖𝑗 “negligible” compared to other factors. However, changing externalities (𝑀𝐶𝑒𝑥𝑡) due to the expected studies over the last 10 years have shown that reduction in the number of cars, and the impact of perceived travel costs can rise by around 50%, or 191 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia even more according to some measures at a crowding intensity of crowding by optimizing vehicle size (𝑙𝑖𝑗) level of 3 passengers/m2. In addition, the original and congestion level (𝑜𝑖𝑗). These optimization model used a nonlinear relationship between conditions take the following form (see Parry and crowding and value of time, which is an inappropriate Small 2009, p. 708): simplification based on the results of the relevant 𝑖𝑗 𝑖𝑗 literature. (Björklund and Swärdh 2015; Hörcher, 𝜚𝑊𝑤𝑖𝑗𝜂𝑤 = 𝜚𝐴𝑎𝑖𝑗𝜂𝑎 ,, (3) Graham, and Anderson 2017; Kroes et al. 2014; 𝜚𝐶𝑐𝑖𝑗𝜂𝑖𝑗𝑜𝑖𝑗 = 𝑡𝑖𝑗𝑘𝑖𝑗𝑛𝑖𝑗, (4) Tirachini et al. 2016; Whelan and Crockett 2009) 𝑐 2 where the factors 𝜚𝑘 express the marginal (monetary) It is difficult to determine the actual characteristics of cost of each type of loss, and 𝜂𝑘 their elasticity (𝑘 = crowding, therefore our approach trace it back to the 𝑊, 𝐴, 𝐶). travel time. When supplemented by a multiplier term the effect of crowding can be calculated with the These equations can be turned into the relation that formula 𝑐𝑖𝑗 = 𝑡𝑖𝑗𝑚(𝑙𝑖𝑗). Since the relationship route density can be increased as long as the between vehicle occupancy and crowding factor can decreasing costs of access, through the fixed quantity already be approximated by a linear function (see the of passenger miles can balance the effect of the results of the above-mentioned literature), a decreasing service frequency, and thus the increasing multiplier function of the form 𝑚(𝑙) = 𝛼𝑐 ∙ 𝑙 has waiting costs. Similarly, the increase in vehicle size been used, where 𝑙 expresses the degree of vehicle pays off as long as the gains from the reduction of occupancy. We assume that the level of crowding crowding can cover the rising operating costs. Based does not affect travel time (i.e. neglecting the on these relationships a generalized user cost gas relationship between the crowding level and the been expressed that can summarize all factors as a aligning time). Thus the elasticity of 𝛼 function of (travel) time: 𝑐 depends solely on the shape of the multiplier function, which 𝑞𝑖𝑗 = 𝑝𝑖𝑗 + 𝜚𝑇𝑡𝑖𝑗 + 𝜚𝑊𝑤𝑖𝑗(1 + 𝜂𝑖𝑗 𝜂𝑖𝑗 ⁄ ) + in this case, due to the linear relationship 𝜂 𝑤 𝑎 𝑐 = 𝑖𝑗 𝑖𝑗 (𝜕𝑚(𝑙)⁄𝜕𝑙) ∙ (𝑙⁄𝑚(𝑙)) = 1 (see the results of the 𝑡𝑖𝑗𝑘 𝑛𝑖𝑗 𝜂 𝑜𝑖𝑗 ⁄ 2 𝑐 . (5) relevant measures (Björklund and Swärdh 2015; In the next chapter, we introduce the suggested Hörcher et al. 2017; Kroes et al. 2014; Tirachini et al. modifications and changes in the aforementioned 2016; Whelan and Crockett 2009). relationships and optimization parameters. The model of Parry and Small (2009) takes only the most important parameters of the transport system 4. OPTIMIZATION PROBLEMS into account, but the model is still simpler, and uses Crowding-related costs are difficult to measure and less data than the detailed traffic demand models. The quantify in the way it was used in the model of Parry analytical relationships of the model counts for and Small. Based on assumptions confirmed by the different modes of traffic (rail, bus, and private car), literature we have modified and extended the original and uses a simple method of welfare optimization model using a linear multiplier function. These while maintaining economic equilibrium. As well as parameters and formulas measuring crowding can be the optimum depends on the costs of travel, it relies used to optimize vehicle size and determine optimal on the consumption of the numeraire good, the fares. In this chapter we will present these questions, passenger miles travelled in the whole system, and and also the related modifications in details. the model also calculates the effects of pollution, accidents and other externalities. 4.1. Vehicle-size optimization The model traces back cost variables (waiting, In the initial model, similarly to the other parameters, access, crowding) to travel characteristics, such as crowding is defined as a specific parameter of waiting time to service frequency, accessibility to distance travelled, i.e. passenger miles. Despite the route density, and it gives the crowding costs as a simplicity of this approach, in our opinion, the effects function of load factor. The model takes travel time of crowding should rather be compared to travel time, patterns into account specialized for every mode of since the extent of profit loss does not depend on the transport. distance travelled, but rather on the duration of discomfort. By the time of Parry and Small’s research there were no available measurement with valid results to An important simplification of the original model is quantify the access costs and the effects of crowding that the effects of crowding can be determined as a on vehicles, therefore the paper replaced this prerequisite for optimizing vehicle size with the indicators with some simplifications. The applied balance of operating and crowding costs. According assumption was that the service provider (travel to this assumption, in order to increase passenger agency) adjusts the vehicle route density (𝐷) and miles, the service provider must optimize the size of service frequency (𝑓𝑖𝑗) according to the changes of vehicles in such a way, that the resulting passenger-travel demand, while responds to the changing side benefits could compensate the increasing operating costs of larger vehicles. This relation would 192 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia be a very practical solution, but its feasibility is 𝜚𝑇𝑡𝛼𝑐𝑚𝑙 = 𝑡𝑘2 highly questionable: it is not possible to react to the 𝜚𝐶𝑐 = 𝜚𝑇𝑡𝑚 = 𝑡𝑘 ⁄ . (13) constant changes in passenger traffic by continuously 2 𝑙 changing the vehicle fleet, or intervening flexibly at Naturally the formula of generalized cost can be used a fixed infrastructure (e.g. by enlarging the stations with the result relationship from both types of of metro networks). Based on this consideration, it is derivation (only if a linear multiplier function is advisable to abandon this equation and approach the used): question from the viewpoint of travel time. 𝑞𝑖𝑗 = 𝑝𝑖𝑗 + 𝜚𝑇𝑡𝑖𝑗 + 𝜚𝑊𝑤𝑖𝑗 + 𝜚𝐴𝑎𝑖𝑗 + 𝜚𝐶𝑐𝑖𝑗 = 𝑝𝑖𝑗 + The cost function was used by Parry and Small 𝜚𝑇𝑡𝑖𝑗(1 + 𝑚) + 𝜚𝑊𝑤𝑖𝑗(1 + 𝜂𝑖𝑗 𝑖𝑗 𝑤 𝜂 ⁄ 𝑎 ) = 𝑝𝑖𝑗 + 𝜚𝑇𝑡𝑖𝑗 + (2009) in the following form: 𝜚𝑊𝑤𝑖𝑗(1 + 𝜂𝑖𝑗 𝑖𝑗 ⁄ 𝑤 𝜂 ⁄ 𝑎 ) + 𝑡𝑘2 𝑙. (14) Γ(∑ 𝑡𝑖𝑗𝑀𝑖𝑗 𝑖𝑗 , ∑ 𝑤𝑖𝑗𝑀𝑖𝑗 𝑖𝑗≠𝑖𝐻 , ∑ 𝑎𝑖𝑗𝑀𝑖𝑗 𝑖𝑗≠𝑖𝐻 , ∑ 𝑐𝑖𝑗𝑀𝑖𝑗 𝑖𝑗≠𝑖𝐻 ) It must be mentioned that a tractable formula can be (6) derived for the costs of crowding, which directly However, after substituting the term 𝑐𝑖𝑗 = 𝑡𝑖𝑗𝑚(𝑙𝑖𝑗), gives the cost of crowding using only the travel time only the parameters in time dimension will remain: and the parameters 𝑚 or 𝑙 (latter is the widely used load factor). Γ (∑𝑖𝑗 𝑡𝑖𝑗 (1 + 𝑚(𝑙𝑖𝑗)) , ∑ 𝑤𝑖𝑗𝑀𝑖𝑗 𝑖𝑗≠𝑖𝐻 , ∑ 𝑎𝑖𝑗𝑀𝑖𝑗 𝑖𝑗≠𝑖𝐻 ) 4.2. Determining the optimal transit fares (7) The other formula where crowding plays a significant In Equation (7) the measure of congestion is already role is the one to determine optimal fares and subsidy time-related, hence the marginal monetary cost of rates. In this case it is necessary to examine where the travel time equals crowding’s (𝜚𝐶 = 𝜚𝑇). If we derive extreme value of the indirect utility function is, where the utility function supplemented with the crowding the equation 𝜕𝑈 ̃ 𝜕𝑝𝑃𝑅 ⁄ = 0 is met. expression in a similar way to the term covering all travel costs expressed in Equation (5), and we derive In addition to vehicle size optimization, crowding the optimal vehicle size from this transformed utility parameters also play important role in the marginal function, then the maximum of the indirect utility welfare formula’s (A6) CROWD + VEHSIZE function is as follows: component. This sum contains the formula of the marginal welfare’s crowding-dependent part, and the 𝜕𝑈 ̃ 𝜕𝑚 𝜕𝑙 d𝐾 0 = = −𝜆𝜚𝑇𝑡𝑀 − 𝜆𝑉𝑡 . (8) 𝜕𝑛 𝜕𝑙 𝜕𝑛 d𝑛 equation derived from the tax-relationship (d𝑇𝐴𝑋 d𝑝𝑃𝑅 ⁄ ), which is an indicator of the transport By substituting 𝜕𝑙⁄𝜕𝑛 = − 𝑙⁄𝑛 and d𝐾⁄d𝑛 = 𝑘2in agency’s operating cost, and therefore also depends Equation (8), we can get a simplified form: on the vehicle size: 𝜕𝑚 𝑙 0 = −𝜆𝜚𝑇𝑡𝑀 − 𝜆𝑉𝑡𝑘 𝜕𝑙 𝑛 2. (9) ∑ d𝑛𝑖𝑗 𝑖𝑗≠𝑖𝐻 [𝜚𝐶 𝑑𝑐𝑖𝑗 𝑀𝑖𝑗 + 𝑡𝑖𝑗𝑉𝑖𝑗 𝑘 ]. (15) 𝑑𝑝𝑃𝑅 2 d𝑝𝑃𝑅 According to the formula (A7a) given in the appendix of Parry and Small (2009), the elasticity of crowding The further transformations of the equation require cost is: the relationships for elasticity that can be found in the appendix of Parry and Small (2009). Using these 𝜕𝑐𝑖𝑗 𝑙 𝑡 𝜕𝑚(𝑙) 𝑙 𝜕𝑚(𝑙) 𝑙 𝜕𝑚 𝑙 𝜂𝑖𝑗 formulae, Equation (15) can be transformed: 𝑐 = = = = . (10) 𝜕𝑙𝑖𝑗 𝑐 𝜕𝑙 𝑡 𝑚(𝑙) 𝜕𝑙 𝑚(𝑙) 𝜕𝑙 𝑚 If we use Equation (8) and (10) together, we get: = ∑ 𝑛𝑡𝑘2 ( 𝑖𝑗≠𝑖𝐻 1 − 𝜀 . (16) 𝑜 𝑉) d𝑀 d𝑝𝑃𝑅 0 = −𝜆𝜚𝑇𝑡𝑀𝜂𝑖𝑗 1 From this equation, the form of the crowding factor 𝑐 𝑚 − 𝜆𝑉𝑡𝑘 𝑛 2. (11) (marginal cost of increased vehicle occupancy) is Rearranging Equation (11), dividing both sides by already apparent: 𝜆𝑀, then rearranging it again to get 𝜚𝑇𝑡𝑚, which is 𝑖𝑗 𝑛𝑡𝑘2 practically speaking equals with 𝜚𝐶𝑐, one can get the 𝑀𝐶 ( 𝑜𝑐𝑐 = 1 − 𝜀 𝑜 𝑉). (17) formula that can be substituted in Equation (5) of the If we approach the question from the viewpoint of generalized costs. crowding multiplier, we get a similar result, but the 𝜚𝐶𝑐 = 𝜚𝑇𝑡𝑚 = 𝑛𝑡𝑘2⁄𝜂𝑐𝑜. (12) parameters are divided into several other factors. At first, with only the CROWD component counted in, Actually, we can express the result by using other we will get the terms, if we use the multiplier function 𝑚(𝑙) = 𝛼𝑐 ∙ 𝑙 directly, since: 𝜕𝑡 𝜕𝑚 𝜚𝐶 d𝑐 𝑀 = 𝜚𝑇𝑚 𝑀 + 𝜚𝑇𝑡 𝑀 (18) d𝑝𝑃𝑅 𝜕𝑝𝑃𝑅 𝜕𝑝𝑃𝑅 𝜕𝑚 𝑙 𝜆𝜚𝑇𝑡𝑀 = 𝜆𝑉𝑡𝑘 𝜕𝑙 𝑛 2 equation, where the first term represents the 𝜆𝜚𝑇𝑡𝑀𝛼 crowding costs due to variable travel time, which, 𝑐𝑙⁄𝑛 = 𝜆𝑉𝑡𝑘2 due to the similarity can be linked to the USERTIM 𝜚𝑇𝑡𝛼𝑐𝑙𝑜⁄𝑛 = 𝑡𝑘2 designation in the marginal utility equation of the 193 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia model. Thus, the crowding multiplier also appears in 5. APPLICATION OF THE MODEL IN the relationship of marginal congestion costs A HUNGARIAN INTERURBAN 𝑀𝐶𝑖𝐻 𝑖𝑘 𝑖𝐵 ENVIRONMENT 𝑐𝑜𝑛𝑔 = ∑ 𝑡𝐻 𝜚𝑇(1 + 𝑚)𝑀𝑖𝑘 𝑘=𝐻,𝐵 + 𝑡𝐻 𝐾𝑖𝐵𝑉𝑖𝐵, (19) The uniqueness of the Hungarian settlement system that can later be used to calculate the marginal is its strong capitalization, the effect of which can be congestion costs of bus transport: 𝑀𝐶𝑖𝐵 obserbed both in the structure of the transport 𝑐𝑜𝑛𝑔 = 𝛼 𝑖𝐻 network and also in the main connections of transport 𝐵𝑀𝐶𝑐𝑜𝑛𝑔. services. Due to the country’s radially structured The second component of the sum expresses that the transport network, the lack of transverse connections, change of fares also has an impact on the crowding and the poor permeability of the Danube River— multiplier; the equation can be further adjusted as which divides the country in halves—the vast follows: majority of road and rail traffic flows directly 𝜕𝑚 𝜕𝑙 𝜕𝑜 d𝑀 through, or in the immediate vicinity of the capital. 𝜚𝑇𝑡 𝑀 = 𝜚𝑇𝑡𝛼 𝑀. (20) 𝜕𝑝𝑃𝑅 𝑐 𝜕o 𝜕M d𝑝𝑃𝑅 This effect puts a huge weight on the road and rail network of Central Hungary. One quarter of the By substituting the modified Equation (20) in the country’s population lives in Budapest and its CROWD + VEHSIZE component used in Equation metropolitan area (Hungary had a population of 9.77 (15), the derivation can be continued: million in 2019). Beside the complex and ∑ 𝜕𝑚 𝜕𝑛 overcongested transportation system of Budapest and 𝑖𝑗≠𝑖𝐻 [𝜚𝑇𝑡 𝑀 + 𝑡𝑉𝑘 ] = 𝜕𝑝𝑃𝑅 2 𝜕𝑝𝑃𝑅 its agglomeration, the biggest traffic appears on the ∑ 𝜕𝑙 𝜕𝑜 d𝑀 𝜕𝑛 𝜕𝑜 𝜕𝑀 𝑖𝑗≠𝑖𝐻 [𝜚𝑇𝑡𝛼𝑐 𝑀 + 𝑡𝑉𝑘 ]. 𝜕o 𝜕M d𝑝𝑃𝑅 2 𝜕𝑜 𝜕𝑀 𝜕𝑝𝑃𝑅 roads and rails between the central region of Hungary and the bigger cities of the country. Although of (21) lesser importance in terms of traffic volume, railway All the partial derivatives, 𝜕𝑙/𝜕o = (1 − 𝜀𝑛) ∙ 𝑙⁄𝑜, lines and bus routes serving rural settlements also 𝜕𝑛/𝜕o = 𝜀𝑛 ∙ 𝑛⁄𝑜 and 𝜕𝑜/𝜕M = (1 − 𝜀𝑉) ∙ 𝑜⁄𝑀 play an important role in the country's transport can be substituted in Equation (21), we get a quite system. compact relationship for the marginal welfare effects: An important feature of public transport in Hungary 𝜕𝑙 𝜕𝑛 𝜕𝑜 d𝑀 = ∑ is that state-owned service providers still dominate 𝑖𝑗≠𝑖𝐻 [(𝜚𝑇𝛼𝑐 𝑀 + 𝑉𝑘 ) 𝑡 ] = 𝜕o 2 𝜕𝑜 𝜕M d𝑝𝑃𝑅 the market. The Hungarian state-owned company ∑ d𝑀 𝑖𝑗≠𝑖𝐻 [(𝜚𝑇𝛼𝑐(1 − 𝜀𝑛)𝑙 + 𝑘2𝜀𝑛𝑛⁄𝑜)𝑡(1 − 𝜀𝑉) ∙ ]. 𝑑𝑝𝑃𝑅 group, the MÁV-Volán Group’s subsidiaries are (22) responsible for the passenger transport on rail and for the interurban bus transit—and in some cities for the It is important to emphasize that in order to carry out local transport service also, on the basis of a contract the optimization, it is necessary to determine the between the municipality and the company. The occupancy of the vehicles, i.e. the load factor (𝑙). MÁV-Volán Group has an important role to play in Since the crowding level is often expressed by providing public transport in cities and metropolitan fraction of the number of passengers and the vehicle’s (floor) size, it is necessary to determine this areas, between the regions of the country, and as well as in connecting villages, towns and cities to the quotient in order to the further use of these transportation system. parameters. Accordingly, the quotient of the average number of passengers and the average surface area of The examination of the subsidization issue is the vehicles (calculated for the vehicles of each city), complicated by the fact that in the European Union, or even the formula (𝑚𝑙 = 𝑘2 𝜚𝑇 ⁄ = 𝛼𝑐𝑙2) retrieved hence in Hungary also, transport service providers from Equation (13) can give us the followings: often operate on a regulated market within the framework of public service. In a regulated market 𝑙 = √𝑘2/(𝜚𝑇𝛼𝑐). (23) like this, there is competition when one comes to Applying one of the two ways of approach entering the market, but companies already in the (calculation of vehicle capacity for the whole vehicle market are providing the service on a yet exclusive fleet, or using Equation (27) (23) with the parameter basis. In addition, the state, the procurer of the values defined in the model) can determine the transport services may impose a public service required values of vehicle occupancy and the load obligation (according to Regulation (EC) No factor, so the value of the crowding factor can also be 1370/2007), but in this case the transport company’s calculated. burdens resulting from this service has to be balanced. (Jászberényi and Pálfalvi 2009) In other words, it is a practical solution for a state to sustain public transport services through subsidization, but this method of operation also fixes the dependence of the service providers on state resources. 194 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia Table 1: Parameter values used by the calculations for the Hungarian interurban transport system Hungary Parameters Unit Source of data Rail Bus Median wage rate 7340 7340 HUF/hour KSH, 2017. Number of unlinked trips 146.9 490.6 millions/year KSH, 2017. Annual passenger miles 7366 7397 millions ITM, 2017. Annual rail car / bus miles 82 367 millions ITM, 2017. Fleet size 1677 5239 – ITM, 2017. Transit speed 51.9 35.5 km/h MÁV-Start, 2016., KTI webpages of vehicle Purchase cost of rail car or bus 350 62 million HUF manufacturers Total operating cost 243 161 billion HUF ITM, 2017. Total fare revenues 039 57 billion HUF ITM, 2017. Road transport Parameters Unit Source of data Peak Off-peak Annual vehicle miles 82.2 million/year (Magyar Közút 2018) Average trip length 41.1 km KTI, 2016. Fuel tax 228.0 HUF/liter estimation previous results of Hungarian Occupancy 1.32 1.32 pass/vehicle measurements (from 2016) Auto average speed 50 60 km/h estimation Fuel efficiency 6 l/100 km estimation Among the European Union member states Hungary does not include a detailed analysis of the domestic has one of the highest rates of public bus transport circumstances and characteristics of the local (Statistical Office of the European Communities transport system. For the purpose of calculation and 2019), a kind of service that is available at all the correct interpretation of results, it should be also settlement. This level of coverage is probably caused noted that the data used for the calculations mainly by the fact, that the motorization level is well below describe the characteristics of the interurban transport the European average, and the fares are cheaper system. These data also describe all settlements compared to other countries’. Moreover, the where the service was ran by state-owned bus and Hungarian State provides travel discounts to several train operators. social groups. For example, citizens under the age of 6 and over 65 can use almost all kind of public 6. RESULTS OF THE CALCULATION transport for free, and people with a student card also The results of the calculations are discussed in two receive a 50% discount on their travel. sections, differentiating the effects of modifying the Although this paper primarily deals with the crowding formulas and the results of the model’s development of methodology of calculating optimal application for the transport system of Hungary. public transport subsidies, the Hungarian adaptation 6.1. Adaptation of crowding relationships of the modified calculation method was also an important element of the research. The application of Reconsidering the relationships and formulas used in the methodology for Hungary has been carried out on the model was a practical decision. The original study a national level, primarily in order to ensure that the stated that the costs of congestion were “relatively local characteristics of the large-scale transport small”, and since then a couple of measurements system do not significantly influence the results of found the exact opposite, i.e. that the economic losses the calculations. caused by crowding cannot be neglected, so a revision of the crowding relationships and the model By the time of our research, the operating and budget was highly appropriate. After the modifications, the data of the national service providers were available costs of crowding can be included as a stand-alone for the years 2016/17, so we were able to perform element of travel decisions, rather than a proportion domestic calculations with quite recent data. To of other travel-time components. As a result, the describe the operating conditions of the Hungarian previously presented crowding parameters also transport system we used the parameters and the values shown in Table 1, the other parameter values appears in the 𝑀𝑊 formula (see Eq. (2)), creating a direct relationship between user costs and objective were the same as in the original model. We used primarily the data provided by the Ministry of congestion metrics. Innovation and Technology, transport service The results obtained by the model using the modified providers, the Central Statistical Office (KSH) and cost and crowding formulas are summarized in Table the Institute of Transport Science (KTI). 2. Compared to the results of Parry and Small (2009), it can be seen that the discomfort caused by crowding It is important to emphasize that the Hungarian already accounts for a significant proportion of travel adaptation mainly remains on theoretical level, and 195 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia Table 2: Results of modelling with modified cost and congestion relationships costs (+20–60% increase), even at moderate 6.2. Results of the model’s application in congestion levels. It can also be observed that a Hungary significant increase in crowding-related (marginal) costs is to be expected, regardless of city, transport With the use of the presented parameter values (Table mode and travel period. The growing number of 1), the results of the optimization show that a subsidy travels in the off-peak period may be more likely to rate of around 90% should be applied at almost all of be caused by the “replenishment” of crowding costs the cases. There is an exception of the peak period that were completely neglected before. The results bus transit, where a subsidy rate of 74% is show that marginal welfare gains, optimal subsidies recommended. This is 2 percentage points lower than and the expected passenger number are reduced in the applied value of subsidy.) The forecast for change most of the studied cases. The direction (and of passenger miles are in line with the change of magnitude) of the subsidy change is in line with subsidy rate. It predicts lower growth in the rail previous expectations, as the transport system is able sector, while the rise by off-peak period bus travel, to optimally operate with fewer passengers due to the with the highest increase of subsidy that exceeds increase of travel costs, where the losses are mostly 90%. (A smaller decline in passenger miles is caused by crowding disutility. expected at the peak period bus transport, because of the optimal subsidy rate is smaller than the current The exception is the off-peak bus service of Los one.) Angeles—we can also observe a minimal increase in London’s peak rail marginal welfare gains—the The disaggregated results of Table 3 also show that model predicts significant subsidy and passenger similarly to the studied cities abroad, the benefits of growth compared to previous results. This may be the reduction of externalities also play a decisive role caused by the fact that decreasing peak-time in the Hungarian transport system. Compared to the subsidies and significantly increasing off-peak period other transport systems however, in Hungary subsidies may lead to a higher proportion of trips relatively high subsidy rates are applied, so the taking place in the latter time period. situation does not provide the opportunity for a significant increase of subsidy—with the exception Comparing the results with the unmodified model’s, of off-peak bus transport. Therefore the congestion it can be clearly seen that even in cases of moderate losses can virtually neutralize the benefits of scale congestion and crowding level, the passengers’ travel economies. It can be undoubtedly concluded, that costs significantly increased. As a result, in most similarly to other cities, a high subsidy rate is a cases a substantial part of the benefits arising from necessity of the Hungarian interurban public scale economies are neutralized by crowding losses. transport system. In all the modelled cities crowding losses approach the level of peak rail transport’s benefits coming from The results of the modelling with Hungarian data scale economies, and in the case of London it also show many similarities with the characteristics of outweighs it. Therefore, as a results of the model foreign cities, although the model basically aims to modification, we conclude that at all (public) examine the transport system of the urban and transport systems with heavy traffic the negative suburban environment. The similarity between the effects of crowding, and all the related losses should results of the model variants (the original and the always be taken into account. modified version) shows that the model can be 196 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia Table 3: Results of the application of the model in Hungary Hungary Rail Bus Peak Off-peak Peak Off-peak Current subsidy, percent of operating costs 87 87 76 64 Marginal welfare effects 𝑀𝑊/𝑊 at current subsidy 0.27 0.01 –0.02 0.38 Marginal cost/price gap –0.85 –1.74 –0.71 –0.69 Net scale economy 0.02 0.13 0.12 0.69 Crowding costs –0.11 –0.09 –0.15 –0.12 Externality 1.01 0.64 0.58 0.05 Other transit 0.20 1.07 0.15 0.46 Optimum subsidy, percent of operating costs >90 88 74 >90 Percent change is passenger miles 5.7a 3.8a –3.2a 92.4a adapted and used in a geographical, social, economic, 7.2. Interpretation of results in Hungary technical environment that is significantly different from the original cities. Naturally, the accuracy of the The results of adapting the model for Hungary results could be improved by specifying the data, or showed that although the model is primarily designed by a detailed review of the factors adjusted to the for urban and suburban traffic, it can be applied to features of other (Hungarian) cities. Nonetheless, the national level service providers as well. As the calculation method and results are perfectly collected technical and economic data primarily consistent with the aims of the study. described the operation of the national railway and bus transport system, we could analyze the data of 7. SUMMARY, CONCLUSIONS, interurban traffic at a national level. In our opinion, OPPORTUNITIES OF the most important requirement for the model is that data shold come from cities, counties, regions etc. DEVELOPMENT with similar economic, social, and transport The results of the paper has been presented in two conditions—this limitation prevents important clusters. Firstly, we described the results of the features from being “averaged”, otherwise the data transformations of the crowding formulae and all the would already bring distortions into the calculation. modifications carried out in the model. Second, we The results for the Hungarian transport system are presented the conclusions of the Hungarian clustered around very high subsidy rates. For adaptation. Finally, the suggestions and possible example, providing a larger subsidy rate in the off- directions of further development of the model and peak period could cause the higher occupancy of the research area are presented. vehicles outside peak hours—for example, by 7.1. Effects of crowding shifting a part of peak period bus travels to the off- peak period. According to the model this The revision of the original study have shown that by transmission could utilize idle capacities, and thus it taking the costs of crowding into account, it would be would make a substantial improvement of the advisable to apply a smaller subsidy at peak periods. transport system and would also achieve the This would obviously lead to an increase in fares, reduction of social losses. resulting in fewer passengers and thus less crowding on public transport vehicles. At the same time, higher We suggest that the state should also promote the off-peak subsidy levels and cheaper fares would more even use of capacities in other ways to prevent allow some travels to be shifted to this period, which harmful levels of crowding. Gradual shift start would provide better overall transport conditions— supported by the state, postponing the beginning of naturally these provisions would require time- school, or by streamlining the flow of information by differentiated pricing which is currently unavailable organizing and standardizing a real-time information in Hungary. It is also important to note that in management systems―e.g. presenting the usual Hungary this theory may be hampered by the fact that occupancy of vehicles, or the actual values using those social groups, whose travels could be relatively ticket purchase or data of vehicle sensors. easily redistributed, such as people over the age of Additionally, transformations caused by the 65, currently do not have to pay fares, and hence the coronavirus epidemic coluld be also good examples. changes of pricing would have no effect on their Changing working, commuting and travel habits, travel habits. accelerating digitalization and the possibility od home office provide an opportunity to a major review From the results of the model modification and of transport strategies. supplementation, we concluded that in these cases of heavy traffic and crowded transport systems the 7.3. Directions for further improvement negative effects of the crowding phenomenon and the By the further development of the model, it could resulting losses should always be taken into account. more effectively predict the impact of strategic 197 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia pricing decisions at a given city or at a regional REFERENCES (national) level. Currently the model cannot really [1] Björklund, Gunilla, and Jan-Erik Swärdh. 2015. account for the capacity constraints of the transport ‘Valuing In-Vehicle Comfort and Crowding system. Thus, for example with increasing peak Reduction in Public Transport’. Working Papers frequencies the user can’t be certain that the system in Transport Economics 2012(12). can manage the increasing demand without costly [2] Boss, Alfred, and Astrid Rosenschon. 2008. Der infrastructure (or traffic organizational and Kieler Subventionsbericht: eine Aktualisierung. controlling) investments. Take the rail network of Kiel: Inst. für Weltwirtschaft. Budapest for example, which is currently suffering [3] Doll, Claus, and Huib van Essen. 2008. Road from the lack of capacity, so the model would also Infrastructure Cost and Revenue in Europe. require a parallel examination of road and rail Delft, The Netherlands: CE Delft. capacities. Despite the model can control the [4] Hörcher, Daniel, Daniel J. Graham, and Richard occupancy of a particular traffic mode and in a given J. Anderson. 2017. ‘Crowding Cost Estimation period by optimizing crowding, but there may be with Large Scale Smart Card and Vehicle situations where this method is no longer sufficient Location Data’. Transportation Research Part B: and enough for an optimal solution. Methodological 95:105–25. doi: 10.1016/j.trb.2016.10.015. The main question of the modelling carried out [5] International Transport Forum. 2018. ‘Zero during the research was to examine the kind of Value of Time Roundtable’. ITF. Retrieved 6 pricing that could be used to achieve the (second best) November 2019 (https://www.itf-oecd.org/zero- social optimum. Though, changing prices necessarily value-time-roundtable). imply some kind of traffic reorganization, e.g. the [6] Jansson, Jan Owen. 1979. ‘Marginal Cost Pricing increasing occupancy of public transport due to the of Scheduled Transport Services’. Journal of reduction of cars in areas prone to congestion, but Transport Economics and Policy 13(3):268–94. there are indirect effects that do not appear in the [7] Jászberényi, Melinda, and József Pálfalvi, eds. model. The calculation does not include the gains 2009. Nemzetközi közlekedés és turizmus. from detached and unused parking and transport Budapest: Aula Kiadó. spaces in downtown areas due to reduced road traffic, [8] Keserű, Imre, Jeroen Bulckaen, Cathy Macharis, or the potential benefits of using these yet free spaces Joeri Minnen, Ignace Glorieux, and Pieter Theun for other purposes—the benefits of increased local van Tienoven. 2015. ‘Is Travel Time Wasted? trade due to bigger pedestrian and bicycle traffic, or Evidence from a Time Use Survey in Flanders, the rise in property prices. Belgium’. Windsor, United Kingdom. [9] Keserű, Imre, and Cathy Macharis. 2018. Further questions are raised, since due to the use of ‘Travel-Based Multitasking: Review of the aggregated data, it is not possible to select or further Empirical Evidence’. Transport Reviews divide users, i.e. the impact of pricing cannot be 38(2):162–83. doi: examined separately on several social groups. It 10.1080/01441647.2017.1317048. would be important to analyse in detail how social [10] Kroes, Eric, Marco Kouwenhoven, Laurence groups respond to a transport policy pricing decision, Debrincat, and Nicolas Pauget. 2014. ‘Value of since in most cases there is a correlation between the Crowding on Public Transport in Île-de-France, quality of transport system and the settlement France’. Transportation Research Record: environment, thus housing prices, and so indirectly Journal of the Transportation Research Board with the financial situation of residents. For these 2417(1):37–45. doi: 10.3141/2417-05. reasons, it is not certain that groups which are still [11] Mohring, Herbert. 1972. ‘Optimization and Scale distinguishable in reality will behave according to the Economies in Urban Bus Transportation’. The model as a result of pricing. Therefore, it is American Economic Review 62(4):591–604. questionable whether the social and the political- [12] Mohring, Herbert. 1976. Transportation economic optimum do not slip, or intentionally can Economics. Ballinger Publishing Company. be slipped away. For example, the price-sensitive, but [13] Munkácsy, András, Imre Keserű, and Miklós typically more mobile people can be “supported” by Siska. n.d. ‘Travel-Based Multitasking on Public the overcharging of transport services for people with Transport: An Empirical Research in Hungary’. less income and reduced mobility. But this method Periodica Polytechnica Transportation Engineering (megjelenés alatt). can be used for good reasons as well, if we influence the travel patterns of the more price-sensitive social [14] Parry, Ian W. H., and Kenneth A. Small. 2009. ‘Should Urban Transit Subsidies Be Reduced?’ groups by pricing to disencumber the overcongested The American Economic Review 99(3):700–724. and crowded peak periods. doi: https://doi.org/10.1257/aer.99.3.70. Acknowledgement [15] Perone, Jennifer S. 2002. Advantages and Disadvantages of Fare-Free Transit Policy. I am grateful to my supervisors, András Munkácsy Tampa, FL: University of South Florida. and Dániel Hörcher for their help and support. 198 YRS 2021 Tamás Strommer 15. -17. September EFFECTS OF CROWDING ON THE OPTIMAL SUBSIDY OF PUBLIC TRANSPORT Portoroz, Slovenia [16] Statistical Office of the European Communities. [19] Turvey, Ralph, and Herbert Mohring. 1975. 2019. Energy, Transport and Environment ‘Optimal Bus Fares’. Journal of Transport Statistics. Economics and Policy 9(3):280–86. [17] Tirachini, Alejandro, Lijun Sun, Alexander [20] Vickrey, William. 1980. ‘Optimal Transit Erath, and Artem Chakirov. 2016. ‘Valuation of Subsidy Policy’. Transportation 9(4):389–409. Sitting and Standing in Metro Trains Using doi: 10.1007/BF00177700. Revealed Preferences’. Transport Policy 47:94– [21] Whelan, Gerard, and Jon Crockett. 2009. ‘An 104. doi: 10.1016/j.tranpol.2015.12.004. Investigation of the Willingness to Pay to Reduce [18] Tscharaktschiew, Stefan, and Georg Hirte. 2011. Rail Overcrowding’. Harrogate, England. ‘Should Subsidies to Urban Passenger Transport Be Increased? A Spatial CGE Analysis for a German Metropolitan Area’. Technische Universität Dresden, Faculty of Business and Economics. 199 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Daniel Szabo, Mgr.1 Transport Research Centre Lisenska 33a, 636 00 Brno, Czech Republic daniel.szabo@cdv.cz ABSTRACT Technological and social progress in bicycle use and vehicle versatility (smart and social innovations, namely electrification, smart infrastructure, bicycle-based logistics systems or crowd-shipping) and landscape pressures, connected to GHG emissions and changing patterns of virtual and physical mobility, have the potential to radically tip the scales of the dominant socio-technical transport regimes in personal mobility. The article utilizes a mixed-theory approach, based on analyzing the data, collected through the Czech National Travel Survey (Česko v pohybu), conducted in 2019, that feed into a multi-level perspective (MLP) model of sustainability transition pathways. The article aims to identify core regime factors of bicycle mode choice in the Czech Republic through controlling for variables such as length and time of the journeys, daily and weekly variations, activities, connected to bicycle use, trip chain structure, and socio-geographical variables, such as urban structure and density, amenity density, socio-demographic variables and others. The results of the statistical analysis form a basis for an MLP model, that analyses current and future niche, regime and landscape factors and potential pathways towards increasing bicycle mobility, motility, universality and users' diversity, concerning systemic interplays between infrastructure, users, governments and other human and non-human actors. Through identifying the possible development scenarios and viability of achieving the goals, defined in national strategies and urban policies and exploring the potential for hybridization of the components of the regime, that are the most sensitive to change, the article further proposes the space for new potential policies to support broader viability of velomobility for the general population. Keywords: Sustainable urban mobility, Velomobility, Mode choice, Multi-level perspective, Socio-technical innovations 1. INTRODUCTION has been looming on the horizon for at least a decade – so-called peak travel, or peak car (Millard‐Ball, There are few singularly more important urban 2011) – even Czech motorization rate is climbing mobility goals than promoting cycling as a still, and is projected to do so . universally viable choice mode. The evidence is compelling and mounting, that the current (and Cycling mobility consists of a wide spectrum of accelerating) automobile-dominated mobility system activities and experience, where the dominant is unsustainable from a wide range of criteria – narratives – usually of highly skilled, but unobtrusive environmental, social (equity) or public health – and cyclists, and cycling as an activity as a goal – might a change in course is direly needed. In the presence serve as a gatekeepers both for cyclists and of decades of strategies aiming to attain just that8, the infrastructure design. Focusing on too a narrow range indicators, connected to macroscopical cycling of cycling patterns (represented e.g. by data, mobility goals (accident incidence rates, modal split) collected by fitness applications, such as Strava in the Czech Republic are slow to progress, even Metro) or indicators (such as vehicle miles travelled) receding at times – we might speak of a velomobility is doomed to be counterproductive, when assessing inertia. On the other hand, another significant trend 1E.g. through the National strategy for development of cycling mobility 2013-2020. 200 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia for a softer, more equitable mobility – not a transportation modelling – which is the core part of performance one. the process of deciding on the location and design of road network – to a large extent lacks detail and The first Czech National Household Travel Survey regard for soft modes. (NHTS) (Centrum dopravního výzkumu, 2020) is a publicly available dataset on harmonized personal One of the core aspects of a viable cycling system is mobility data, that hasn't been used for a research a relatively dense, connected and safe linear into mobility patterns yet. The aim of the article is infrastructure – i.e. segregated cycleways. From this thus to utilize the available data to analyze the current standpoint, it is safe to say that segregated cycling factors, influencing modal choices, and the infrastructure throughout Czech Republic is border to stabilizing and destabilizing factors of the state in non-existent. There are less segregated cycleways order to identify possible transition pathways towards (lengthwise) in Czech Republic, than there are cycling culture, open and accessible for all people highways (1.94 % of length compared as of 2020). and all purposes. There are still less segregated cycleways and mixed cycle-pedestrian ways combined, than highways The paper is structured as follows. Chapter 1 (92.5 % of length compared as of 2020 (CEDA, (Introduction) presents the general overview of the 2020)) . Given the low share of highways on the road current state of the velomobility and its confrontation network and magnitude of difference in construction, with the incumbent mobility regimes in order to operational and externalities costs, the magnitude of better outline the landscape in which the mobility difference is incomparable – and still growing. The choices are being made. Chapter 2 (Methodology) rate of growth of the network is further concerns the methodological background on external geographically uneven, and still mostly concentrated environmental factors on the bicycle mode choice along a few long-distance routes (such as Eurovelo), and presents the data, used for the analyses. Chapter even negative at times (e.g. via recategorization). The 3 (The binomial regression model) presents the difference is the highest in the urban areas, where analyses and results that form a base for the multi-protected measures are scarce, due to spatial level perspective model. Chapter 4 (The multi-level conflicts, or outright dangerous due to a common lack perspective model) outlines the potential for urban of skilled cycling infrastructure planners. and national policies and policymakers to adopt support for the regime change. Chapter 5 discusses The mean increase of segregated cycling and mixed the results and conclusions in the light of the potential cycling-pedestrian paths has been 28 meters per for a politically-driven change. municipality per year for the last three years – although this is a substantially higher growth rate 1.1. The context of cyclist infrastructure than for the whole road network (8.7 % per year, planning in Czech Republic compared to 0.08 % per year). New cycle-only roads represent a staggering 26.75 % relative annual 1.1.1. The overview growth rate, which only illuminates the paradox, Nello-Deakin (Nello-Deakin, 2020) has recently where even small projects represent a tremendous proposed not only that practical (ie. technical or relative leap in the overall network capacity. The vast economical) aspects are not the main barrier for majority (84 %) of the municipalities haven't cycling infrastructure provision, but that the recorded any change in their segregated cycling theoretical knowledge has expanded enough to a infrastructure. point, where there is little else to explore to further Reliable data on softer (non-segregated) cycling illuminate the forming policies – and multiplying infrastructure are not yet being collected – while the research in underlying factors of cycling might be in OpenStreetMap data structure is exhausting, the fact counterproductive. completeness of mapping cannot be estimated (albeit Indeed, the aim of this article is not primarily to it is probably fairly high, and shows a similar pattern: uncover an previously unknown factor, that would a dozen of cases of active municipalities with once again "solve" the problems of geographically relatively higher share of cycle lanes). The support of and societally uneven propensity to cycle. On the cycling measures is highly geographically uneven other hand, where the general theory of largely and dependent on the proactive approach of the universally applicable cycling factors is in practice respective municipalities, public pressure and frowned upon i.a. due to the lack of data, an analysis cooperation of several administrative bodies to push of both the mobility patterns and the specific and for individual projects' approval. varying problems, the Czech municipalities face, is not misplaced. 1.1.2. The municipal planning Moreover, regardless of the deemed lack of gaps in The cycling infrastructure is planned mainly on local theoretical research, the practical application of (municipal), regional/long-distance (NUTS 3) level, proven factors of modal choice through or both in coordination. 201 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia  Technical standards for cycling infrastructure segregated infrastructure (especially more often than not deter from pursuing the recreational) does not necessarily translate safer (or any) cycling solutions, by raising the into a well-serving infrastructure. bar of costs, land-use/spatial requirements, and technical and safety conditions, that can 1.1.3. The role of expertise in sustaining a be objected to. locked in automobility regime  The focus on isolated cycling infrastructure The current state constitutes a locked-in regime of projects is favoured both by funding schemes automobility – that is a socio-technical regime (Geels both on national and European level (given F. W., 2007), sustained through shared expert the high threshold of project preparation practices (ie. solutions heuristics) but are also costs). Needless to say, the projects, that embedded in the societal and cultural practices and favour recreational cycling, might produce an rules, or built infrastructure and its obduracy (Urry, uneven and damaging effect on various non- 2004) (Hommels, 2005) (Hoffmann, 2017) . The recreational cycling/cyclists and their dominance of the regime reverberates in different presence in the vast majority of non-cycling stages of transport policies creation or enactment, road space.  even if its effects are variegated as a result of groups The focus on large, linear infrastructure and networks interactions. While it is outside of the projects disregards the value and qualities of scope of this article, it is worth noting that policies, surrounding space, that are paramount to a that favour growth of automobility (or stymie human movement, and the vast majority of alternatives), pervade through different sectors, daily travel patterns: the street permeability, mostly connected to transport and construction. The accessibility of destinations (including examples may draw from the legal basis for parking provision and safety), or presence of requirements for parking place provision (in different amenities.  context, although to an extent applicable, famously Both cycling and non-cycling projects consist analyzed in (Shoup, 2011)) – or in practice relaxed mostly of new, added corridors, rather than interpretation of technical norms, that are intended to transformation of the existing space. If serve active or sustainable modes of mobility. unmitigated, this adds capacity to the road network, inducing new automobile demand The automobile regime lock-in has found ways to (such as in the case of ring roads (Drabicki, trickle down further to transport modelling or mode Kucharski, & Szarata, 2020)), reduces land choice modelling: use density and increases environmental costs 1. for one, it appears as fixed automobility (e.g. habitat fragmentation, road runoff growth coefficients (even where the pollution). automobility growth directly contradicts the  Whether the cycling infrastructure stimulates long-term transportation policies' goals) cycling demand or vice versa, there is a both in the rate of motorization and in the positive correlation – even if this is city- rate of traffic intensities and vehicle miles specific, and many municipalities sustain high travelled. The macroscopic pressure levels of cycling without any segregated translates into increasing the network infrastructure; and, on the other hand, capacity, where lack thereof is projected. 2. It is expressed in involuntarily skewed data from national travel surveys (travel diaries), where lower validity of active trips leads to their underrepresentation (e.g. there is 3.5 % higher representation and 0.6% lower representation of car/bicycle trips respectively in our valid dataset, compared to the raw dataset, see Table 1 ). Even further, the active trips are underrepresented by involuntarily lacklustre reporting of the travellers (as the less important, forgettable trips), or where they exhibit a higher degree of diffusion between the trips and the activities (resulting in lower trip rate for active modes, connected to several activities). Furthermore, in activity-based Figure 1: The correlation between modal share of cycling models, non-trip activities are not and the share of segregated cycling infrastructure in the considered at all, inflating the significance Czech cities 202 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia of the major trips – and reducing sensitivity which the modelled travellers' choices are being to the shift in virtual mobility/presence. decided. Multinomial logistic models have been a preferred approach to targeting variables influencing 3. The role of a technical expert obfuscates the transportation mode choice, mostly due to their leeway that exists when creating or explanatory power, that can be utilized for assessing interpreting the models and selecting the future infrastructure or non-infrastructural measures, proposed measures. The scientific such as traffic calming. knowledge – be it expertise on human health, air pollution, transport modelling or The approach has several possible pitfalls that econometrics – is shaped by the alternatives demonstrably influence decisions on transportation – and means for pressing towards the pre- projects: selected alternatives are manifold – their 1. The transportation models are implicitly (lack of) accuracy is therefore predictable motivated to skew the results in order to (Flyvbjerg, Holm, & Buhl, 2007). On the favour new development. This is a function other hand, where conflicting expertises of the power imbalance between the large accumulate and are politicized (usually in technical infrastructure construction sector the cases of controversial variants, or and marginalized actors and interests (such exceeding uncertainty), the expertise may as environmental protection), where be bypassed by the decision-makers transport models are often created as a (Konopásek, Stöckelová, & Zamykalová, means to substantiate new development, not 2008). as a tool for their evaluation. Road 4. At last (and this is the starting point of our infrastructure plans in Czech republic are article), the impact is no less present in the construed upon an idea of perpetually common functions for mode choice, that delayed completeness that pervades through stem from the four step models' different sectors, but is probably most requirements for reducing the scope of tangibly exemplified through the core modelled variables and spatial detail. The highway and motorway network, where active modes of transport are still often pressures towards new development have omitted, or left behind, often due to lack of petrified in different socio-technical data collected (e.g. pedestrian counts conflicts (Konopásek, Stöckelová, & observations), simplicity of the network Zamykalová, 2008) and are present as a (e.g. lacking paths) or reducing the permanent struggle to unstring the existing complexity of the model. Not including legal protections and public participation factors, relevant for specific mode choice representation, such as in the case of the decisions then leads to underestimation or latest ongoing Construction law reform. discarding of planning policies and 2. As (Cervero, 2002) notes, the underlying strategies, that influence the cyclability mode choice models are in practice usually (walkability) of the area. One of the seminal underfitted, lacking plenty of possible works in this area (Katz, 1995) highlights available factors, apart from generalized the practical aspects of modelling costs – such as built form, land use specifically for the bicycle (even if such characteristics or network quality. Omitting case can be made for any mode of significant factors from the models leads to transportation) – quantifying mutual causal a distortion of the effect of the other factors relationships between urban planning of mode choice. processes, individual choices (such as 3. The model's purpose ends at identifying the household-work location) and factors, related to discrete choices based on transportation mode choices raises the statistical probability. The material, or even probability of bringing about the change, as political reality cannot be easily bent to it allows for their explicit inclusion. From attain specified goals or even reshaped to fit the built form point of view, especially the models, as it also would have to important are the microscale and urban form disregard the trade-offs between the factors, that are not reducible to time costs, identified factors. Even on the most simple such as network design and suitability, terms, the policy trade-offs are unavoidable: density or directness (Rybarczyk & increasing average traffic speed to reduce Changshan, 2014). travel time leads to higher accidents* rates When dealing with transportation project decisions, and higher investments toward safety and public authorities will refer to transport modelling, the users side, which further deter new users typically based on multinomial logit models, derived (such is the case of mandatory helmets on from travel surveys data and data about the existing bicycles or potentially mandatory, even if infrastructure, that, combined, form the landscape in non-existent helmets in cars). Land use 203 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia density (or rather suburbanization 2. METHODOLOGY processes) are a results of decades-long deregulation of land use regulation and 2.1. Built form and Land-use factors pressures toward profit maximization influence on travel behaviour (Maier, 2012). Land value and land use The relationship between various exogenous factors regulations usually decrease with increasing – land use, urban morphology, or endogenous factors distance from the urban core, leaving more (such as risk perception or attitudes towards a room for pedestrian or cycling infrastructure transportation mode) have been studied extensively – in areas with low density and transit on various scales for several decades (Rybarczyk & accessibility, increasing car dependency. Changshan, 2014). The shortcomings that (Cervero, 2002) has identified There are additional studies rooted in Czech context in mode choice modelling are very much present in (e.g. (Gabrhel, 2019)) that propose extending the the current state of the art in Czech Republic, despite travel surveys to include psychological or stated the theoretical advances and even though the preference factors, based on the individual predictors availability and completeness of the datasets on for mode (bicycle) choice. This approach, while possible influencing factors, such as accidents rates, providing useful insights, relies on extending the land use, density, bicycle facilities or urban design, scape of the national travel surveys' travel diaries, has improved considerably. which already battle with low return/fill rates due to But neither changes in the bicycle infrastructure time required and lack of motivation. provision, nor soft, persuasive policy changes, might On the other hand, if shown any significance (esp. on be the leading factor, influencing the proclivity to the national level), extending the existing models cycle. The disruption of public transport services, with additional available data could potentially aid to drops in road usage and shifts in daily mobility create more precise models without increasing the patterns due to recent Covid-19 pandemics have in burden on either side. many metropolitan areas worldwide resulted in a sudden spur of new, often pop-up bicycle The primary candidates for such (quantifiable) infrastructure or bicycle-aiding schemes, such as factors would be those, that can be aggregated and low-traffic neighbourhoods. Other major pressures – connected to the trips (typically routes or origin- climate change etc. – may break alone, or via major destination zones), usually based on geographical actors (e.g. European union – see (Hoffmann, 2017)) proximity or zonal aggregation – factors, such as land into a new momentum and a more permanent regime use, built urban form, or accident rates. Identifying change. Such "regime erosions" may stem from two more detailed parameters of either the network or intersecting paths: the "1) bottom-up mobilization, built environment requires increasingly unobtainable which gains momentum through positive policy consistency and validity of the geo-spatial data. feedbacks, and 2) weakening policy regimes due to The chosen approach allows to control the correlation negative policy feedbacks. These conditions enable of aggregated socio-demographic factors, such as policymakers to defect from the eroded policy regime automobility rates or income with urban-transport to the alternative, which is accompanied by major structures (e.g. car/transit accessibility), but also policy change." (Roberts & Geels, F.W., 2019, p. 4) individual characteristics, that form the mobility Indeed, it is impossible to make reliable predictions decisions on the individual and trip level. The about future human behaviour in tumultuous times of relationship between different aspects of built potentially critical junctures (ibid.). Yet, precisely environment may not be linear, or transparent – due the critical junctures in history present the highest to the diversity and variance in land use and travel potential for disruptive policies to gather support and behaviour through a whole republic. Research in the create positive feedback loops – drawing in new influence of breadth of various factors onto mode users, providing fast, demand-driven improvements. choice (or vehicle kilometres travelled) usually focuses on a strictly geographically delimited area (or Yet – are there any potential paths – or breaking comparison of multiple such areas (Cervero, 2002)). points forward for velomobility in Czech Republic precisely in the light of conflicting interests and At the same time, some of the factors might not show potential regime instability? The aim of this article is hypothesized effect, because of their form or to bring a multi-level perspective model of socio-localization. The typical example for this effect technical innovations approach to bicycle would be localization of bicycle infrastructure, that infrastructure planning. The model is in the first tends to avoid busy, quoditian connections due to phase informed by a regression analysis of possible spatial conflicts with motorized road infrastructure, macroscopical factors, involving urban environment the effect being meagre, or even negative estimates and individual/household characteristics, influencing of the presence of segregated infrastructure on the current individual mobility choices for the general mode choice, as the misplaced infrastructure is not population. being actively used. 204 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia Controlling for other multiple variables, such as The NHTS has been conducted as a continuous destination activity of the trip, land use and POI survey from May 2018 to April 2019, except for the density along the route, may improve the predicting months of December and January via probability- precision of the model. However, for the second step proportional-to-size sampling with sample size of of the analysis, the substantial significance is more 40 000 households with assumed sample response important, than atheoretical modelling precision. rate of 25 %, resulting in the final sample of 9419 households, contacted in person, following a PAPI or Yet even if untangled, the possible relationship CAPI survey for all the members and vehicles of the between land use characteristics and mode choice is household. still not easily translatable into land use or transportation plans. Even detailed data on various Table 1: Individual (socio-demography) variables of geographically aggregated indicators cannot easily the sample translate into individual decision making, bar viable Variable Description Sample public policy. (n=22,122) As shown in the Introduction, it's not the land use Gender 0:Male 10,647(48.13%) 1:Female 11,472(51.87%) factors, or land use policies, that form mode choices Age Median 45 – the different factors, influencing mode choice are Education 0: Unfinished 2,746(12.4%) themselves a result of interplay of broader network of primary actors, influencing technology, infrastructure, 1: Primary 2,292(10.36%) 2: Vocational 6,537 (29.55%) politics or culture. 3: High school 7,157 (32.3%) 4: Vocational 526 (2.38%) The identified predictors are henceforth not primarily college used as factors, stimulating bicycle use, on the 5: University 2,859 (12.9%) contrary: the general assumption being the bicycle is Work status 0: Not working 10,602(47.9%) under favourable conditions usable universally, many 1: Working 11,514(52.1%) of the factors are better in explaining the missing gaps Car availability 0: none 7,659 (34.6%) 1: any 14,463 (55.4%) in broader adoption of the bicycle. Bicycle 0: none 11,497 (51.9%) availability 1: any 10,625 (48.1%) 2.2. Sampling and data Driving license 0: none 8,628 (39%) The data, used for the analyses are as follows: availability 1: any 13,494 (61%) 1. Czech National Household Travel Survey (NHTS, so called Czechia on the move) The NHTS data structure is split into several (Centrum dopravního výzkumu, 2020). The dataframes, including: data have been collected through random  Households sampling through years 2018-2019, using  Persons standardized approach, based on European  Trips harmonization of travel surveys  Cars (vehicles) (Armoogum, 2014) and Austrian methodology KOMOD, that has been Each of the dataframes have been checked and adapted for Czech context. validated for completeness, range validity, data 2. The Czech Global Network (GNDB) integrity, data usability and logical consistency. database, created by Road and Motorway Missing values were checked and imputed or Directorate of the Czech Republic. corrected, according to typical errors present. The built environment and road network data have The trips have been geolocated via identification on been aggregated on the municipal level with the various levels (the most precise being house level) following considerations: with the official registry of addresses, combined with  The uneven precision of geographical API geolocation approach. The probable routing of localization might misplace trips origin- the trip has been calculated, based on constructed destination TAZ, while this is far less network and a separate API for public transport probable on the level of a city. queries. Validity of the geolocation of origin-  The urban mobility policies tend to be more or destination points and the validity of less coherent. On the other hand, there is high computed/declared time and distance of the trip has variability within the city both for the built been cross-checked, based on relative and absolute environment and for the mobility patterns thresholds of difference between computed and (e.g. urban core and periphery/residential declared time for each mode, which leads to zones), and this variability will increase with relatively lower share of active modes. the size of the city. This may be a reasonable argument for aggregating the data on the level of urban districts, amending the analysis. 205 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia Table 2: Proportion table for the raw/validated sample The model, estimating the choice of bicycle for the of the trips Czech residents has been constructed as: Main mode All trips Reliable trips 𝑃(𝑏𝑖𝑐𝑦𝑐𝑙𝑒 = 𝑦𝑒𝑠) [%] [%] 𝑃𝑛 = log [ ] = 𝛼 + 𝛽 1 − 𝑃(𝑏𝑖𝑐𝑦𝑐𝑙𝑒 = 𝑦𝑒𝑠) 1( 𝑇𝑑) Car driver 29.2 31.6 + 𝛽2( 𝐻𝑣) + 𝛽3(𝑟𝑥) + 𝛽4( 𝐶𝑥) Car passenger 9.9 11 + 𝛽5( 𝑃𝑐) + 𝛽6( 𝐻𝑑) + 𝛽7( 𝑑𝑝𝑥) Bus 3.8 3.8 + 𝛽8( 𝑇𝑚) + 𝛽9( 𝑆𝑙𝑜𝑝𝑒𝑎) + 𝛽10( 𝑃𝑎) + 𝛽 Bicycle 4.5 4.2 11( 𝑊𝑚) PT 14.6 16.1 Where probability P for person n of choosing the Other 0.8 0.5 alternative (bicycle or non-bicycle mode) is a linear Walk 35.4 31.1 combination of 𝛽1−𝑝 predictors, that is a vector of Train 1.8 1.6 parameters estimating the probability of choosing the alternative. Overall, the share of time, distance and location-wise validated trips is 69.86 % (15,498 invalid trips, with The following predictors/categories were considered, critically corrupted structure, out of 51,434). Since all based on existing theory and available data: of the variables are critical for our analyses and there 1. Land-use attributes, urban morphology: is no assumption of non-standard distribution of the the diversity, density and design of the omitted data, we have decided further to work with origin/destination area are hypothesized to the validated data only. increase the probability to choose bicycle as Main mode for the trip, where combination of modes a main mode of transport. Average slope of has occurred, has been chosen, according to the the streets of the city or mean altitude of the general practice (train > coach > car > public city may play a role in mobility choices. transport > cycling > walking). 2. The urban bicycle/sustainable mode policy attributes proxies: The density of Table 3: Trip variables distribution in the sample network of either segregated bicycle Variable Description Sample network or mixed bicycle/pedestrian (n=35,936) network in the origin/destination area. The Distance Mean 3.29 km ratio of traffic-calmed streets (less than 50 measured km/h) to the residential streets length may Main mode Bicycle 4.22% be correlated with more favourable Car-driver 31.62% conditions for the cyclists. The ratio of Car-pass. 11.01% Coach 3.84% pedestrian network to residential streets Public transport 16.05% length. Train 3. Extended costs of the trip, where possible Walk 16.31% to identify: steepness, altitude difference, Other 31.08% 5.23% detour factor, overall length, overall time. Destination H: Home 42.77% 4. Character of the trip: intra or inter-urban; (activity) purpose W: Work 19.78% destination activity (purpose) – considering S: Shopping, 10.12% utilitarian vs. non-utilitarian trips. Services L: Leisure 8.34% 5. The traveller's personal attributes: E: Education 6.39% car/bicycle and driving license availability, M: Personal 7.66% income, gender, age. B: Business/work 2.35% 6. Safety: Areas with higher share of serious trips F: Food RTA (with/without pedestrian/cyclist O: Other 10.38% victim) are hypothesized to reduce 15.22% propensity to choose bicycle as a mode of transport. 3. THE BINOMIAL REGRESSION 7. Weather: While tracing back weather MODEL patterns for the concrete locations several years back would prove hard to obtain and 3.1. Analyses imprecise due to the publicly available data The framework, used for estimating the propensity of structure, average annual Czech residents to cycle is based on and extended weather/temperature variations have been binomial regression model, that tries to improve the assumed. goodness of fit when predicting bicycle mode choice, The following variables have been selected as of yet, based on additional factors. The binomial regression based on available or derivable data, valid with model was fitted through pscl package (Jackman, reasonable confidence: 2008). 206 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia Table 4: Comparison of additional potential predictors Table 5: Binomial logit model for predicting bicycle for bicycle mode choice only mode choice in Czech population Operationalizatio Availability Model I Null Model Identified Category n / (Intercept) -2.46 *** -3.38 *** predictors Source (0.13) (0.09) Traffic Urban Share of speed Distance travelled -0.10 *** -0.10 *** calming policy limited (< 50 GN 2018 (0.01) (0.01) km/h) streets in (0.00) the city of Work status 0.38 *** 0.37 *** residence (%) (0.09) (0.08) Quality of Urban Share of cycling GN 2018 / Gender -0.24 *** -0.23 *** cycling policy infrastructure per OSM (0.06) (0.06) infrastructur non-service roads Productive age -0.37 *** -0.37 *** e in the city (0.08) (0.08) Road Geograph Average slope of GN 2018 College Education -0.04 -0.16 grading y distinct road (0.09) (0.09) segments in the Absence of public transport 0.27 *** 0.39 *** city membership disposition Safety Risk Traffic accidents Police of (0.07) (0.06) perception severity index the Czech Bicycle availability 0.43 *** 0.44 *** Republic (0.02) (0.02) POI density Urban Number of non- GN Absence of driver’s license 0.33 *** 0.33 *** structure transportation 2018/OSM (0.07) (0.07) related POIs in the Number of children in the household -0.53 *** -0.55 *** vicinity of origin (0.04) (0.04) of the trip (< 300 Weather (summer) 0.59 *** m) (0.06) Weather Comfort Months of - Share of segregated cycling 0.03 *** July/August infrastructure in the origin city cadaster (0.01) Segregated, any or contra-flow cycling infrastructure Traffic accident severity index of -0.05 *** is rare in Czech cities: the contra-flow lanes or roads cyclists (Reinhold) per population in don't exceed length proportion of 2.5 % in any of the the origin TAZ (0.02) cities, i.e. the policy is scarce even in the cities that The average slope of streets in the -0.63 *** adopt it (when the general recommendation of city of origin universal adoption has been debated (Methorst, (0.06) 2017)). The indicator has not been included in the POI count in the origin TAZ -0.01 *** Share of speed limited (< 50 km/h) -0.05 *** model, since its validity cannot be estimated due to streets in the city of origin the impossibility of obtaining accurate historical data (0.00) for the sampled year. nobs 35125 35203 null.deviance 12420.22 12427.04 The urban design, density and diversity variables, AIC 10775.40 11163.55 along with public transport accessibility, were BIC 10910.87 11248.24 disregarded for the time being, as different df.residual 35109.00 35193.00 Goodness-of-fit (pseudo-R²) 0.16 0.12 approaches to their operationalization have yielded McFadden 0.14 0.10 wildly varying results. χ² 1676.82 (15) 1283.49 (9) *** p < 0.001; ** p < 0.01; * p < 0.05. The model performs better, than the null model, albeit not significantly. This is not surprising, given the variability of cities and the individual variability that exceeds the explanatory potential of aggregated characteristics. To a large extent, cycling shares the effect of the observed factors with walking, which significantly reduces the presicion of the model, but creates a space to explore the synergy between the two modes and the possible co-dependence of their promotion. Figure 2: OR plot of factors for binary mode choice The following table shows the results of the fitted model without beta standardization of the coefficients: 207 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia The road network used (CEDA, 2018) is collected twice a year for the use of The Road and Motorway Directorate of the Czech Republic, responsible solely for management, security, administration, maintenance and repair of motorways and main roads, hence the lack of details on the cycling infrastructure. Respective municipalities conduct independent surveys, collect and maintain the data on the infrastructure in their own structures, and are required to inventarize their own infrastructure. Direct demand models for bicycle use (e.g. (Fagnant, 2016)), or models based on cycling traffic level stress (such as BLOS) usually utilize more precise data on Figure 3: Modelled effect of segregated cycle factors, such as cycle lane width or type, traffic infrastructure share on mode choice. intensities, or weather. The identified infrastructural factors do not explain a significant part of the variability, ie. the There is an overlap between theoretical models and macroscopical reasons, why people in some increasing demand for better cycle routing (stemming municipalities cycle more frequent and for longer from the evolution of cycling-specific navigation). trips, than others. This is not surprising, given both This has led to a steady co-evolution of public the unreliable and aggregated character of the data domain data on the infrastructure (such as on the cycling infrastructure, and their variability on OpenStreetMaps), private-domain, pay-per-request the national level, even in design, which is highly data (such as HERE or Google API) and land use and city-specific. their usage in applications, allowing for adding more and more cycle or walk-specific criteria to the routing Additionally, there is a non-standard distribution of choices, such as safety, elevation, avoidance criteria, the used indices, such as bicycle infrastructure, or amenities, or even real time traffic data. traffic calming measures ratios. These measures are typically present either in a larger scale, or, for the 3.2.2. Accidents majority of municipalities, are non-existent, leading to significantly right-skewed distribution. This (Bílová, Bíl, & Müller, 2010), analyzing cyclists' presents an additional potential for model and spatial fatalities in Czech republic have identified car aggregation improvement. Since multiple speeding as the most significant cause of cyclists' observations (trips) have been made for each fatalities, where segregated infrastructure and/or individual, the generalized linear model that assume enforced traffic calming, primarily on the critical independence of the observations might not be the "hotspots" is identified as the main (or rather the ideal solution. only) solution. There is a significant potential for the model The proposed model observes a relationship between improvement regarding several factors that may be (cycling) mode choice and overall, geographically considered in follow-up: weather (obtaining precise aggregated accidents rate, where higher accidents data for the specific trips, regarding rainfall, rate is hypothesized to deter potential cyclists and temperatures with high variability due to continuous correlate with more dangerous road conditions in design of the study); slope (obtaining estimated general. While higher cycling rates lead to higher slopes and cumulative ascent/descent for each trip). absolute incidence rate, it is probable, that the relative incidence rate decreases with higher modal share. 3.2. The potential for model and data collection improvement 3.2.3. Routing 3.2.1. Road network Due to various considerations, it is not practical for travel diaries to be geolocated or routed via GPS Apart from classification problems, the road network aided devices (Kouřil, 2020). The data collected via lacks more detailed information on non-segregated geolocating applications (such as Strava) additionally cycling infrastructure. Crowdsourced projects, such usually suffer from a lack of representativeness of the as OSM, tend to improve both in the depth of detail, sample (e.g. bias towards certain socio-demographic and in the completeness of mapping of the current groups) (Lee, 2021). state of the infrastructure, but the overall level of accuracy (completeness and classification precision) The process of geolocation of the origin-destination is impossible to estimate, and might vary across the points and the routes of travel diaries in the NHTS is country and across different categories of data. highly dependent on the completeness of the respective travel diaries, in the latter case joined by uncertainty about precise routing due to variability of 208 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia the individual choices, or other factors, that are hidden from the sight and do not follow the shortest route preference (e.g. joining several activity-trips into one trip, detouring or delaying for various non- utilitary reasons). Arguably, the very principle of forcedly utilitarian route and activity choice discriminates against the active modes, that from their nature deny such classification. 3.2.4. Weather patterns Especially active modes exhibit variable elasticity for changing weather patterns, which may be locally specific and correlate with the network quality and age (Goldmann, 2020). In case of uncontrolled weather variables, this further deepens the crevasse between the "cyclist-friendly" cities and the rest. All in all, we should not expect the model to perform significantly well, given the identified limitations. Still, the model presents conclusions, that allow for a Figure 4: Mode choice distribution per trip count. modestly confident inclusion into prospective urban sustainable mobility policies. 3.2.5. Urban structure While the model incorporates a generic correlate of POI accessibility in the direct vicinity of the origin, a more detailed approach to the factors such as land- use density, amenity accessibility, road network density/block density along the routes or at the origin/destination TAZ would yield more detailed results and a better representation of the dimension. 3.3. Discussion and inclusion into the MLP model The distance travelled has a negative, or rather cutoff effect on choosing the bicycle as a mode for transport. The means to attain higher cycling levels are in general connected to the short-distance city measures, ie. urban structure densification, yet pervious – connected via direct, safe routes. Technological innovation penetration, such as pedelecs, tend to further raise the threshold of Figure 5: Measured distance distribution per transport willingness to commute both for the distance and mode slope of the route. Contrary to the popular belief, size of the households is positively associated with bicycle use, while number of children in the household is only mildly negatively associated with bicycle use – mainly for four children or more. While cycling is a popular mode for leisure (be it as a leisure activity itself, or as a means to it), the dominant purpose of bicycle use remains work (15.8 to 20.76 %). 209 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia significant portion of the traffic-calmed zones, such as living streets, that are usually located in the suburbian, highly car-dependent neighbourhoods (as shown in the spike in the Figure 5 often with no additional changes to the street profile. The results call for further analysis of covariance of additional urban morphology measures, in order to identify the relationship more precisely. As traffic calming is one of the preferred means to achieve both lower road traffic injuries and deaths and higher active mobility rates, a close attention should be paid to the form, placement, scope and enforcement of the calming measures. On the contrary, segregated bicycle infrastructure has positive, if small impact on the bicycle mode choice. This is a likely result of mutual support of the mass demand and infrastructure provision. On the other hand, spatial conflicts with other land uses (such as Figure 6: Mode co-occurence distribution per trip road infrastructure) push cycling infrastructure on the count for bicycle fringes of effectivity (i.e., long routes, distanced and There is a stronger correlation between trip count disconnected from the urban core, recreational (personal mobility ratio) and mode choice (Figure 2); cycling only), thus reducing the estimated effect. with the former increasing, the probability of Alas, the cycling infrastructure in Czechia is too choosing car also increases. The co-occurrence of scarce and disconnected to estimate more detailed modes within-person or within-household (i.e. choice aspects of usability on the network level. homogeneity) has not been analyzed, mainly due to Number of traffic fatalities and injuries on bicycle- the overall low number of trips recorded per person. allowed roads has a negative impact on the bicycle For cycling persons, the prevalent co-occuring mode mode choice. It is presumed, that segregation and is walking, regardless of the number of overall trips. traffic calming are the most effective measures for Along with the distribution of cycling-only persons, reducing both physical risk of cyclists' injuries and it appears, that cycling-active modes are the primary the perceived risk when choosing whether to cycle mode of transportation for the majority of those and where to cycle. persons, who cycle – or at least on the day they Availibility of a bicycle (number of bicycles in the cycled. This defies a general notion of bicycle as a household) is also positively associated with bicycle spare time, or secondary vehicle (Figure 4). mode choice. While the significance of the parameter is usually understated in traffic modelling, not having the option to cycle in a first place is naturally the most probable barrier, non-cyclists will face. While growing number of policies or initiatives worldwide are targeting this either via tax deduction schemes or direct subsidies with significant results (e.g. (Caulfield & Leahy, 2011)), there has not yet been such effort on national level in Czech republic. Another means for targeting new users are bike sharing schemes, albeit these operate to a limited degree, predominantly in the larger cities. The analyzed predictors thus can, to a degree, form the ground for inclusion into a MLP model, considering the mutually co-evolving scape of actors and processes. Figure 7: The density distribution of binary choice A more precise analyses in the vein of national (blue=non-cycle) per a) share of cycling infrastructure bicycle level of service equivalent (as is present in b) traffic fatalities per population c) distance travelled d) several nations) could yield more precise elasticities traffic calming share (left top to bottom right) for wider range of factors, that could be considered As for the road infrastructure, the paradoxical as a predictive tool for possible variances and relation between traffic calming measures and mode developments in bicycle services provision. choice could be explainable by the character of 210 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia 4. THE MULTI-LEVEL PERSPECTIVE 4.1. Barriers and pressures for and against MODEL velomobility Yet, the bicycle services provision is firmly rooted in 4.1.1. Infrastructure development the long-term constellations of actors and processes, that have established their positions and have a As has been mentioned above, the development of significant influence on the public policies. cycling infrastructure has been stagnant both on the spatial and temporal level, with only recent and length-wise relative spikes in development. Both regional and municipal level strategies have since 2007 renewed focus on recreational cycling as the main driver of new infrastructure construction, being financed from national and European funds. 4.1.2. Policy changes The pressures towards new infrastructure have resulted in amendments of technical standards for cycling infrastructure, that draws broadly from the Dutch Design manual for cycling traffic (CROW). While on the municipal level, since early 2000s, several Czech cities have adopted newly conceptualized Plans for the development of cycling Figure 8: Conceptualizing political defection infrastructure, their contents and realization have (represented by the fat arrow) in the multi-level been a subject of permanent public and political perspective on transitions. Source: Geels and Roberts 2019:3. struggle for space, along the lines of spatial conflicts, traffic safety and control. The multi-level perspective on sustainability transitions has been developed mainly by F. W. Since 2017, the first Sustainable urban mobility plans Geels, mainly focusing on energy systems (low- (SUMP) have been created in cities with population carbon) transitions and mobility systems transitions exceeding 40,000, where the plans are seen as a (Geels F. W., 2007). condition for receiving public funds for infrastructure development. While the principles of the SUMPs Originally and dominantly a historical analysis, the promote sustainable mobility first approach, there are framework has evolved into several strains and no mechanisms present that would enforce either approaches, and has been also used to analyze current inclusion of cycling measures, or condition trends and unfolding pathways in innovation-lock-in development of road infrastructure and parking dynamics and policy implications (Geels F. W., provision with sustainable measures. For example, 2019). In recent years, the framework has gained while parking regulation is more and more present in popularity in identifying the dynamics of the mobility the city cores, this is a result of internal pressures regimes and their stabilizing and destabilizing between commuters and residents, short-term and factors, as a means to analyze potential for long-term parking, that leads little to no results in sustainable mobility transitions (Hopkins, 2017). freeing up space for alternative modes. Recently, the framework has been adapted for analyzing public policy (policymakers') adaptation to On the other hand, SUMPs shed some light on the the system evolution and feedback via crossover with soft mobility patterns on the urban level, and their historical institutionalism analysis (Roberts & Geels, potential for a growth, possibly countering the 2019). narrative of efficient urban mobility system as one, based on automobile. The MLP analysis, operating on different levels (micro-niche, meso-regime and macro-landscape) is 4.1.3. New bicycle niches and municipal employed to conceptualize the interactions between policy feedbacks two possible regimes – one, the incumbent automobility regime, and the second, the promising New institutional arrangement velomobility regime, that is represented by rapidly Several new forms of velomobility have pervaded in growing number of bicycle-centered innovations on recent years, meeting with mixed responses from the technological-vehicle level, infrastructure and policymakers: organizational level, and on practice-culture level.  Informal cycling clubs and bike sharing schemes have been formalized in many cities, mainly through entering the market. Public 211 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia transport services providers and New technological arrangements municipalities are beginning to cooperate with  E-bicycles are rapidly gaining popularity due two types of private companies: to increasing effectivity, range of use and - The communal bike clubs and bike availability. Given the high effect of hilliness sharing schemes, whose impact tend to go on the probability to cycle in general, pedelecs beyond the most basic sharing model, as are highly relevant in overcoming the distance they are more dependent and responsive to and altitude effort barriers, accounting for the the local environment. Such schemes are value of time as well. more active in provisioning additional  Cargo bicycles have achieved a small, but services (such as bike repair workshop, significant niche emergence via delivery and lectures, critical masses), and are postal services, where the impetus for change significant actors in their own respect in stems rather from operational costs and the process of planning and constructing international commitments than from local new infrastructure. They actively pressures, albeit there have been multiple participate in the planning processes, be it feasibility studies conducted for urban via consultation and expertise, helping the consolidation centres. On the individual level, municipalities e.g. with project rapid growth in effectivity and decline in the preparation – or, on the other hand, price of electric cargo-bicycles has made both lawsuits. privately owned or shared cargo bicycles - The private companies' bike-sharing more broadly available, allowing for and renting schemes draw from a market- increased flexibility. based "predict and provide" approach.  Unclear or unevenly enforced legislation, Since the rules for providing these services concerning pedelecs, scooters and other types (such as using publicly available cycle of small electric vehicles have seen the parking) vary, these services tend to vehicles' counts rise and drop sharply where achieve early market penetration and drawing pushbacks on the municipal level, or domination also via formal or informal simply falling out of the hype cycle. agreements with the local municipalities. Regarding cycling measures, the municipalities often  The sporadically filled positions of urban face internal conflicts between sectors, that represent coordinator of cycling mobility are associated different interests – local or national police with more systematic action for supporting department, building authority, land use plans, velomobility and inclusion of its facilities into political parties or traffic safety engineers, local projects (such as new development or road residents organizations and factions. reconstruction) on both on small and large scale. On the other hand, the added costs of an The crucial factor here is the non-existence of optional position are spreading the budgets of standard cycling infrastructure provision – albeit many, especially smaller cities thin; and such is codified through technical norms and without strictly formalizing the standards in increasing detail, it is not present or responsibilities, the coordinator has to battle prevalent in the streets. The scarcity of actual (any) for acceptance in the decision making – or for or high-quality cycling infrastructure is shaping both even being invited to the decision making at users' expectations and behaviour and increases all. This has been accompanied by a growing capacity for both on and off-the-road conflicts – there number of plans and strategies for the might be no "one size fits all" for cycling promotion development of cycling infrastructure for a in general – as travelling patterns, experience and growing number of cities and regions; hence requirements radically differ – but there although these also struggle with deeper certainly is one for the core safety and usability of inclusion into the land use and transport cycling infrastructure. planning in general, mainly reserving to This has been recently exemplified on a national isolated projects (cycleways, parking level, when a call for codified safe passing distance provision), rather than transforming the and other protecting measures has reverberated in general transportation plans approach and both legal, technical and public debate, following practice. such measures in neighbouring states (e.g. Germany). These arrangements have added to an increasingly Nevertheless, the rationale for the safety measure – stable and growing institutional substrate for the main reason for not passing such law, is that the applying the principles for dense and protected, impossibility of complying with the safe passing connected and direct cycling infrastructure and distance is hardwired into the current built form of cycling facilities provision, as stated in the technical the roads, and therefore unenforceable . norms. 212 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia This illuminates the main both strengthening and automobile. The crisis and subsequent recovery weakening factors of the velomobility niches. promise have allowed the incumbent coalition actors Cycling is engrained to a large extent as popular to rally round the non-controversial road leisure or sport, weekend activity, that is not in construction, rather than to succumb to its contradiction with car use (except for occasional road nevertheless growing instability (due to significant conflicts, that can be subjectively attributed to landscape pressures towards net carbon neutrality). individual behaviour) – as is after all discourse, preferred by major Czech car manufacturers. On the 5. DISCUSSION other hand, utility, everyday cycling, that enters the The lack of detail of the regression model, based on domain of both space and traffic rule organization the national level still results in several departure and affects the motorized transport flows and points, from where the current and emerging forms capacities, is considered a structural, coordinated and determinants of velomobility can be studied. network, that undermines the hard-fought victories of roads provision. Velomobility currently faces the main barrier of low, uneven infrastructure coverage, when compared with 4.1.4. Feedbacks of the road transport any other major transportation mode. This state has infrastructure development policies been sustained by perpetual spatial conflicts with both motorized and pedestrian-designated space and While most of the settlements have been developed lack of attention to the details in their realization. in close connection to busy roads, continuous increase in road transport intensities have been While several promising niches emerge, to an extent steadily pressuring local municipalities into supported on national and transnational level through negotiations on regional and national level to plan expertise sharing and financial support on European and construct bypasses and/or ringroads. Indeed, level, there appear to be several caveats, that reject presence of higher-category road is a prerequisite for the hypothesis of looming regime change or major multiple traffic calming policies, such as Low- breaks: emission zones (LEZ), but more importantly, the  The conditions for rapid growth of delayed promise of complete infrastructure allows velomobility have been arguably the most for postponing other sustainable urban mobility ideal in the recent era, due to significant policies for at least several decades. Even when the changes in mobility patterns. However, the municipalities are prone to support sustainable new influx of travellers resulting from mobility measures, the system's orientation on disruption of public transport services appears expansion of the road network is draining their to favour car-based mobility. This is partially capacity, pressuring them to join forces in the due to the structure of origin-destination coalitions in the favour of the construction of the new locations (density, structure, work/home road. Only where the variants of the possible choice considerations), resulting from long- development lead to uneven distribution of term defined and stabilized mobility patterns, externalities (such as noise levels or air pollution to a large extent appear beyond the individual levels for different localities), the actors in conflict threshold for utility cycling. reach out to technical or scientific knowledge, in  The propensity to develop new cycling order to strengthen their position – and hinder the infrastructure is to a large extent predicted by development – either way, there is no tangible contra- the urban morphology – preferring wide force to the primacy of constructing new roads as the corridors – negatively correlating with the ailment for motorized transport externalities. bicycle-friendly (i.e. dense and This has been made tangible in the omission of polyfunctional) urban morphology and cycle- velomobility (or sustainable mobility for that matter) friendly street profiles (narrower streets, from the Czech National recovery plan. Moreover, allowing for freely dispatching and crossing there has been a growth of share in the national the streets). This renders any more significant budget present for the years 2020 and 2021, that has changes towards reducing the road space in seen a record investment into transport infrastructure, favour of active modes (i.e. road diets) more prevailing over other sectors reforms, again lacking unlikely. any substantial sustainable mobility support. At the  The opportunity to take up the newly available same time, the Covid-19 pandemic has resulted in road space for newly designated cycling unforeseen losses in public transport, including the infrastructure has been largely missed and services provided or covered by municipalities that restricted to the existing plans. are continually destabilized, reducing their services  Given the low prevalence of cycling frequency, coverage, accessibility and capacity. infrastructure even in the majority of the The current crisis thus can be paradoxically identified cities, where the measures are broadly as rapidly accelerating the dominance of the supported, the form of infrastructure- 213 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia behaviour patterns has not been stabilized, [5] CEDA. (2018). Global network v2018v.6. resulting in conflicts in use (e.g. pedestrian- Czechia. cyclist conflicts on shared-use paths) and [6] CEDA. (2020). Global Network v2020.06. conflicts regarding its presence (prevalence of Czechia. efforts to scrap the newly established contra- [7] Centrum dopravního výzkumu. (2020). Česko v flow cycling roads/lanes in several cities).  pohybu. Retrieved from The existing strategies on both national and https://www.ceskovpohybu.cz/ municipal level to support urban cycling lack measurable targets and clear plans for their [8] Cervero, R. (2002). Built environments and mode choice: toward a normative framework. achievement, even where the means for Transportation Research Part D 7 , pp. 265–284. achieving the support couldn't be more clear. [9] Dendy Jr., J. E., Swartz, B., & Wendroff, B. The accelerating rate and unrelenting governmental (1977). Computing travelling wave solutions of a focus on road infrastructure growth, that dwarfs nonlinear heat equation. In J. H. Miller (Ed.), investments into cycling infrastructure, will have far- Topics in Numerical Analysis (Vol. III, pp. 447- reaching consequences for the prospects of 463). London: Academic Press. sustainable mobility. The projects will not be unbuilt; [10] Drabicki, A., Kucharski, R., & Szarata, A. the resources will not be recovered; on the contrary – (2020). Suppressing the effects of induced traffic road construction begets road construction. We in urban road systems: impact assessment with return to the Nello-Deakin's (Nello-Deakin, 2020) macrosimulation tools - results from the city of proposed turn to researching the traffic disappearance Krakow (Poland). (F. S. Esteve Codina, Ed.) as the main theoretical and practical impetus for Transportation Research Procedia, pp. 31-138. promoting the viability of cycling infrastructure doi:https://doi.org/10.1016/j.trpro.2020.03.085 planning; where even the possibility of reducing the [11] Fagnant, D. J. (2016). A direct-demand model for car utilization rates is unaccounted for, the bicycle counts: The impacts of level of service sustainable modes are silently omitted. and other factors. . Environment and Planning B: Planning and Design(1), pp. 93–107. However, there is a sliver of hope, that even if not doi:https://doi.org/10.1177/0265813515602568 imminently destabilizing, the new prospects in active and hybridized velomobility (more acceptable by the [12] Flyvbjerg, B., Holm, M., & Buhl, S. (2007, 11 26). How (In)accurate Are Demand Forecasts in regime insiders) will continue to garner popularity Public Works Projects?: The Case of and shape the mobility and built environment Transportation. Journal of the American patterns. The local initiatives are gaining momentum. Planning Association, pp. 131-146. More research is needed on whether – and how this power can be facilitated. [13] Gabrhel, V. (2019). Feeling like cycling? Psychological factors related to cycling as a This article was produced with the financial support mode choice. Transactions on Transport of the Ministry of Transport within the programme of Sciences(1), pp. 19-30. long-term conceptual development of research doi:https://doi.org/10.5507/tots.2019.006 institutions. [14] Geels, F. W. (2007). Transformations of Large Technical Systems: A Multilevel Analysis of the REFERENCES Dutch Highway System (1950-2000). Science, Technology, & Human Values(2), pp. 123–149. [1] Armoogum, J. e. (2014). Survey Harmonisation doi:https://doi.org/10.1177/0162243906293883 with New Technologies Improvement – SHANTI. Les collections de L'INRETS. [15] Geels, F. W. (2019). Socio-technical transitions to sustainability: A review of criticisms and [2] Baltus, R., & Rubbers, A. (1999). Piping elaborations of the Multi-Level Perspective. reliability improvement through passive seismic Current Opinion in Environmental supports . Nuclear Energy in Central Europe '99 Sustainability, pp. 187–201. (pp. 251-258). Portorož, Slovenia: Nuclear doi:https://doi.org/10.1016/j.cosust.2019.06.009 Society of Slovenia. [16] Goldmann, K. &. (2020). Some people feel the [3] Bílová, M., Bíl, M., & Müller, I. (2010). Critical rain, others just get wet: An analysis of regional factors in fatal collisions of adult cyclists with differences in the effects of weather on cycling. . automobiles. Accident Analysis & Research in Transportation Business & Prevention, (6), pp. 1632–1636. Management. doi:https://doi.org/10.1016/j.aap.2010.04.001 doi:https://doi.org/10.1016/j.rtbm.2020.1005 [4] Caulfield, B., & Leahy, J. (2011). Learning to [17] Hoffmann, S. W. (2017). Discontinuation of the cycle again: Examining the benefits of providing automobility regime? An integrated approach to tax-free loans to purchase new bicycles. multi-level governance. Transportation Research in Transportation Business & Research Part A: Policy and Practice, pp. 391– Management, pp. 42–47. 408. doi:https://doi.org/10.1016/j.tra.2017.06.016 214 YRS 2021 Daniel Szabó 15. -17. September A MULTI-LEVEL PERSPECTIVE ON VELOMOBILITY IN CZECH REPUBLIC Portoroz, Slovenia [18] Hommels, A. (2005). Studying Obduracy in the [27] Methorst, R. S. (2017). Can cycling safety be City: Toward a Productive Fusion between improved by opening all unidirectional cycle Technology Studies and Urban Studies. Science, paths for cycle traffic in both directions? A Technology, & Human Values(3), pp. 323–351. theoretical examination of available literature doi:https://doi.org/10.1177/0162243904271759 and data. Accident Analysis & Prevention, pp. 38-43. [19] Hopkins, D. (2017). Destabilizing automobility? doi:https://doi.org/10.1016/j.aap.2016.05.018 The emergent mobilities of generation. Ambio(3), pp. 371–383. [28] Millard‐Ball, A. &. (2011). Are We Reaching doi:https://doi.org/10.1007/s13280-016-0841-2 Peak Travel? Trends in Passenger Transport in Eight Industrialized Countries. . Transport [20] Jackman, S. (2008). pscl: Classes and Methods Reviews (3), pp. 357–378. for R Developed in the Political Science doi:https://doi.org/10.1080/01441647.2010.518 Computational Laboratory, Stanford University. 291 Department of Political Science, Stanford University, Stanford, California. Retrieved from [29] Nello-Deakin, S. (2020). Environmental http://CRAN.R-project.org/package=pscl determinants of cycling: Not seeing the forest for the trees? Journal of Transport Geography. [21] Katz, R. (1995). Modeling Bicycle Demand as a doi:10.1016/j.jtrangeo.2020.102704 Mainstream. Transportation Research Record, pp. 12-28. [30] Roberts, C., & Geels, F. (2019). Conditions for politically accelerated transitions: Historical [22] Konopásek, Z., Stöckelová, T., & Zamykalová, institutionalism, the multi-level perspective, and L. (2008). Making Pure Science and Pure two historical case studies in transport and Politics: On the Expertise of Bypass and the agriculture. Technological Forecasting and Bypass of Expertise. Science, Technology & Social Change, pp. 221–240. Human Values(4), pp. 529-553. doi:https://doi.org/10.1016/j.techfore.2018.11.0 doi:10.1177/0162243907306699 1 [23] Kouřil, P. &. (2020). Usability of Wi-Fi [31] Roberts, C., & Geels, F.W. (2019). Conditions fingerprint approach for place departure for politically accelerated transitions: Historical recognition in travel surveys. Travel Behaviour institutionalism, the multi-level perspective, and and Society, pp. 83–93. two historical case studies in transport and doi:https://doi.org/10.1016/j.tbs.2019.10.004 agriculture. Technological Forecasting & Social [24] Langhaar, H. L., & Chu, S. C. (1970). Change, pp. 221-240. Development in theoretical and applied [32] Rybarczyk, G., & Changshan, W. (2014). mecanics. New York: Pergamon. Examining the Impact of Urban Morphology on [25] Lee, K. &. (2021). Strava Metro data for bicycle Bicycle Mode Choice. Examining the Impact of monitoring: A literature review. . Transport Urban Morphology on Bicycle Mode Choice(2). Reviews(1), pp. 27–47. doi:https://doi.org/10.1068/b37133 doi:https://doi.org/10.1080/01441647.2020.179 [33] Shoup, D. (2011). The High Cost of Free 8558 Parking. Routledge. [26] Maier, K. (2012). Nástroje územního plánování k [34] Urry, J. (2004). The 'System' of Automobility. regulaci suburbanizace. Urbanismus a územní Theory, Culture & Society(4-5), pp. 25–39. . rozvoj. doi:https://doi.org/10.1177/0263276404046059 [35] Vo, T. V., & Edwards, D. R. (1994). Development of In-Service Inspection Priorities for PWR. Nucl. Technol., 106(110), 253-259. 215 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Teun Uijtdewilligen, MSc SWOV Institute for Road Safety Research PO Box 93113, 2509 AC The Hague, The Netherlands teun.uijtdewilligen@swov.nl ABSTRACT While bicycle levels in Dutch cities increased rapidly the last few years, it is still debatable what the effects of increasing bicycle volumes are on road safety. On the one hand, a safety-in-numbers effect exists, meaning that increasing cycling levels lead to a less increasing risk for cyclists. On the other hand, recent studies found a hazard-in-numbers effect, revealing that more cyclists on the road leads to a large increase in serious bicycle crashes and a slight increase in bicycle fatalities. A forthcoming study by the author contributes to this debate by investigating the effect of hourly variation in bicycle and motorised traffic volume on bicycle crashes related to type of bicycle facility. But why is hourly variaton in bicycle traffic so important to investigate road safety for cyclists in Dutch cities, the author’s PhD topic? This is discussed in the present paper. Broadly speaking, most studies related to bicycle crash risk make use of the average annual daily bicyclists and the average annual daily traffic to predict bicycle crashes, barring additional factors. This is very helpful to understand the effects of bicycle volume and motorised traffic volume on road safety for cyclists. However, such analyses do not grasp the temporal variation in daily bicycle volume, daily traffic volume and bicycle crashes. To analyse varying loads of the road network at different times of the day, and how this relates to road safety for cyclists more detailed bicycle and traffic volume data must be used. The present paper discusses how this could help for bicycle safety studies in Dutch cities, by evaluating the usefulness of hourly bicycle and traffic count data from count stations in a mature cycling city in the Netherlands. Keywords: road safety, bicycle volume, cycling, safe cycling infrastructure, urban road network 1. INTRODUCTION populated (de Jong & Daalhuizen, 2014). These new inhabitants have to find their way in the city. After years of progression, road safety becomes a Secondly, bicycle use in the Netherlands increases, growing problem for cyclists in the Netherlands especially in the four major Randstad cities (Harms again. Over the past decade, the number of bicycle et al., 2014; Jonkeren et al., 2019). As a result, the crash related casualties increased. While there were Dutch vehicle fleet becomes more bicycle, and in 162 bicycle fatalities in 2010 (25.3% of all road particular e-bike, oriented (BOVAG, 2019). These fatalities), in 2018 there were 228 bicycle fatalities two processes combined lead to an intensely used (33.6% of all road fatalities) (CBS, 2019; OECD, road network and have unknown implications for 2018; Weijermars et al., 2019). Furthermore, the road safety of cyclists. Despite these unknown number of seriously injured cyclists increased as well implications, the fact is that the number of severe and (43% of all reported serious road injuries in 2008 fatal bicycle crashes increases (CBS, 2019; against 53% of all reported serious road injuries in Weijermars et al., 2019). This shortly summarizes the 2017), especially caused by the increasing number of main topic of the author’s PhD research, which single-bicycle crashes (SWOV, 2018; Wegman et al., investigates road safety for cyclists in Dutch cities. 2012; Weijermars et al., 2019). The effect of increasing bicycle volumes on road At the same time, two trends happen in the four major safety is widely known as the safety-in-numbers Randstad cities in the Netherlands: Amsterdam, effect, as introduced by Jacobsen (2015). Safety in Rotterdam, The Hague and Utrecht. Firstly, these numbers can be explained by the non-linearity of risk cities experience rapidly increasing numbers of (Elvik, 2009). This means that the number of bicycle inhabitants and, subsequently, become more densely crashes increases proportionally less than the 216 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia increase in bicycle volume (Schepers et al., 2014). measures of exposue, which in most cases are the Thus, cyclists are safer when there are more cyclists main predictors for bicycle safety (Gomes et al., on the road (Carlson et al., 2019). Conversely, 2019; Kamel et al., 2020; Mukoko & Pulugurtha, cyclists are more vulnerable than car users, which 2020; Osama & Sayed, 2016; Strauss et al., 2015). may imply that more cyclists lead to a decrease in These are often present as the daily average bicycle road safety (Wegman et al., 2012). Several Dutch counts or the amount of bicycle kilometres travelled, studies found a large increase in severe bicycle as well as the daily average counts of motorized crashes and a small increase in bicycle fatalityies traffic, the amount of motor vehicle kilometres after a modal shift toward more bicycle use, travelled or the time being in traffic (Dijkstra, 2011). especially due to an increase in single-bicycel crashes While this is very helpful in explaining the effect of (Schepers & Heinen, 2013; Stipdonk & Reurings, bicycle volume and traffic volume on road safety for 2012; van Wee & Ettema, 2016). In Berlin, Lücken cyclists, it may not clarify the temporal variation in and Wagner (2020) even found a hazard-in-numbers bicycle volume, traffic volume, and eventually effect after rescaling from a yearly scale to a monthly bicycle crash risk. Therefore, it is necessary to use scale. These studies imply the cause of the detailed bicycle volume data as well as detailed abovementioned increasing seriously injured cyclists traffic volume data. Broadly speaking, it is assumed in the Netherlands. that detailed traffic volume data is more widely available and easier to collect compared to bicycle Although a lot has been done in the area of bicycle volume data. Therefore, in the remainder of this volumes and cyclist safety, there are still some gaps paper, the focus is on bicycle volume data. in the literature. Firstly, as mentioned above, it is still debatable what the effects of increasing bicycle First of all, one reason for why such detailed bicycle volumes are on road safety. Secondly, most safety- volume data is not always used is that this type of data in-numbers literature aims to understand its effect at is often unavailable or does not even exist. When this an aggregated level, such as a city-wide safety-in- is the case, it is important to collect data in the field numbers effect (Aldred et al., 2018; Dozza, 2017), or or to find other sources which collect such data. One even a country-wide safety-in-numbers effect (Tin way to overcome this problem is to make use of GPS Tin et al., 2013). The present study aims to persuade data. the need of very detailed bicycle and traffic data in In the last few years, using GPS data as a source for order to understand the safety-in-numbers effect on detailed bicycle volume data became more urban network level. In this way, the purpose of this widespread. Several studies explain how GPS data study is to elaborate on why it is valuable to use can be used to estimate bicycle volumes (Jestico et detailed temporal variation in bicycle volume in order al., 2016; Pogodzinska et al., 2020; Strauss et al., to carry out a proper bicycle safety study. This is 2015). Sometimes the GPS data is collected by the especially true for the amount of bicycle traffic in authors themselves, while in other studies crowd Dutch cities, which fits in the scope of the author’s PhD research: ‘ sourced data is used. Sources like the sports tracker Road safety for cyclists in Dutch app STRAVA are well-known sources for crowd cities’. The detailed traffic data used in the present sourced bicycle GPS data. In many cases, this data is study comes from permanent count stations for then calibrated by using data from permanent bicycle cyclists in Utrecht, while the hourly traffic data count stations or pneumatic tubes. Another method comes from traffic counts at traffic lights. Bicycle can be to only use data from count stations and crash data is derived from the Dutch national predict bicycle volumes in the surrounding areas by database for traffic crashes: BRON. the use of a transport model (Beitel et al., 2017). In the remainder of this paper three arguments will be The result of collecting detailed bicycle volume data discussed about how the temporal variation in bicycle and its comparison to traffic volume data is shown in traffic and motorized traffic play a role in explaining Figure 1. The figure illustrates an average week on an biycle crash frequency in Dutch cities. Firstly, hourly scale from a permanent bicycle count station variaton in bicycle volume during the day and on in Utrecht in the Netherlands. The pattern of the different weekdays will be discussed. Secondly, the bicycle volume shows a similar pattern as the traffic unique case of the four major cities in the Netherlands volume. Although, it seems that motorized traffic has will be discussed in terms of bicycle volumes. Lastly, a larger afternoon peak and during midday the it is showed how hourly variation in bicycle and volume does not decrease as much as the bicycle motorized traffic relates to bicycle crash frequency. volume. Furthermore, the pattern clearly shows the 2. HOURLY VARIATION IN BICYCLE morning and the afternoon peak hours on the weekdays and an overall peak during the shopping VOLUME times on the weekenddays. What can be seen in the In the majority of bicycle road safety research, the figure as well is that bicycle volume varies during the annual average daily bicyclists (AADB) and the day. This is important information because when the average annual daily traffic (AADT) are used as goal is to investigate different loads of traffic in a 217 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia network, such kind of data is very useful. Especially Station for trains, as well as the city its main bus when its effects on road safety for cyclists is studied. terminal and main tram station. A large share of users of this public transport hub uses cycling as an access or egress mode. On the other hand, an inner-city ring road attracts more motorized traffic compared to cyclists, because these roads are well-connected to the highways and connect different neighborhoods. Therefore, traffic volume levels are higher in Figure 1 and bicycle volume levels are higher in Figure 2. In order to compare Figure 2 to other European cities, Figure 3 shows the modal split for twenty major European cities. This figure illustrates that only Figure 1: Average hourly bicycle volume and traffic Copenhagen has a larger share of cyclsits than from a permanent count station in the Venuslaan in Utrecht and most other European cities are very low Utrecht, the Netherlands in their share of cycists. While it is not hourly data, it may give an idea of how the composition of traffic in 3. DUTCH BICYCLE LEVELS these cities looks like. This illustrates why Dutch While bicycle levels between different countries in cities are a special case compared to urban areas from the world may vary, the Dutch situation is something other European countries. Moreover, this may clarify special compared to most other countries. In the most the large difference in bicycle traffic and motorized urbanised areas of the Netherlands, the Randstad with traffic in Figure 2. its capitals Amsterdam, Rotterdam, The Hague and Utrecht, almost 40% of the trips between 1 and 7 kilometer is traveled by bicycle in 2015 (Jonkeren et al., 2019). This is a large share compared to other modalities. On street level, this may result in a large difference between motorized traffic and bicycle traffic. Figure 2 shows the average hourly bicycle volume compared to the average hourly traffic volume in 2018 for a single permanent count location in Utrecht. The AADB for this location is 16,536 while the AADT is 10,852. It seems that bicycle volume may exceed traffic volume. Figure 3: Modal split in percentage of trips in twenty large European cities (EPOMM, 2019). * PT: Public transport Figure 2: Difference in bicycle volume and traffic volume on street level on the Amsterdamsestraatweg The Netherlands is unique in terms of bicycle traffic in Utrecht, the Netherlands and infrastructure. Cycling levels in most other countries are incomparable due to too low bicycle The difference between Figure 1 and Figure 2 can be volumes compared to the Dutch bicycle demand. explained by the location of the permanent count Moreover, the Netherlands has a well-designed station. While the location of the count station from bicycle network containing more bicycle facilities Figure 2 is close to the city centre and Utrecht’s main than any other country in the world, resulting in an public transport hub, the count station from Figure 1 unique cycling environment. Therefore, studying is located further away from the city centre and lies bicycle traffic and its safety in the Netherlands needs on an inner-city ring road. The city centre in Utrecht a different approach and more detailed bicycle and attracts large numbers of cyclists as a destination for traffic data . cyclists itself, and because it contains the main corridor to the Utrecht Science Park for cyclists. Moreover, the the public transport hub is located next to the city centre and contains the main Central 218 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia 4. DIFFERENCE IN BICYCLE CRASH outpace motorized traffic in Utrecht. This may be of NUMBERS special interest when looking to a safety-in-numbers effect for cyclists in Dutch cities. Therefore, hourly As discussed above, bicycle volumes vary during the variation in bicycle traffic, motorized traffic and day and during different days of the week. How does bicycle crashes is an important perspective to this apply for bicycle crashes? Do they follow a investigate road safety for cyclists in Dutch cities. similar distribution during the week or do they differ? A first look on aggregated hourly bicycle crashes 5. CONCLUSIONS from 2015-2019 in Utrecht creates Figure 4. The figure illustrates how bicycle crashes in Utrecht The present paper tried to argue why hourly variation follow the distribution of cyclists and motorized in bicycle traffic, motorized traffic and bicycle traffic during the days of the week. crashes is important in explaining road safety for cyclists in Dutch cities. Three arguments were given. Firstly, detailed bicycle volulme data sheds light on how the amount of cyclists varies during the day and between different days of the week. This may give other results than looking to average annual daily bicycle traffic when trying to explain bicycle road safety. Secondly, compared to other countries, the Netherlands is a special case in terms of bicycle demand. The share of bicyclists in the four major Dutch cities for short trips is higher than for cars. This relates to the difference in hourly bicycle traffic and Figure 4: Average hourly bicycle volumes and traffic hourly motorized traffic at street level. In this way, a volumes for seven permanent count stations in study about road safety for cyclists in Dutch cities Utrecht, including hourly bicycle crashes needs a different approach compared to studies using According to this figure, it may be indicated that conventional crash frequency models. The temporal hourly variation in exposure plays a large role in aspect may be very important to include and this explaining bicycle crashes in Dutch cities. On needs a different perspective. Lastly, it is made clear weekdays, when exposure increases during rush that for most hours in a week the distribution of hours, bicycle crashes increase as well. Whereas bicycle crashes follows the distribution of bicycle during off-peak hours and in the evening and nights, volume and traffic volume. Therefore, hourly crash frequency is much lower. At these times, exposure data may help to grasp the temporal aspect especially in the evening and at night, exposure has in road safety for cyclists in Dutch cities. Although, also decreased compared to daytime hours. This is other factors may play a role as well. especially true when night times are compared to The results from this study are helpful for transport peak hours. planners from municipalities. These insights may be On weekenddays, a similar pattern is found, although an argument to even more try to reduce bicycle at different times. During shopping hours, both crashes, especially during peak hours. One solution exposure and bicycle crash frequency increase, while could be to focus on disentangeling cycling in the evening and during night times both decrease. infrastructure and motorized traffic infrastructure One remarkable issue is the large crash number on (Schepers, 2013). This leads to separating cyclists Friday night in relation to the low bicycle and traffic from heavy and fast motorized traffic. Another volume during these hours. This is the moment where solution may be to decrease maximum speeds of other factors than exposure start to play a role. For motorized traffic. In the Netherlands, there is an this particular case, alcohol may be part of the ongoing discussion to decrease all maximum speeds explanation as Friday nights are often the nights in urban areas to 30 km/h. This speed is seen as a safe when people go to bars and clubs. Interestingly, speed limit for vulnerable road users (Twisk et al., similar findings come from a study by Dozza (2017). 2013). He found that during weekend nights, crash numbers The findings in this paper are useful in further cannot be explained solely by exposure anymore. research to road safety in Dutch cities, the author’s However, one drawback of this study is that the PhD topic. In this project, the presented data from the author only used bicycle volume data and tries to present study will be input for a crash frequency explain bicycle-motorized vehicle crashes. model where the temporal aspect is the main interest. Interestingly, Figure 4 also shows that the average Instead of using daily average traffic and bicycle bicycle volume of seven permanent count stations counts for specific road segments, hourly estimates exceeds the average traffic volume for the same are the guiding principle. For every road link, it is locations during some time frames. This is in line known how, for an average week, the hourly with what Figure 2 revealed, that cyclists may variation in cyclists and traffic looks like. 219 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia Furthermore, for the bicycle crashes it is also known [8] Dozza, M. (2017). Crash risk: How cycling flow in which hour and on which day of the week they can help explain crash data. Accident Analysis & occured. This information will be aggregated. In this Prevention, 105, 21-29, way, the whole road network becomes one road https://doi.org/http://dx.doi.org/10.1016/j.aap.20 16.04.033 segment with an aggregated hourly bicycle count, traffic count and bicycle crash count. The resulting [9] Elvik, R. (2009). The non-linearity of risk and the data frame therefore has 168 rows, being every hour promotion of environmentally sustainable in a week (Monday 00:00-01:00 till Sunday 23:00- transport. Accident Analysis & Prevention, 41(4), 00:00). This will be the starting point for the 849-855, modeling in the author’s PhD research, where a https://doi.org/https://doi.org/10.1016/j.aap.200 9.04.009 Generalized Linear Model will be used. [10] EPOMM. (2019). TEMS - The EPOMM Modal In conclusion, in order to find out if more cyclists on Split Tool. the road leads to decreasing road safety for cyclists, http://tems.epomm.eu/compare_cities.phtml it is necessary to take hourly variation in bicycle [11] Gomes, M. M., Pirdavani, A., Brijs, T., & Souza volume and traffic volume into account. Especially Pitombo, C. (2019). Assessing the impacts of for cases where bicycle use exceeds car use, like in enriched information on crash prediction Dutch cities. The analyzes in this study make it performance. Accident Analysis and Prevention, necessary to examine correlations between hourly 122, 162-171, bicycle volume, traffic volume and bicycle crashes in https://doi.org/10.1016/j.aap.2018.10.004 more detail. This will derive statements concerning [12] Harms, L., Bertolini, L., & te Brömmelstroet, M. bicycle road safety in urban areas, which is part of the (2014). Spatial and social variations in cycling author’s PhD thesis. patterns in a mature cycling country exploring differences and trends. Journal of Transport and REFERENCES Health, 1(4), 232-242, https://doi.org/10.1016/j.jth.2014.09.012 [1] Aldred, R., Goodman, A., Gulliver, J., & Woodcock, J. (2018). Cycling injury risk in [13] Jacobsen, P. L. (2015). Safety in numbers: more London: A case-control study exploring the walkers and bicyclists, safer walking and impact of cycle volumes, motor vehicle volumes, bicycling. Injury Prevention, 21(4), 271-275, and road characteristics including speed limits. https://doi.org/10.1136/ip.9.3.205rep Accident Analysis and Prevention, 117, 75-84, https://doi.org/10.1016/j.aap.2018.03.003 [14] Jestico, B., Nelson, T., & Winters, M. (2016). Mapping ridership using crowdsourced cycling [2] Beitel, D., McNee, S., & Miranda-Moreno, L. F. data. Journal of Transport Geography, 52, 90- (2017). Quality measure of short-duration 97, bicycle counts. In Transportation Research https://doi.org/https://doi.org/10.1016/j.jtrangeo. Record (Vol. 2644, pp. 64-71). 2016.03.006 [3] BOVAG. (2019). Zonnig 2018 stuwt omzet [15] Jonkeren, O., Wust, H., & de Haas, M. (2019). fietsbranche naar record. Mobiliteit in stedelijk Nederland. The Hague: https://mijn.bovag.nl/actueel/nieuws/2019/maart Kennisinstituut voor Mobiliteitsbeleid. /zonnig-2018-stuwt-omzet-fietsbranche-naar- record [16] Kamel, M. B., Sayed, T., & Bigazzi, A. (2020). A composite zonal index for biking [4] Carlson, K., Ermagun, A., Murphy, B., Owen, attractiveness and safety. Accident Analysis and A., & Levinson, D. (2019). Safety in Numbers Prevention, 137, for Bicyclists at Urban Intersections. In https://doi.org/10.1016/j.aap.2020.105439 Transportation Research Record (Vol. 2673, pp. 677-684). [17] Lücken, L., & Wagner, P. (2020). On the relation between bicycle volumes and individual risks for [5] CBS. (2019, 18 April 2019). Overledenen, doden bicyclists in Berlin. Journal of Transportation door verkeersongeval in Nederland, wijze van Safety & Security, 12(1), 210-223, deelname. https://doi.org/10.1080/19439962.2019.1597003 https://opendata.cbs.nl/statline/#/CBS/nl/dataset /71936ned/table?ts=1574759336784 [18] Mukoko, K. K., & Pulugurtha, S. S. (2020). Examining the influence of network, land use, [6] de Jong, A., & Daalhuizen, F. (2014). De and demographic characteristics to estimate the Nederlandse bevolking in beeld: verlden, heden, number of bicycle-vehicle crashes on urban toekomst. The Hague: Planbureau voor de roads. IATSS Research, 44(1), 8-16, Leefomgeving. https://doi.org/10.1016/j.iatssr.2019.04.001 [7] Dijkstra, A. (2011). En route to safer roads: How [19] OECD. (2018). Road Safety Annual Report road structure and road classification can affect 2018: The Netherlands. Paris: International road safety [Doctoral thesis, University of Transport Forum/OECD. Twente]. Leidschendam: SWOV. 220 Teun Uijtdewilligen YRS 2021 PLACING HOURLY VARIATION IN BICYCLE VOLUME AND CRASHES IN THE BIGGER 15. -17. September PERSPECTIVE OF ROAD SAFETY FOR CYCLISTS IN DUTCH CITIES Portoroz, Slovenia [20] Osama, A., & Sayed, T. (2016). Evaluating the [26] Strauss, J., Miranda-Moreno, L. F., & Morency, impact of bike network indicators on cyclist P. (2015). Mapping cyclist activity and injury safety using macro-level collision prediction risk in a network combining smartphone GPS models. Accident Analysis & Prevention, 97, 28-data and bicycle counts. Accident Analysis and 37, Prevention, 83, 132-142, https://doi.org/https://doi.org/10.1016/j.aap.201 https://doi.org/10.1016/j.aap.2015.07.014 6.08.010 [27] SWOV. (2018). Ernstig verkeersgewonden in [21] Pogodzinska, S., Kiec, M., & D’Agostino, C. Nederland. In. The Hague: SWOV. (2020). Bicycle Traffic Volume Estimation [28] Tin Tin, S., Woodward, A., & Ameratunga, S. Based on GPS Data. Transportation Research (2013). Incidence, risk, and protective factors of Procedia, 45, 874-881, bicycle crashes: Findings from a prospective https://doi.org/https://doi.org/10.1016/j.trpro.20 cohort study in New Zealand. Preventive 20.02.081 Medicine, 57(3), 152-161, [22] Schepers, P. (2013). A safer road environment https://doi.org/https://doi.org/10.1016/j.ypmed.2 for cyclists [Doctotal Thesis, Delft University of 013.05.001 Technology]. Leidschendam: SWOV. [29] Twisk, D. A. M., Vlakveld, W. P., Dijkstra, A., [23] Schepers, P., Hagenzieker, M. P., Methorst, R., Reurings, M., & Wijnen, W. (2013). From van Wee, B., & Wegman, F. (2014). A bicycle crashes to measures. Leidschendam: conceptual framework for road safety and SWOV. mobility applied to cycling safety. Accident [30] van Wee, B., & Ettema, D. (2016). Travel Analysis and Prevention, 62, 331-340, behaviour and health: A conceptual model and https://doi.org/10.1016/j.aap.2013.03.032 research agenda. Journal of Transport & Health, [24] Schepers, P., & Heinen, E. (2013). How does a 3(3), 240-248, modal shift from short car trips to cycling affect https://doi.org/https://doi.org/10.1016/j.jth.2016. road safety? Accident Analysis & Prevention, 50, 07.003 1118-1127, [31] Wegman, F., Zhang, F., & Dijkstra, A. (2012). https://doi.org/http://dx.doi.org/10.1016/j.aap.20 How to make more cycling good for road safety? 12.09.004 Accident Analysis & Prevention, 44(1), 19-29, [25] Stipdonk, H., & Reurings, M. (2012). The Effect https://doi.org/https://doi.org/10.1016/j.aap.201 on Road Safety of a Modal Shift From Car to 0.11.010 Bicycle. Traffic Injury Prevention, 13(4), 412- [32] Weijermars, W., de Goede, M., Goldenbeld, C., 421, Decae, R. J., Schepers, P., Wijlhuizen, G. J., https://doi.org/10.1080/15389588.2012.660661 Mons, C., Hoekstra, A. T. G., & Dijkstra, A. (2019). Monitor verkeersveiligheid 2019 - Achtergrondinfromatie en onderzoeksverantwoording. The Hague: SWOV. 221 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN UNSTRUCTURED ENVIRONMENTS Alfredo Valle Barrio, Phd Student in Artificial Intelligence Alberto Cruz Ruiz, Master in Artificial Intelligence Victor San Telesforo García, Degree student in Computer Science José Eugenio Naranjo Hernandez, Professor in the UPM Felipe Jiménez Alonso, Full Professor in the UPM INSIA, UPM, Campus Sur UPM Carretera de Valencia km 7, 28031, Madrid, España alfredo.valle@upm.es ABSTRACT Today we are used to hear news related to vehicle automation and its integration on roads in different parts of the world. These news items mainly concern the automation of highly sensorised cars for conventional roads, which must be perfectly signposted and maintained. However, there is still a long way to reach this point in commercial vehicle automation, especially in their use in unstructured environments. The proposal reflected in this paper is the automation of an industrial vehicle, specifically a Volvo A25, to use it in unstructured environments such as the excavation of a tunnel. The design is focused on carrying out automatic debris removal tasks without need a driver or supervisor. To this goal, the essential vehicle functions have been automated, such as speed control, steering, gear change, and tilting the dump box, which allows the system to have total control over the vehicle's capabilities. In addition to this, it has been equipped with an inertial-differential GPS that provides it with the capacity of sub metric self-location. Thanks to this GPS, the vehicle has the capacity to circulate autonomously on a pre-recorded map, maintaining its trajectory. In the tests carried out on the INSIA track, the maneuvering capacity that has been implemented in the vehicle has been verified, obtaining as a result that it is capable of driving for an unlimited time on a pre-recorded map, carrying out the controlled braking maneuvers and resuming the route, either due to an event defined or due to the intention of an operator to control the vehicle remotely. Keywords: Connected and automated vehicles, autonomous vehicles, Intelligent transportation systems, Robot Operating System (ROS) Research domain: The work included in this paper is closely related to two of the seminar's research topics. These are firstly Intelligent transport systems ITS & Traffic management. Secondly, Transport civil and road engineering. 1. INTRODUCTION was due to the difficulty and danger of drilling walls for explosive charges or spraying concrete. The application of autonomous vehicles is expanding its frontiers beyond those seen in the Gradually, we are starting to hear from commercial news, passenger transport. There are many areas vehicle construction companies that are more where they can be applied, such as agriculture, frequently incorporating driver assistance systems military scenarios, or public works (Jiménez, 2018). or low-level automation. These companies' problem The use of autonomous machinery in such is that trucks or equipment of this type tend to have environments was strictly restricted to specific a relatively long service life, making upgrading to repetitive tasks requiring machine movement. This new automated vehicles difficult. 222 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia Companies such as Aurora, Caterpillar, and Virgin the Polytechnic University of Madrid. These Hyperloop have already shown their intention to facilities are located in the south of Madrid, Spain, introduce industrial vehicles in unstructured at the junction of the M.40 road with the A-3 environments such as mines or excavations. Some of motorway. Figure 2 shows an aerial image of the the drawbacks of these private companies' proposals INSIA on whose test track can be distinguished in are economical, as they sell a closed system that is the image next to the M-40 road, where the tests not very flexible and requires a significant have shown at the end of this article has been carried deployment of communications infrastructure. out. Due to the complexity and cost of developing this Another of the problems encountered when working type of vehicle and technology, more and more with such a large vehicle is that the track is too small, projects are attempting to take automation outside as it has a diameter of just 40 meters at its widest the field of passenger transport. One of these is the part, which does not leave much room for TUNNELAD project, which includes the maneuvering. development of this article. This is a pilot project for Although the INSIA test track is not designed to the development of autonomous industrial vehicles accommodate and maneuver such large vehicles, the for unstructured environments. availability of this track on the outskirts of Madrid This article wants to show the possibility of has allowed to carry out many tests, especially those automating such a commercial vehicle to a high of vehicle guidance by GPS, without having the level of automation that requires minimal interaction problem that the urban canyon situation poses for the with personnel (Jiménez et al., 2020). This allows GPS. The possible implementation of this project the vehicles to be reused for a long time while will be carried out in open mines for the most part, increasing safety (Pereira de Oliveira et al., 2020) in so GPS is essential. the hazardous environment in which they work. Economically, it reduces maintenance costs, personnel costs, and potential accidents. It is also highlighting noting that this type of installation does not require infrastructure deployment as all the equipment is embedded in the vehicle, which is a requirement in many public works projects. This article will detail the architecture, both hardware and software, developed for this project (Naranjo et al., 2020) and the result of the automation and guidance tests carried out on the south campus of the Polytechnic University of Madrid at the INSIA. Figure 2: Bird view of INSIA The project's natural development requires to test the 2. PLATFORM AND TEST TRACK system in a more realistic environment, so the The work developed in this paper has been vehicle's transport and testing at the project implemented on an industrial vehicle, a Volvo A-25 contractor's facilities in Noblejas, Toledo, Spain, has truck, as shown in Figure 1. The automation of this already been planned. vehicle has been a challenge due to its large cargo, More demanding tests will be carried out at these size, and, precisely, the way it turns, since, in this facilities. They will include autonomous driving, as case, the wheels do not behave like those of a typical shown in this article, loading/unloading the vehicle, vehicle. The wheels move in solidarity with the head taking control by an operator during vehicle to turn the vehicle, so it pivots on its axis. operation, and even simulating entry into a tunnel without GPS coverage, where the vehicle will have to be guided by Lidar. In this new facility, the vehicle will face several challenges. On the one hand, speed control, having to overcome steep ascents and descents, whether the vehicle is loaded with up to 25 tons of earth. The difficulty of keeping such a large vehicle in a confined space at a considerable speed of up to 40 Figure 1: Volvo A-25, vehicle to automate km per hour. All this during a 2 km journey where interaction with an operator will be minimal. This project's development and testing have been carried out at the INSIA facilities, which belong to 223 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia 3. AUTOMATION AND HARDWARE ARCHITECTURE For this project, complete automation of the vehicle has been carried out, consisting of the following sections:  Steering  Speed - Throttle - Brake  Gears  Dump box The following sections will show in detail how each actuator's automation required for the vehicle's complete control has been carried out. 3.1. Steering To control the vehicle's steering, (Jiménez et al., Steering position 2013) (Alonso et al., 2014) a device has been sensor attached to the steering column. On the upper side of Figure 3 shows the patented device that will perform the steering wheel turning functions. This has an electromagnet that will act as a magnetic clutch to take control of the steering. The lower side of the image shows how the device is attached to the steering rod. A signal controls this device through a CAN BUS Figure 3: Detail of the steering automation assembly that will indicate the angle at which it should turn. This angle is calculated based on the reading of a 3.2. Speed permanent turn sensor attached to the steering rod, The speed control of the vehicle is done by because the reading of this position is not available controlling two actuators. On the one hand, the brake through the vehicle's CAN. The magnetic clutch is control and, on the other hand, the throttle control. controlled by a digital signal obtained from the The automation of these actuators is described engine itself, activated or deactivated via a CAN below. BUS signal. 3.2.1. Brake The brake control is pretty similar to the steering wheel control. The same motor is used in the steering control device. The brake pedal is actuated mechanically using a chain. In contrast to the steering control device, the brake pedal's pressure does not require measuring the brake pedal position. This is because its actuation is controlled dynamically by a PID from the speed value. 224 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia 3.3. Gears Since the vehicle will have to perform complex manoeuvres, including changing gears to make tight turns or load/unload the truck, the gears' control has been automated. In the case of this vehicle, gear shifting is effortless to automate. Although the gearshift lever may look like a too mechanical device, it is a digital signal selector that tells the control unit which gear to select. Figure 6 shows the relay board's control of the gears with a tester and the final installation inside a watertight box with standard connections. The Figure 4: Detail of brake assembly control of these relays is carried out employing CAN BUS communication that selects the relays' state of 3.2.2. Throttle excitation. In this vehicle's case, the accelerator pedal is a potentiometer that generates a voltage proportional to the pedal position. To control the throttle and return control to an operator when necessary, a set of relays selects the original throttle signal's sending or the signal provided by an analogue board. Figure 5 shows the INSIA designed board, which has two inputs corresponding to the throttle potentiometer signal and the analogue board signal, Relays test and an analogue output to be sent to the vehicle board ECU. Figure 6: Installation of the gear control relays 3.4. Dump box Dump box control has been made with a stepper motor, coupled to a device consisting of a worm screw and two metallic guides to avoid oscillations. The worm screw has an associated anchorage to the dump box control lever, making the dump box move in conjunction with the motor's rotation. An end-of- Figure 5: Throttle signal bridging board stroke sensor has also been installed to calibrate the bowl positions. 225 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia The assembly of this device can be seen in Figure 7. In the lower right part of the image, we can see an Observing the image, it can be seen that the anchor electronic card that functions as a power distributor. that joins the control lever with the worm screw is From this card, the power supplies of all the motors not welded to the lever but grips it in such a way that and the computers are controlled. it can be unscrewed to carry out manual control of The electronic card anchored to the case's cover is a the box. card designed at INSIA, responsible for monitoring and managing the CAN BUS control and guaranteeing a possible emergency always stop. Finally, a fanless computer is installed on top of the Stepper suitcase, where the system's software control is motor installed. This oversees the necessary commands to the actuators and, in turn, executes the high-level systems such as waypoint and later the obstacle detection by Lidar. This computer is connected via Ethernet to a switch that manages the connections between the two Ethernet-CAN cards and the computer. 4. SOFTWARE ARCHITECTURE Figure 7: Dump box tooling The entire software architecture has been implemented in the ROS 2.0 framework, in its 3.5. Control Board "eloquent" version. This framework has been used To unify the control of all the devices mentioned as the basis for the entire development because above, they are connected via a single CAN BUS to modular, scalable, and standard implementation the control board, which can be seen in Figure 8: develop will be used in all our vehicles. Control suitcase. ROS provides several operational advantages because of its modular functionality. On the one hand, it provides the usual features of an operating DIO-AIO system, such as hardware abstraction and message passing between processes. Simultaneously, it output abstracts from the computer's operating system, which is especially important since several computers, possibly with different operating systems, will be used to develop the project. On the other hand, it allows having several people PIC develop code for the exact vehicle, specializing in a specific model and only worrying about maintaining consistency in the name of the topics. In the case of this project, we will use ROS on the Ubuntu 18.04 operating system. We have chosen CAN-Ethernet this operating system because it is the one that works best with ROS and at the same time allows better Cards management of resources and networks. It is also a much lighter operating system than Windows, which allows to use less power and consequently cheaper computers. In the following, the implemented modules will be Power developed, and the data flow between them. The board system consists of 5 ROS packages and seven nodes divided as follows: Figure 8: Control suitcase  CAN Package This suitcase consists of one hand of 2 Ethernet- - Control Node CAN cards. One for collecting data from the  Decision Package vehicle's CAN, precisely the speed, and the other for - Decision Node transmitting data to the devices mentioned above.  Sensors Package 226 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia - GPS Node use the reader on the lever and not on the relay  Waypoints Package board. - Waypoints Node Finally, in the case of the tank, it sends a CAN - Recorder Node  packet to the arduino board that controls the stepper Teleoperation Package motor's movement coupled to the tank movement - Keyboard Node lever. Unlike the previous devices, the dump box - Tablet Node cannot be enabled and disabled by software. The Figure 9 shows the general system scheme and its lever is mechanically associated with the motor, so data flow. if you wanted to disable the vat control, you would have to disassemble it manually. The steering, throttle, and brake commands are controlled at 20 Hz, while the gears and dump box operate at 10 Hz. The maximum frequency at which CAN packets are sent from all devices together has been calculated. This does not coincide with the frequencies mentioned above because these frequencies only control the devices, not the number of packets required to carry out this control. The frequency at which packets should be sent to the Ethernet-CAN card has been defined as 300 Hz. 4.2. Decision Package Figure 9: Overview of software architecture The decision package contains only the decision 4.1. CAN Package node. It has two main functions. The CAN package contains only one node, which is Firstly, it is responsible for collecting the speed, the control node. It must open TCP connections to direction, and gear setpoints from those nodes that the Ethernet-CAN cards to transmit messages or publish these parameters. Secondly, it is responsible receive messages from the vehicle. It is also for sending a single speed/steering/gear setpoint to responsible for converting the speed and direction the vehicle's low-level control at a constant commands to values that the actuators can frequency through a predefined priority system. understand. This frequency has been empirically adjusted on the The low-level control performs, as the following: vehicle in such a way as to allow rapid action on the vehicle control without saturating the control line To send steering commands, it takes the encoder's components (ethernet-can cards, motors, and low- value coupled to the steering rod. It generates a CAN level computer). packet encapsulated in a TCP packet, which contains the instructions for a relative rotation of the motor 4.3. Sensors Package concerning the current position. Enabling/disabling Currently, only the GPS node is implemented in this the steering control is done by the same CAN package, including sensors such as Lidar, Mobileye packets that enable/disable the digital signals camera, or stereo camera later. controlling the electromagnet coupled to the steering link. The GPS node connects via a TCP client to even differential sub-metric GPS, which continuously To send speed commands, it has a PID controller, provides us with the vehicle's position at a which controls the throttle and brake. In the throttle configurable rate of between 10 and 20 Hz. Besides, case, a packet is sent to the analogue card that it provides various signal quality parameters to controls the signal sent to the ECU. In the brake identify whether we are entering a tunnel or safe to case, a packet similar to that of the steering motor is continue autonomous driving. sent, but in this case, in absolute format. In the brake, disabling the speed control means moving the brake 4.4. Waypoints Package to the zero position, not pressed. In the throttle case, it is rerouting the signal from the accelerator pedal The waypoints package is composed of 2 nodes, the back to the ECU. recorder node, and the waypoint node. The gear commands are sent via a CAN packet to 4.4.1. Recorder Node the board controlling the relay board that selects the gear engaged. To enable/disable the gear control, a The recorder node has as input the GPS position of CAN packet is sent to tell the gear control board to the vehicle and a processing trigger flag. 227 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia The node collects all messages received from the vehicle. From these points, a Bézier curve is GPS node and checks if the received position is inferred, which will provide us with the target point within a minimum distance from the last saved at which we have to focus the vehicle. With this position. This distance is manually defined in the Bézier curve, we get the vehicle to make the turns node configuration file. more smoothly and continuously. Once the trigger flag is enabled, it continuously To finish calculating the target point, this is obtained saves the vehicle's GPS position and its speed in a by segmenting the Bézier curve obtained in 4 text file. The file follows its own structure defined in segments and defining the target point as the third the code, although it is easily configurable. point of the curve. Finally, the recording has two modes of operation. The vehicle trajectory's error angle with respect to Static speed recording means that the recording will the target point is calculated to finalize the path record a fixed speed in the entire route's file. tracking. This angle is obtained by trigonometric Recording with dynamic speed means that the map calculations corresponding to the following formula: will be recorded with the same speed as it is travelled during the recording. It is respecting the margins 𝛼 defined in the configuration file. (2) 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒_𝑡𝑜_𝑡𝑎𝑟𝑔𝑒𝑡_𝑝𝑜𝑖𝑛𝑡 = 𝑎𝑠𝑖𝑛 ( ) 4.5. Waypoints Node 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒(𝑡𝑎𝑟𝑔𝑒𝑡_𝑝𝑜𝑖𝑛𝑡, 𝑐𝑢𝑟𝑟𝑒𝑛𝑡_𝑝𝑜𝑖𝑛𝑡 ) The Waypoints node is in charge of loading and In addition, the Waypoint node has the task of replaying the route recorded by the recorder node. It generating the acceleration ramp at the start of the also performs the path tracking calculation to keep journey and the deceleration ramp at the end of the the vehicle on the previously recorded path. journey. The operating cycle of this node is the following: 4.6. Tele-Operation Package Firstly, the selected map is loaded using a topic. In The teleoperation package contains two nodes. On case that a map has not been selected, a warning is the one hand, the tablet node allows the vehicle to be emitted, and a default map is loaded. controlled from an external application not Next, a self-localization on the map is carried out. implemented in Ros, hosted on a Windows tablet. To do this, the closest point on the map to the current On the other hand, a keyboard node allows us to test position is searched for. Two situations can occur at the systems' operation. this point. 4.6.1. Tablet Node If the Waypoints has just been started and there is no previous position of the vehicle to calculate its The tablet link node is used to open a series of UDP trajectory, the closest point on the map to the connections to an external application, not vehicle's current position will be searched. If we do implemented in ROS, hosted on a Windows 10 have information of the trajectory followed by the tablet. vehicle, the nearest point on the map will be This node allows high-level control of the vehicle's searched within the following n points from the last operation. Its capabilities include: nearest point found. If none of these points meets the validation range as the closest point, the closest  Enable/disable autonomous steering wheel point on the map will be chosen, even if it does not control. meet the validation range.  Enable/disable autonomous speed control  Emergency braking The next step is to calculate the target point to which  Remote control of speed and steering wheel the vehicle is to be directed. To do this, the target rotation distance or look ahead is calculated, which is  Select the route to follow defined by the following formula:  Activate autonomous guidance  Make new map recordings 𝑙𝑜𝑜𝑘𝑎ℎ𝑒𝑎𝑑 = (𝑚 ∗ 𝑐𝑢𝑟𝑟𝑒𝑛𝑡𝑠𝑝𝑒𝑒𝑑) + 𝑛 (1) Once the system is launched on the low-level Where m and n are manually defined parameters in computer, the Tablet has total vehicle control, and it the configuration file and change based on the is a vehicle information interface. The connection vehicle. Moreover, current_speed is the current with the ROS node is made through a wifi access vehicle speed collected from the BUS CAN. point installed in the vehicle so that the Tablet can connect to any automated vehicle within its wifi Once the target distance is calculated, those points signal range. between the vehicle's current position and this one are collected, always following the trajectory of the 228 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia There are two scenarios if wifi is disconnected. In constant speed throughout the journey, and brake case that the vehicle is working in autonomous gently at the end of the journey. mode, nothing happens. The vehicle will continue The steering control is judged on its ability to follow driving until the end of the route and it will stop the line marked by the route. It includes two sharp automatically there of if it detects any obstacle in the left turns, but the real challenge is to follow straight route. In case that the wifi connection is broken lines with little margin for lateral error. during while the vehicle is manually controlled, through the tablet, it will stop until the connection is Finally, the gear changing system demonstrates how fixed. it works: at the start of the journey, the vehicle starts in neutral gear, automatically shifts into direct gear, In dynamic tests carried out in Noblejas, it has been and at the end of the journey, it must be shifted back possible to control the vehicle at around 50 m if there into neutral gear. is direct visibility between the vehicle and the Tablet. The dump box operation is not tested because it is always carried out at a standstill and under an 4.6.2. Keyboard Node operator's supervision outside the vehicle. Keyboard node contains a minimal implementation 5.2. Results of the controls necessary to operate the vehicle. It allows to control the speed, the steering wheel The results obtained are shown below, which show rotation, the control of the gears, and the dump box's the following route superimposed on the recorded operation with the keyboard arrows and some route, the speed profile followed to check the correct alphanumeric keys. functioning of the throttle and brake control and finally the steering wheel turning profile. 5. TEST AND RESULTS The following route is shown in Figure 11. The In this section, the results of the tests carried out at vehicle starts at a different point from the beginning INSIA will be shown. These tests have allowed us to of the route, it calculates its trajectory to return to the check the system's good functioning during the route and follows it until the end of the route where initial development phase, and later we have used it stops a few meters before the end point. them to check the system's accuracy and The vehicle makes two left turns, which, being very performance. tight for the size of it, it tends to take them a little on 5.1. Test track the inside, although within the permitted margins. In the image is a change of colour in the route followed The route followed for the tests reflected in this showing the change of speed. paper is as shown in the Figure 10. Figure 11: Followed route Figure 10: Recorded route Figure 12 shows the speed setpoint together with the This route allows has been used to check the actual speed of the vehicle. It can be seen that during operation and accuracy of: the first few seconds it takes a little while to react, this is due to the fact that during this time the vehicle  Speed control is changing gear, so it keeps its speed at zero until it  Steering control does so.  Gear shift It then makes a speed ramp up to 8 km per hour, The speed control should behave as follows. It must which is the target speed for this route. At the end accelerate progressively at the start, maintain a we can see how the vehicle slows down. The speed 229 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia setpoint decreases sharply from 8 km/h to 0 km/h, transport of material along a defined route is because the jump is very small but the vehicle brakes common to many civil engineering works, so its smoothly, thanks to the control of the brake motor. application would be immediate. With this development, the aim is to be able to take advantage of vehicles, so that after a process of automation, they are capable of carrying out this task, reducing costs, maintenance and increasing safety. The INSIA tests reflected in this article show the system's capacity to perform the task for which it has been developed. Apart from the visualisation of the route followed, data from the steering wheel and speed controllers have been included, demonstrating its good response even in a vehicle of these characteristics. The project is now fully developed and ready for its Figure 12: Speed profile planned field tests at the contractor's facilities in Noblejas, Toledo, Spain. The final validation tests Finally, Figure 13 Figure 13: Steering profileshows will be carried out there and the vehicle's capabilities the steering profile. This shows how at the beginning will be extended with a Lidar sensor. of the route it performs a little rougher manoeuvres due to the lack of GPS data, but during the route both Aknowledgement the setpoint and the actual steering wheel angle This research has been partially funded by the move smoothly and very evenly. Spanish Ministerio de Ciencia, Innovación y At the end of the trip there is a significant Universidades (TUNNELAD porject, reference disturbance in the speed control, which is due to RTC-2017- 6382-4), Ministerio de Ciencia e having to calculate the target point extremely close Innovación (CCAD project: reference PID2019- and having the vehicle at a very low speed the GPS 104793RB-C33) and Comunidad de Madrid error increases. Even so, although the setpoint tends (SEGVAUTO 4.0: reference S2018/EMT-4362) to shoot up, the steering wheel remains within a small range. REFERENCES [1] Alonso, F. J., Hernández, J. E. N., & Casado, Ó. G. (2014). Dispositivo universal para el control automatico de la direccion de un vehiculo. DYNA, 89(4), 398–404. [2] Jiménez, F. (2018). Aplicaciones especiales de la conducción autónoma. Revista de Obras Públicas: Organo Profesional de Los Ingenieros de Caminos, Canales y Puertos 3604, 76--82. https://dialnet.unirioja.es/servlet/articulo?codig Figure 13: Steering profile o=6765303 [3] Jiménez, F., Martín, M., Anguera, A., Gómez, 6. CONCLUSIONS O., Sánchez, S., Cruz, A., Valle, A., & Naranjo, This paper has presented the automation, software J. E. (2020). Operación autónoma coordinada de vehículos de obras públicas en entornos architecture and INSIA testing and results of an complejos. ITS España. industrial vehicle for unstructured environments. This vehicle attempts to take a step forward in the [4] Jiménez, F., Naranjo, J. E., González, M., & development of autonomous and driver assistance Gómez, O. (2013). Equipo para controlar systems for industrial environments. It is ahead of automáticamente la dirección de un vehículo. Patente ES2516568. the development of industrial vehicle production companies that are starting to market vehicles with driver assistance systems. The development of this project has focused on the automation of this vehicle with a view to its future application in mines and unstructured environments with highly repetitive work. The need for continuous 230 Alfredo Valle, Alberto Cruz, Víctor San Telesforo, Eugenio Naranjo, Felipe Jiménez YRS 2021 APPLICATION OF AUTOMATION OF PUBLIC ROAD WORKS VEHICLES IN 15. -17. September UNSTRUCTURED ENVIRONMENTS Portoroz, Slovenia [5] Naranjo, J. E., Jimenez, F., Anguita, M., & [6] Pereira de Oliveira, L., Jiménez Alonso, F., Rivera, J. L. (2020). Automation Kit for Dual- Vieira da Silva, M. A., Tostes de Gomes Garcia, Mode Military Unmanned Ground Vehicle for B., & Messias Lopes, D. M. (2020). Analysis of Surveillance Missions. IEEE Intelligent the Influence of Training and Feedback Based Transportation Systems Magazine, 12(4), 125– on Event Data Recorder Information to Improve 137. Safety, Operational and Economic Performance https://doi.org/10.1109/MITS.2018.2880274 of Road Freight Transport in Brazil. In Sustainability (Vol. 12, Issue 19). https://doi.org/10.3390/su12198139 231 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD OBSERVATIONS Kartik Varma Université Gustave Eiffel, Laboratoire SPLOTT Groupe RENAULT, Département Innovation et Recherche François Combes Université Gustave Eiffel, Laboratoire SPLOTT Pierre Eykerman Groupe RENAULT, Département Innovation et Recherche ABSTRACT The aim of this paper is twofold. Firstly, it aims to deepen the understanding of the process of urban deliveries through lessons from field observations of the current scenario, and secondly, to use this knowledge to envision and examine a scenario for urban deliveries involving autonomous vehicles, remotely monitored by a human operator, from both a practical and financial perspective. In the introduction, urban logistics, its importance, and trends are briefly described. Questions of desirability of the use of autonomous vehicles, from the perspective of both customers and logistic suppliers, are inspected. Through field observations areas of crucial human intervention are identified. A scenario accomplishing similar delivery characteristics, but utilizing autonomous vehicles instead is developed. Questions and calculations of practicality and cost along with and recommendations for future research are discussed. Keywords: Autonomous Vehicles, Urban Logistics, Field Observations 1. INTRODUCTION – AECS possibilities of the next generation of vehicles. These factors, taken together, hint at a paradigm VEHICLES AND URBAN shift characterized by, among other changes, LOGISTICS different patterns of human-vehicle interaction (i.e. The commonplace ‘car’ may be at the cusp of the non-necessity of a human driver), stricter change; actors across the automobile industry are emission norms, changed ownership scenarios (i.e. aware that the next generation of vehicles – due to a sharing instead of owning) and a higher ratio of time combination of factors – might make the act of utilized per day of vehicles. driving a ‘leisure’ activity rather than a necessity. This transition to AECS will affect vehicle classes This may disrupt both the automobile and mobility differently†; AECS cars, buses, trucks and Light industry. Billions of dollars have been allocated to Commercial Vehicles (LCV’s) can each be expected the development of Autonomous Vehicles (Varma to be used in a singular way, different than currently & Combes, 2020b). McKinsey and Company refers used, and used differently than each other. This to these disruptive factors as ‘diverse mobility, difference of usage will stem from not only the autonomous driving, electrification, and transition to ACES, but also the context of usage of connectivity’( Disruptive Trends That Will the vehicle (ex. Commute/mobility or freight Transform the Auto Industry | McKinsey, n.d.), transport.) While Infosys refers to them as the ‘electric, autonomous, connected (EAC) vehicle’. Drawing from both academic and non-academic literature, these factors can be summed up under the acronym ‘AECS’ which refer to the Automation, Electrification, Connectivity, and Sharing 232 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS 1.1. Urban Logistics 1.2. Autonomous, Electric, Connected and Shared Vehicles 1.1.1. What is Urban Logistics? While autonomous vehicles have had a rich history According to Laetitia Dablanc, Urban or City (ex. The EUREKA funded PROMETHEUS Logistics “can be defined as any service provision project), due to a combination of factors including, contributing to efficiently managing the movements but not limited to, advances in telecommunications, of goods in cities and providing innovative widespread GPS availability, increasingly reliable responses to customer demands. The objective of and fast Internet connectivity, increased processing city logistics is to make deliveries to urban residents, power at reduced costs, and increased sensitivity and visitors, and businesses possible at the highest desire for sustainability, this trend has re-entered economic, social, and environmental standards. importance. Various large OEM’s have invested in City logistics includes physical operations such as this technology – and start-ups have raised money – order preparation and packaging, transportation both to the tunes of billions of dollars (Varma & and deliveries (including homedeliveries), short-Combes, 2020b). term storage of goods, management of drop- off/pick-up boxes for parcels, return of goods, waste However, it is important to consider that the next management including the management of recycled paradigm shift may be a combination of 4 factors; goods, and the management of empty pallets and the removal of the human component, the packages. City logistics makes use of sophisticated electrification of the drive train, the connection of information and communication technologies to vehicles with each other and other technologies, and increase coordination and efficiency and ensure the changing ownership models – the possibility of adequate enforcement of urban delivery sharing instead of owning - brought about due to operations.” (Dablanc, 2019) various connected technologies. These factors are considered together as they are complementary disruptors; the potential disruption is ‘multiplied’ Trends like increasing urbanization ( 68% of the World Population Projected to Live in Urban Areas when these factors are considered together. In more by 2050, Says UN, 2018), increasing e-commerce, concrete terms, if a vehicle is considered ‘autonomous’ in the strictest sense, the only value negative externalities linked to Urban Logistics (congestion, pollution, etc.), coupled with the addition it brings is the removal of a human driver – inherent importance and problematic nature of however, adding the possibility to be connected to Urban Logistics have generated many studies the internet enables a host of economically lucrative addressing various issues. (Holguín-Veras et al., possibilities (ex. Robo taxis). However, this 2018a) and (Holguín-Veras et al., 2018b) through a understanding is not always reflected in the review of the public-sector initiatives that could be literature. Out of various articles consulted for this used to improve freight activity in metropolitan paper, only 1 refers to all 4 aspects together. areas produce a ranking of suggested initiatives. (Another component not mentioned is the changing (Combes, 2019) presents a structural shape of the vehicles, however it is out of the scope microeconomic model of the choice of warehouse of this article and has not much literature on it) location in urban logistics, which focus on Based on the available literature, it is difficult to operational constraints, their diversity, and their argue that autonomous vehicles may not be both influence on costs to arrive at an understanding of useful or used in logistics. (Bucsky, 2018) inspects current trends and recommendation for land-use the potential of autonomous vehicles across planning. (Quak et al., 2014) discusses the business different freight modes (road, rail and urban aspects that need to be considered when logistics) and concludes that in urban logistics, the implementing or scaling up city logistics solutions connection between the transport of individuals and by using the business model canvas, which is freight transport can offer efficiency gains, though accompanied by data on field demonstrations which regulation and safety can be a bigger challenge, show improvements on operational efficiency, however market push effects of the rise of e- flexibility and environmental impact. (Gatta et al., commerce will help developments in this field. 2019) make an exploratory analysis which aims at (Meldert & Boeck, 2016) study vehicle automation understanding and evaluating the environmental and technology, liability and legislative challenges, economic impacts of a crowd shipping platform in along with the ethics & human factors challenges urban areas through modelling on the city of Rome. applied to the usage and potential consequences of The literature review above aims, through AVs for the logistics industry and conclude adoption mentioning a small number of papers, to show the of AVs holds the promise of innovating the way in diverse and eclectic nature of this subfield of which mobility and transportation logistics are dealt transportation research. with. Studies inspecting the impact on business 233 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS models have also emerged (ex. (Fritschy & Spinler, service which integrates a drone and an autonomous 2019)) indicating interest across the ecosystem. Light Commercial Vehcle (LCV) to deliver small parcels. ( Mercedes-Benz, n.d.) Renault, under the Apart from companies such as Waymo which are aegis of CityPod has been working on a modular working on automation of personnel transport, other autonomous vehicle, with similar loading volumes companies are working specially on delivering as an LCV. These vehicles can be broadly classified freight. For example, Nuro a start-up based in into 3 types (Figure 3); Multiple delivery vehicles Arizona recently raised almost a billion dollars for (Ex. Nuro), Single delivery pods (ex. Amazon their local goods delivery vehicle. ( Nuro — Product, Scout), and non-standard delivery bots (ex. Ford’s n.d.) Mercedes has successfully demonstrated a Digit). Image 1: Nuro, Scout and Digit (L-R) 1.3. Research Questions driver is handed their round for the day in the form of slips of paper. Each of these slips represents an In light of recent trends linked to urbanization, urban individual delivery and contains the address and logistics and AECS vehicles (mentioned above), contact details for the respective client. Also, these through this paper, we examine the question of the slips are presented to the driver as per a computer suitability of AECS vehicles for urban deliveries, specified order (an optimal round). However, at this notably Autonomous Multiple Delivery Vehicles. particular logistic hub, the opening/closing time of Using, as the foundation of our understanding, data the enterprises are not considered by the computer at from field observations, we closely inspect the the time of generating the round. Using this (time of process of urban deliveries to identify the suitability opening hours) and other factors, the drivers of substitution of conventional vehicles by AECS rearrange their round. In the image above, a driver vehicles. We examine with attention the role of the (Abdel) is reorganizing his round. In general, a driver of the vehicle in the delivery process in order driver receives roughly 50 slips per round in this to better envision autonomous deliveries. facility. At the end of this reclassification, the drivers start 2. METHODOLOGY/RESULTS loading their vehicles in the warehouse. While conditional on the number of doors on the vehicles, In this section, first the field observations, and the objects are generally loaded ‘first in last out’. engendered questions – linked to the application of This process takes roughly 60-90 minutes. autonomous vehicles in urban logistics – are Sometimes there are instances when the objects, described. Areas of necessary human intervention according to the computer specified round, do not fit during the delivery process are delineated. Finally, in the vehicle. In these cases, the driver does not load financial aspects of different scenarios, including these objects, and leaves out the corresponding slips when the delivery process is replaced with an from the delivery round. When the vehicle is loaded, autonomous vehicle with the aspects of human the driver recontacts the manager; the reordering interaction handled remotely, are examined. (along with the final number of deliveries (in case 2.1. Field observations some objects are not loaded) is validated by the manager. At the end of this step, the driver is ready The information collected for this paper was to start the delivery process. gathered during various field visits. Most notably 2 days were spent delivering parcels with a freight The delivery process had several peculiarities. company in the outskirts of Paris, France. Firstly, in both the observation cases (with different drivers and different vehicle types), the drivers were In this organization, at 7 am each working day, the intimately familiar with the area of delivery and the drivers report to a centralized office, where each addresses. In the case of the larger vehicle, since driver is allocated their round for the day. The roles most of the deliveries were to business, the driver, of the ‘drivers’ and ‘managers’ are clearly based on his knowledge of opening times of various demarcated; the managers handle the planning of the enterprises did not have to make as many individual delivery (creation of delivery round, distribution calls as the driver of the smaller vehicle. The driver among drivers) in the office. The drivers handle the of the smaller vehicle, as he was dealing primarily execution of the delivery in the field. Then, the 234 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS with B2C deliveries made calls for almost every In this organization, every working day, delivery delivery. rounds were generated by software. Then, these delivery rounds were allocated to drivers. The Thus, the order of activity is as follows. Once the vehicles are loaded based on the round, following driver has mounted the vehicle, s/he will contact the which the driver starts the delivery process. first client, if the first client is not a business. If the The delivery activities are carried out in the client responds, based on whether s/he is ready, the following order. Clients are contacted and if they are driver will drive to the specific client. The driver did available, the driver drives to them to complete this, in almost all instances while actively engaged either the delivery or the return. In almost all in the act of driving the vehicle. Once the driver instances, the driver contacts the clients arrives at the location s/he will locate the specific simultaneously while driving. Once at the location, parcel in the storage and make the physical delivery the driver locates the specific parcel in the storage (in return for a signature). At the end of a delivery, and make the physical delivery (in return for a the driver reenters the vehicle and repeats the cycle signature). of contacting and delivering. At the end of all deliveries, s/he makes his/her way back to the A rough, qualitative estimate of the time spent per warehouse. In these two observations, only 1 activity in the delivery process is presented in Figure delivery (out of approximately 90) was reattempted 1. on the same day1. The drivers did their best to achieve delivery on first attempt. Figure 1: Time spent per activity 2.2. Deliveries using an autonomous 2.2.1. Type of autonomous vehicles suitable vehicle without physical human assistance for urban deliveries This section discusses the possibility of using an According to the SAE, a level 5 autonomous vehicle autonomous vehicle in the process sequence can accomplish ‘the full-time performance by an described above. automated driving system of all aspects of the dynamic driving task under all roadway and environmental conditions that can be managed by a human driver’. In other words, no human 1 This was a special scenario concerning delivery to a supermarket visually identified based on the number of vehicles queueing to located in a shopping mall. Owing to the uniqueness of this enter), the reattempted delivery was abandoned. It should be location, the entry to the logistic bay was often crowded and could noted that the product delivery was not urgent, and that the driver imply larger wait times, leading to over an hour spent on one makes multiple trips per week to this same supermarket, thus, the single delivery, disrupting the rest of the round. In this instance, marginal cost of transferring this delivery to another day was once the driver attempted delivery as per the order of the round, small. These were factors that the driver was aware of and could and once after finishing some orders, when generally the bay is include as parameters while making decisions in real time. less crowded. However, as it was still crowded (this can be 235 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS intervention for driving is necessary(Varma &  For each delivery, a unique ‘password’ Combes, 2020a). (using a smart device ex. Smart phone, To strategically envision the use of an autonomous computer, etc.) can be issued to the end vehicle, the delivery process was examined based on customer, which enables opening a this definition. Attention is placed on the part after predetermined locker on the autonomous the loaded vehicle along with driver is leaving the vehicle. The merchandise located in the warehouse. appropriate locker is then collected. This takes the place of the ‘signature’ in We imagine a scenario where a level 5 autonomous vehicle, equipped with an appropriately filled traditional deliveries. mobile locker is deployed. We base our analysis on  It can accommodate parcels with different an autonomous vehicle equipped with a mobile physical characteristics. locker as it enables the separation of human intervention. Our choice is based on these three Any other vehicle type (under development or test, reasons: to the best of our knowledge) does not fulfill these criteria. Indeed, in a related article (Figliozzi and  It can accomplish multiple deliveries on Jennings, 2020) come to the same conclusion. the same round (i.e. without returning to the warehouse) Iamge 2: Some autonomous mobile lockers under test and development The following information should be noted.  Single delivery bots (ex. Scout, Figure 1), Cargo Delivery bikes, Drones, On Foot Deliveries, Autonomous Vehicles with a human on board are not considered in our analysis. While these might be theoretical responses to urban deliveries (refer to Figure 2), they are exempted for one or more of the following reasons. - A central focus of this paper is the role of the driver in the urban delivery scenario. Our aim is to elucidate the practical and financial aspects of deliveries without a human driver. Thus, any other alternative with a human operator are not considered. Figure 2: Suitability of various vehicles as a function - To make a relevant comparison with the of distance from warehouse and between deliveries current scenario in which a vehicle 2.2.2. Exchanging information with customers leaves a warehouse, generally situated not close to the sites of delivery, and While this vehicle can drive to various customers, makes multiple deliveries in one round, the question on how these customers might be we do not consider single delivery notified about the imminent arrival is raised. The vehicles. fact that the customer is aware and consents to the time and date of delivery or return is important for at least two reasons: 1. To avoid a delivery/return failure; in case the customer is not ready to accept the delivery of or return some merchandise, it must either be returned to the warehouse 236 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS and accommodated on another round in the These margins heavily depend on the coming days or collected another day. This number of parcels delivered per day. is economically costly. 2. To avoid excessive time spent on a It should be noted that in the field observations of delivery/return; having an unprepared conventional deliveries, the driver of the vehicle is customer will increase the time spent per accomplishing the task of calling the customers and delivery/return. As is, from the supply side making a real time decision on whether to drive to (logistic companies), urban logistics is an them parallel while driving. operation with very narrow profit margins. Table 1: Shift of tasks between conventional and autonomous vehicles S.No Task Handled by in Handled by in Autonomous Conventional Delivery Delivery 1. Navigating Driver Vehicle 2. Calling and notifying customer of arrival Driver Call Center Operator 3. Locating merchandise in storage Driver Customer 4. Unloading merchandise Driver Customer 5. Delivering merchandise to end customer Driver Customer (customer collects it him/herself) 6. Getting validation of successful transaction Driver Vehicle (registers (delivery) from customer opening/closing of door) 7. Latent knowledge Driver Call Center Operator? - Creating personal relationships - Looking for parking - GPS overriding - Memorized entry codes - Real time rerouting To handle this part of the delivery process, our The effect of this latent knowledge on delivery scenario assumes the presence of a remote effectiveness parameters, like time spent on average management system supervised by a human, who per delivery, first time delivery success rates, time apart from remote monitoring and guidance, also spent of average for parking, etc. is, very poorly handles the calling and letting customers know tackled in the literature, and to the best of the about the vehicle arrival and confirms they are ready knowledge of the authors, not reflected in any study for the delivery. For the rest of the document, the doing economic cost considerations. However, the term autonomous delivery is used to represent a effect of this knowledge can be extremely delivery undertaken by an autonomous vehicle with significant on some performance parameters. For remote human supervision (no physical human example, according to one of the rare studies which assistance), and the term conventional delivery is explicitly consider this “the difference between two used to represent a delivery undertaken by a non-of our drivers (D22 and D24) with similar round autonomous vehicle with a human driver. Table 1 sizes and parcel volumes shows a considerable summarizes the shift of tasks from the scenario of a variation in effective- ness, with D22 driving 44% standard delivery to an autonomous delivery less distance, spending 35% less time per parcel, scenario. 29% less driving time per parcel, and 39% less parking time per parcel. The variation in 2.2.3. The role of Latent Knowledge effectiveness of our drivers relates to better route planning, exploitation of accumulated knowledge of Often, logistics companies allocate their drivers to specific ‘delivery rounds’ within the same the round, personal relation- ships with other stakeholders, the amount of time spent at the geographic area or patch over a long period of time. curbside and the influence of walking. These By maintaining them on the same rounds, carriers statistics show that more effective drivers achieve attempt to build up the driver’s familiarity in a round higher rate of delivery of parcels per minute while and surrounding area, helping them learn the most efficient routes, build personal relationships with spending less time driving and parking in the van”.(Bates et al., 2018) The relevant matter of the customers and maintain knowledge of a round. transfer of these significant efficiency gains in the Owing to repeatedly deliveries in the same area, the context of autonomous deliveries is considered in driver develops an extremely useful, practical, the discussion section. understanding of the area and its inhabitants. A non- exhaustive list of these understandings and their applications are reflected in Table 2. 237 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS Table 2: Latent knowledge and its applications S.No Skill/Knowledge Application 1. Perso nal relationships with - Alternate delivery possibilities (ex. If customer is not available, driver various inhabitants delivers to neighbour, based on previous agreements) - Accommodating non prepared return on other rounds (Especially true for business deliveries, If a return form a customer is not prepared, s/he can call the driver and inform them, making the round more efficient) - Concierge 2. Kno wledge of Parking spaces - Reduction of time spent to look for a parking space - Knowledge of parking time restrictions - Alternate parking spots 3. Kno wledge of traffic Ex. Higher traffic in a specific repeated delivery address at a particular conditions and trends time (commercial center) - Reordering delivery order to achieve faster overall delivery 4. Kno wledge of geographical Ex. Access codes, GPS Map Failures : Dead Ends quirks - Driver saves a lot of time by already knowing the access code, or the requirement of it. - Driver aware of GPS failures in certain specific scenarios, ex. dead ends, and avoids them. 2.3. Cost competitiveness and financial and an autonomous vehicle remotely operated by a considerations human operator denoted by 𝐴𝑉. We then incite the following concepts: To have any incentive to switch to autonomous deliveries, we assume that some value must be The vehicle capital cost is denoted by 𝐶𝑘(€/day), the added in the delivery process. This added value operating costs by 𝐶𝑜(€/km) and the driver costs by could be for the end customer, or for the 𝐶𝑤(€/h) organization fulfilling the delivery. The approach distance is denoted by 𝑑𝑎 and the For the customer, this added value could take the distance between deliveries is denoted by 𝑑𝑧. The form of Reduced delivery prices and greater distance per round then becomes 𝐷 = 2. 𝑑𝑎 + 𝑛. 𝑑𝑧 flexibility regarding delivery times. For the logistics The number of rounds per day is denoted by 𝑛 provider, this value could arise from reduced 𝑟, the number of deliveries per round denoted by 𝑛. We delivery costs arising from reduced first time failure thus arrive at the number of deliveries per day, rates and reduced delivery times (as deliveries can denoted by 𝑁 = 𝑛 be scheduled), or from more deliveries per round, 𝑟 × 𝑛 higher energy efficiency of vehicles, owing to The time spent for delivery is denoted by ℎ. We flexible hours (autonomous vehicles can deliver till denote the delivery time window by 𝐻. Some the limits of their autonomy). It could also arise from logistic companies make multiple deliveries rounds better delivery management since the vehicles are per day, thus 𝐻 takes the form of 24 hours. In some connected and can provide real time fulfilment data other cases the logistic company might only make a along with better scheduling and higher customer single round. The round duration3 thus take the form satisfaction (if the customers can, for example, 𝑑 𝑑 of 𝐻 𝑎 𝑧 𝑟 = 2 + 𝑛 (ℎ + ). By extension, the number schedule the delivery more flexibly). 𝑣𝑎 𝑣𝑧 𝐻 of rounds 𝑛𝑟 = . In our analysis, we develop an economic model 𝐻𝑟 focused on the delivery costs. We assume the The user cost per delivery is denoted by 𝑢. it stands demand is fixed. The cost2 is an extremely important for the fact that a) the customer may have to decisional parameter for players in the domain of participate to the realisation of the delivery and b) logistics, ex. for fleet managers the TCO is the most the level of service does not fit their preferences important consideration.(Nesbitt & Sperling, 2001). perfectly. Without loss of generality 𝑢𝑠 = 0. We first make the distinction between a The following operating constraints are assumed. conventional delivery (i.e. a non-autonomous vehicle operated by a human) which is denoted by 𝑆 2 Greater flexibility regarding delivery times for customers, due to the advent of autonomous vehicles remain nebulous and better delivery management and higher customer satisfaction thus difficult to quantify. that may be offered by autonomous vehicles, for example, are 3 This definition of the round duration does not include the also relevant parameters for the logistics provider. However, at charging or the loading and unloading time. The complete this stage of development of autonomous vehicles the magnitude definition thus becomes 𝐻𝑟 = 2 𝑑𝑎 + 𝑛 (ℎ + 𝑑𝑧) + and direction of the changes associated with these parameters 𝑣𝑎 𝑣𝑧 (𝑐ℎ𝑎𝑟𝑔𝑖𝑛𝑔 𝑡𝑖𝑚𝑒) + (𝑙𝑜𝑎𝑑𝑖𝑛𝑔 𝑡𝑖𝑚𝑒) + (𝑢𝑛𝑙𝑜𝑎𝑑𝑖𝑛𝑔 𝑡𝑖𝑚𝑒). 238 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS  In the case of a conventional delivery 𝑆, the 2.3.2. Case 2: Autonomous Delivery delivery time window and round duration The cost of autonomous vehicles is difficult to (delivery time window duration 𝐻(ℎ)) precise. Sensors and technology required would  In the case of autonomous vehicles, 𝐴𝑉, the probably imply a higher purchase cost for operating constraints take the form of the autonomous vehicles. Though, this cost might be max number of shipments carried by the offset by the fact that it can operate continuously vehicle when it leaves from the warehouse. (apart from refueling/charging time). However, This is denoted by 𝑛𝐴𝑉 these savings cost might also be offset by the  We arrive at the Total Cost of Delivery mechanical wear and tear of operating under these (TCD) which is denoted as 𝑇𝐶𝐷 = 𝐶 conditions. This might manifest itself as a shorter 𝑘 + 𝐶 life of the vehicle and a reduced, if any, resale value. 𝑤. ℎ + 𝑛𝑟. 𝐷. 𝐶𝑜  In essence, there are various factors which may And finally, we denote the Cost per contribute to either an increased or decreased total 𝑇𝐶𝐷 Delivery 𝐶 = cost of ownership of an autonomous vehicle. 𝑁 Literature on these is currently sparse, however there 2.3.1. Case 1: Conventional Delivery appears to be a general consensus that the purchase cost would be higher, at least initially, due to the cost In the case of conventional deliveries 𝑆 the TCD of technology.(Bösch et al., 2018) Some of these takes the form of seemingly contradictory forces are delineated in 𝑇𝐶𝐷𝑆 = 𝐶𝑆 𝑆 𝑆 table 3. For our analysis, however, instead of 𝑘 + 𝐶𝑤. ℎ𝑠 + 𝐷. 𝐶𝑜 speculating the direction and magnitude of these In the European Union, a Light Commercial Vehicle forces, we consider the 𝑇𝐶𝐷𝐴𝑉 and 𝑁𝐴𝑉 to vary used in an urban context, has an effective life span from half to three times that of conventional of 10.7 years. Total cost of ownership (TCO) of a vehicles. Thus, the total cost of delivery for light commercial vehicle used in an urban context is autonomous vehicles is represented as; calculated over 10.7 years. (Lebeau et al., 2019). We 𝐴𝑉 𝐴𝑉 𝐴𝑉 𝑆 found the work of the MOBI laboratory in Belgium 𝑇𝐶𝐷𝐴𝑉 = 𝐶𝑘 + 𝐶𝑤 . ℎ𝐴𝑉 + 𝑛𝑟 𝐷. 𝐶𝑜 + 𝑢𝐴𝑉 to be most adapted for the extrapolation of figures In our autonomous delivery scenario, the deliveries for this article as they use a consistent, are remotely supervised, and calls made by a human. comprehensive methodology over various papers We assumer a similar cost structure as assumed in applied to vans in urban logistics. Notably (Lebeau our calculations regarding conventional deliveries. et al., 2019) considers explicitly the case of electrical Thus, 𝐶𝐴𝑉. ℎ𝐴𝑉 = 𝐶𝑆 . ℎ𝑆 = 80€/𝑑𝑎𝑦. It is possible vans in urban deliveries. Electric vans for urban 𝑤 𝑤 deliveries are currently the ‘conventional state of the that the level of service with autonomous deliveries art’, further, as described above, autonomous may not be the same as conventional deliveries, with 𝑢𝐴𝑉 ≠ 0 vehicles may probably be electric. Thus, figures . While the implications of this (and the from this publication are used. difficultly in quantifying it) are discussed in the next section, for our calculations, we assume that 𝑢𝐴𝑉 = They note that the average TCO (in eurocents/km) 𝑢𝑆 = 0. across 18 vehicles is 36.82 €c/km. Further, the average kms travelled per year are found to be 𝑇𝐶𝐷𝐴𝑉 𝐶𝐴𝑉(𝑆𝑖𝑛𝑔𝑙𝑒 𝑉𝑒ℎ𝑖𝑐𝑙𝑒) = 16000, with 260 operating days. Thus, the average 𝑁𝐴𝑉 kms travelled per day are roughly 61.54km/day. There is sparse literature on TCO of autonomous Thus, we note 𝐶𝑆 𝑆 𝑘 + 𝐷. 𝐶𝑜 = 22.66€/𝑑𝑎𝑦. Through vehicles, thus it makes it difficult to have numerical rough salarial estimates obtained during our field values, and instead we address the ratios. observations, we note that the drivers earn roughly 9-12€/hour. We also note that the working hours can Let’s consider the diagram of the division of time vary between 6-10 hours/day. We assume the hourly (Figure 1.). We note that the driver is engaged in wage as 10€/hour and the hours worked per day as calling customers parallelly while driving for an 8. Thus, 𝐶𝑆 average of 180 seconds out of the 691 seconds or for 𝑤. ℎ𝑠 = 80€/𝑑𝑎𝑦 an average of 26% of the time. Assuming contacting 𝑇𝐶𝐷𝑆 = 𝐶𝑆 𝑆 𝑆 𝑘 + 𝐶𝑤. ℎ𝑠 + 𝐷. 𝐶𝑜 = 102.66 €/𝑑𝑎𝑦 the customer takes a similar amount of time in the scenario of an autonomous delivery, the question of We further assume, based on subjective estimates how the human operator uses the rest of (74%) of the from field observations, 𝑛𝑆as 50 and 𝑛𝑆𝑟 as 1. Thus, 102.66 time becomes. We suppose that the economic 𝑁𝑆 = 50. We arrive at 𝐶𝑆 = = 2.0532 €/ 50 incentive lies in that the operator uses this time to 𝑑𝑒𝑙𝑖𝑣𝑒𝑟𝑦 handle other vehicles. Taking 𝑃 to be the number of vehicles handled simultaneously, we arrive at the following cost per autonomous delivery 239 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS 𝐶𝐴𝑉(𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) Thus, 𝑃. (𝐶𝐴𝑉 + 𝑛𝐴𝑉𝐷. 𝐶𝐴𝑉) + 𝐶𝐴𝑉. ℎ𝐴𝑉 = 𝑘 𝑟 𝑜 𝑤 𝑃. 𝑁𝐴𝑉 𝐶𝐴𝑉(𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) 𝑃. (𝐶𝐴𝑉 + 𝑛𝐴𝑉𝐷. 𝐶𝐴𝑉) + 𝐶𝐴. ℎ𝐴𝑉 = 𝑘 𝑟 𝑜 𝑤 2.3.3. Comparison 𝑃. 𝑁𝐴𝑉 For delivery by autonomous vehicles to be cost 𝑃. 𝑎. (𝐶𝑆 + 𝐷. 𝐶𝑆) + 𝐶𝑆 . ℎ𝑆 𝐶𝐴𝑉(𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) = 𝑘 𝑜 𝑤 competitive, 𝑃. 𝑏. 𝑁𝑆 𝐶𝐴𝑉(𝑠𝑖𝑛𝑔𝑙𝑒 𝑜𝑟 𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) ≤ 𝐶𝑆 𝑃. 𝑎. (22.66) + 80 𝐶𝐴𝑉(𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) = 𝑃. 𝑏. 50 We take the following ratios We take that 𝑃 varies from 1 to 4 (I.e. 1 human  (𝐶𝐴𝑉 𝐴𝑉 𝑆 𝑆 𝑆 operator can remotely handle 1-4 vehicles), while 𝑎 𝑘 + 𝑛𝑟 𝐷. 𝐶𝑜 ) = 𝑎. (𝐶𝑘 + 𝐷. 𝐶𝑜 ) varies from 0.5 to 2 (I.e. (𝐶𝐴𝑉 𝐴𝑉 𝑆 𝑘 + 𝑛𝑟 𝐷. 𝐶𝑜 ) is 0.5 –  𝑁𝐴𝑉(𝑠𝑖𝑛𝑔𝑙𝑒 𝑜𝑟 𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) = 𝑏. 𝑁𝑆 2 times (𝐶𝑆 𝑆 𝑘 + 𝐷. 𝐶𝑜 )) and finally 𝑏 varies from 0.25 to 1 (I.e. 𝑁𝐴𝑉(𝑠𝑖𝑛𝑔𝑙𝑒 𝑜𝑟 𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 𝑣𝑒ℎ𝑖𝑐𝑙𝑒𝑠) is 0.25 – 1 times 𝑁𝑆). Figure 3: The ratio of Cost of Autonomous Delivery to the Cost of Conventional Delivery (with P=1 or P=4) In the graph above we consider that the human (ex. (Beziat, 2017) states that this number is close to operator remotely handles either 1 vehicle or 4 30). However, based on various sources, it seems vehicles. We choose the upper limit as 4 as in our that an autonomous multiple locker would make analysis of the time spent per activity of the delivery fewer deliveries, thus we assume that it makes process, the driver is involved in making phone calls between 12 to 50 deliveries. We not that there are a 26% of the time. If a human operator is concerned few cases where autonomous delivery is cost with exactly this part of the delivery process then, competitive and might even be cheaper. This for 26% of the time he is involved in calling, but free generally occurs since the human operator handles the other 74% of the time. Thus, there is economic multiple autonomous vehicles. In other words, incentive for him to handle more vehicles parallelly. conditional on the fact that autonomous vehicles Thus, if it takes on average 26% of the time, on cost the same (unlikely), and can deliver similar average he can handle (close to) 4 vehicles in number of vehicles in the same round (also parallel. unlikely), only the fact that a human handles multiple autonomous vehicles will lead to an We then vary the number of deliveries that can be economic incentive for autonomous deliveries. This accomplished in the same round by an autonomous is assuming no loss of efficiency due to the vehicle. The average number of deliveries in our experience of the driver (which is also unlikely). observations were close to 50. Though there exist other estimates for the average number of deliveries 240 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS 3. DISCUSSION 3.1. Financial calculations In the results section, we present a mathematical Firstly, we discuss some caveats and trends model which compares the cost of delivery in an associated with the parameters used in the model. autonomous scenario to the cost of delivery in the Autonomous vehicle technology is changing at a conventional scenario. This economic model aims to rapid pace. This has implications for the vehicle find a combination of parameters at which, capital cost 𝐶𝑘, the operating costs 𝐶𝑜and theoretically, fleet managers/logisticians are 𝑁𝐴𝑉parameters used in our calculations . Notably, indifferent between using autonomous or both parameters are, in turn, dependent on various conventional vehicles for urban deliveries. other parameters. These other parameters can be There are various caveats to this form of modelling, analytically reasoned to evolve, financially however, especially if this issue is addressed speaking, in opposite directions. A non-exhaustive practically rather than theoretically. These are list of these parameters and their evolutions, and the discussed below. reason for these evolutions, is presented in Table 3 below. Table 3: Evolution of parameters linked to Cost per Autonomous Delivery Parameter in 𝑇𝐶𝐷𝐴𝑉 Expected to Due to Vehicle capital cost 𝐶𝑘 1. Increase 1. Higher purchase cost due to 2. Increase expensive sensors (ex. Lidar) 3. Decrease 2. Higher maintenance cost due to TKD increase (more hours of operation) 3. Lower maintenance cost due to fewer moving parts due to electric Number of Deliveries (𝑁𝐴𝑉) 1. Increase 1. Hours of operation 2. Decrease 2. lower operating speed 3. Decrease 3. smaller volume (ex. Nuro) 4. Increase/Decrease 4. Potentially longer (if accomplished by humans) or shorter (if accomplished by sophisticated machines) loading time Thus, it becomes hard to predict the overall direction engendered by a financial incentive, then, a part of of movement about the Total cost of Ownership of this incentive could be offered to the customer in Autonomous Vehicles, and more so in the context of compensation for the reduced level of service. urban logistics. However, how much does the customer value the service provided in the last few meters of the last (Figliozzi & Jennings, 2020) construct a similar mile? In other words, if the customer could choose financial model considering various financially between a conventional delivery (which covers the relevant parameters. Our work, while modelling entire last mile), and delivery by an AML (which similarly, is idiosyncratic in proposing a solution covers all but the last few meters), what would be that considers real world constraints and fits in the his/her willingness to pay for the former? If a existing urban logistic ecosystem. monetary value, albeit subjective (as the last few 3.2. Level of service and customer meters will have different importance to individuals, willingness to pay for example, with limited mobility compared to agile individuals) can be assigned to it, what if this In a scenario where deliveries are accomplished by number exceeds the total monetary value created by an autonomous mobile locker (AML), it is the adoption of autonomous deliveries? reasonable to consider that there might be a loss of While simplistic, the argument serves to stress the service from the customers perspective. While intricate nature of urban logistics, highlighting that traditionally the customer could expect to be to effectively contribute to an understanding of the delivered at his/her doorstep, with AML’s, s/he possibility of adoption of autonomous vehicles for would be responsible for the last few meters of the urban logistics, examining adoption scenarios on a last mile. This is represented as 𝑢𝐴𝑉 or 𝑢𝑆 in our case to case basis is probably more fruitful; what delivery cost model. may work for transport of hotel linen from the This purported reduction in the level of service laundry service may not work for delivery of parcels raises pertinent economic questions. If, on the to dispersed apartment blocks. This general research supply side, the switch to autonomous vehicles is 241 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS recommendation will also be reflected in our further commerce may render centrally situated warehouses research. cost competitive, leading to reduced delivery times and costs. The general direction of these trends is 3.3. Latent knowledge of the driver comparatively convergent and point in the direction Another relevant aspect, which has been briefly of reduced delivery costs for conventional vehicles. discussed above, are the efficiency gains linked to Another aspect consider is the development of an experienced driver. From a psychological artificial intelligence. The development, perspective, physically interacting with the deployment, and utility of autonomous vehicles is environment, the driver passively gathers and stores symbiotic with that of AI, especially concerned with various cues like social relationships (a non- (but not limited to) the software. For example, exhaustive list has been presented in table 2). These advancements in AI could, theoretically, eliminate understandings are unique to physical interaction; the need of a LIDAR through advances in image the driver would use, for example his/her spatial processing software. Advancements in AI could memory to keep a mental repertoire of parking perhaps also address the efficiency gains linked to spaces, which a human operator remotely the latent knowledge of a driver. The development controlling an autonomous vehicle would be unable of AI, and its application to AV’s is well beyond the to. Further, various studies (ex. (Kolb & Kolb, scope of this article. However, we acknowledge both 2005)) have found that humans encode cues in both the critical place it occupies, and the difficulty in long term and short-term memory when various, prediction of its developmental milestones, linked to rather than a singular, sensory modality is engaged its application to autonomous vehicles, and in a task. Thus, human remote operators, may be subsequently their use in urban logistics. generally overall less efficient at deliveries. As mentioned flowing figure 3, it appears unlikely 4. CONCLUSION that the economic incentive to switch to In this paper, we begin by briefly discussing urban autonomous modes of deliveries would be due to logistics, its contemporary issues and the evolution the removal of the human component. Rather, it of autonomous vehicles. Considering the case of appears that in extremely specific cases, for urban deliveries, we present data from field example for dispersed e-commerce orders with a observations. Based on which, we develop a truly short lead time from a warehouse close to the scenario of urban deliveries involving a remotely final delivery locale (ex. a supermarket), a monitored autonomous mobile locker. We mention premium delivery fee could be charged to realize a associated challenges and present financial delivery with a single delivery bot. We consider calculations based on an economic model. This model and the methodology addressing the subject this as a theme for future research. of cost competitiveness between autonomous and Evolution of Conventional vehicles (and urban conventional deliveries, and its development based deliveries with them) on real world observations are the primary While we have painted a generous picture of the contribution of this paper. We further examine the potential impact of autonomous vehicles and results, their implications and limitations the deliveries on urban logistics, this analysis would be discussion section, concluding that there are various, incomplete without considering the alternative: uncertain factors shaping the adoption and conventional vehicles. development of autonomous vehicles, and highlight the prudence in pondering the issue of adoption on a How are conventional vehicles evolving? And specific, case to case basis, rather than general where will they be in the next few years? Higher fuel calculations, which we also present as a general efficiency can be expected. Parallelly, battery recommendation to the field. technology has been improving exponentially. While studies have already shown (ex.(Figenbaum, REFERENCES 2018)), that battery electric vehicles are already suitable for a number of deployment scenarios in [1] 68% of the world population projected to live in urban logistics, increased battery efficiency may urban areas by 2050, says UN. (2018, May 16). render them extremely cost competitive, maybe UN DESA | United Nations Department of Economic and Social Affairs. even to the point of surpassing convention fuel https://www.un.org/development/desa/en/news/ vehicles. Simultaneously, developments in population/2018-revision-of-world- computer science, communication, and information urbanization-prospects.html technology will continue to yield information flows (ex. inventory management, optimizing order of deliveries, scheduling orders) more efficient, further reducing the cost of deliveries. Continued rise in e- 242 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS [2] Bates, O., Friday, A., Allen, J., Cherrett, T., [12] Gatta, V., Marcucci, E., Nigro, M., Patella, S. McLeod, F., Bektas, T., Nguyen, T., Piecyk, M., M., & Serafini, S. (2019). Public Transport- Piotrowska, M., Wise, S., & Davies, N. (2018, Based Crowdshipping for Sustainable City April 20). Transforming last-mile logistics: Logistics: Assessing Economic and Opportunities for more sustainable deliveries Environmental Impacts. Sustainability, 11(1), [Proceedings paper]. Conference on Human 145. https://doi.org/10.3390/su11010145 Factors in Computing Systems - Proceedings; [13] Holguín-Veras, J., Amaya Leal, J., Sánchez- ACM. http://doi.org/10.1145/3173574.3174100 Diaz, I., Browne, M., & Wojtowicz, J. (2018). [3] Beziat, Adrien. (2017). Approche des liens entre State of the art and practice of urban freight transport de marchandises en ville, formes management: Part I: Infrastructure, vehicle- urbaines et congestion : Le cas de l'Ile-de- related, and traffic operations. Transportation France. Research Part A: Policy and Practice. https://doi.org/10.1016/j.tra.2018.10.037 [4] Bösch, P. M., Becker, F., Becker, H., & Axhausen, K. W. (2018). Cost-based analysis of [14] Holguín-Veras, J., Amaya Leal, J., Sanchez- autonomous mobility services. Transport Diaz, I., Browne, M., & Wojtowicz, J. (2018). Policy, 64, 76–91. State of the art and practice of urban freight https://doi.org/10.1016/j.tranpol.2017.09.005 management Part II: Financial approaches, logistics, and demand management. [5] Bucsky, P. (2018). Autonomous vehicles and Transportation Research Part A: Policy and freight traffic: Towards better efficiency of Practice. road, rail or urban logistics? Urban https://doi.org/10.1016/j.tra.2018.10.036 Development Issues, 58(1), 41–52. https://doi.org/10.2478/udi-2018-0022 [15] Kolb, A. Y., & Kolb, D. A. (2005). Learning Styles and Learning Spaces: Enhancing [6] Combes, F. (2019). Equilibrium and Optimal Experiential Learning in Higher Education. Location of Warehouses in Urban Areas: A Academy of Management Learning & Theoretical Analysis with Implications for Education, 4(2), 193–212. Urban Logistics. Transportation Research https://doi.org/10.5465/amle.2005.17268566 Record, 0361198119838859. https://doi.org/10.1177/0361198119838859 [16] Lebeau, P., Macharis, C., & Van Mierlo, J. (2019). How to Improve the Total Cost of [7] Dablanc, L. (2019). City Logistics. In Ownership of Electric Vehicles: An Analysis of International Encyclopedia of Geography (pp. the Light Commercial Vehicle Segment. World 1–8). American Cancer Society. Electric Vehicle Journal, 10(4), 90. https://doi.org/10.1002/9781118786352.wbieg0 https://doi.org/10.3390/wevj10040090 137.pub2 [17] Meldert, B. V., & Boeck, L. D. (2016). [8] Disruptive trends that will transform the auto Introducing autonomous vehicles in logistics: A industry | McKinsey. (n.d.). Retrieved March 5, review from a broad perspective (No. 543558; 2019, from Working Papers Department of Decision https://www.mckinsey.com/industries/automoti Sciences and Information Management). KU ve-and-assembly/our-insights/disruptive- Leuven, Faculty of Economics and Business, trends-that-will-transform-the-auto-industry Department of Decision Sciences and [9] Figenbaum, E. (2018). Can battery electric light Information Management. commercial vehicles work for craftsmen and https://ideas.repec.org/p/ete/kbiper/543558.htm service enterprises? Energy Policy, 120, 58–72. l https://doi.org/10.1016/j.enpol.2018.04.076 [18] Mercedes-Benz: Vans & Drones. (n.d.). [10] Figliozzi, M., & Jennings, D. (2020). Retrieved February 12, 2019, from Autonomous delivery robots and their potential /content/com/en/mercedes- impacts on urban freight energy consumption benz/vehicles/transporter/vans-drones-in- and emissions. Transportation Research zurich.html Procedia, 46, 21–28. [19] Nesbitt, K., & Sperling, D. (2001). Fleet https://doi.org/10.1016/j.trpro.2020.03.159 purchase behavior: Decision processes and [11] Fritschy, C., & Spinler, S. (2019). The impact of implications for new vehicle technologies and autonomous trucks on business models in the fuels. Transportation Research Part C: automotive and logistics industry–a Delphi- Emerging Technologies, 9(5), 297–318. based scenario study. Technological https://doi.org/10.1016/S0968-090X(00)00035- Forecasting and Social Change, 148, 119736. 8 https://doi.org/10.1016/j.techfore.2019.119736 [20] Nuro—Product. (n.d.). Nuro. Retrieved February 12, 2019, from https://nuro.ai/product/ 243 Kartik Varma, François Combes, Pierre Eykerman YRS 2021 SUITABILITY OF HUMAN MONITORED AUTONOMOUS LOCKERS FOR URBAN 15. -17. September DELIVERIES: A COMPARISON WITH THE CURRENT SCENARIO THROUGH FIELD Portoroz, Slovenia OBSERVATIONS [21] Quak, H., Balm, S., & Posthumus, B. (2014). [22] Varma, K., & Combes, F. (2020a, March 12). Evaluation of City Logistics Solutions with LCV’s in Urban Logistics: Are AECS vehicles Business Model Analysis. Procedia - Social and conceivable? Behavioral Sciences, 125, 111–124. [23] Varma, K., & Combes, F. (2020b, December 7). https://doi.org/10.1016/j.sbspro.2014.01.1460 What’s the Buzz About? Analyzing acquisitions by automotive companies. Gerpisa colloquium, Paris (2020), Paris. http://gerpisa.org/node/6002 244 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Cyril Veve Nicolas Chiabaut Ecole Nationale des Travaux Publics de l’Etat 3 Rue Maurice Audin, 69120 Vaulx-en-Velin, France cyril.veve@entpe.fr; nicolas.chiabaut@entpe.fr ABSTRACT Shared mobility services are announced as a game-changer in transportation and a promising solution to reduce congestion and improve the performance of urban mobility. They could prefigure the arrival of autonomous vehicles. Modeling of these new services is a real challenge, especially because existing approaches are mainly an adaptation of methods devoted to classic transportation services. Consequently, this paper introduces a new data-driven optimization method fully devoted to shared mobility service. First, the proposed approach decomposes the recurrent demand based on its spatio-temporal features to overcome the drawbacks of the existing methods. Notably, it makes it possible to consider larger instances and to build robust solutions. Thus, recurrent demand patterns are identified to capture the potential demand of shared mobility services using a tailored clustering process. Second, a variant of Dial-a-Ride Problem is implemented to design robust lines to serve this demand. Such a hybrid method makes it possible to define relatively massive transport lines while maintaining spatial and temporal proximity to users’ real demand. The method is then tested with an open-source dataset released by the New York City Taxi and Limousine Commission. Keywords: Clustering, Mobility pattern, Dial-a-Ride Problem, Similarity, Ride-sharing 1. INTRODUCTION These approaches are well known in the literature for many years. However such methods involve long Shared mobility services are announced as a game- term demand changes. Indeed, the deployment of changer in transportation and a promising solution such lines affects the socio-economic development to reduce congestion and improve the performance around the lines stops, and so affects the demand of of urban mobility. Moreover, it appears that recent mobility. Thus, unlike dynamic methods, studies on shared mobility, particularly regarding conventional methods are not adapted to an the real-time satisfaction of the demand, could instantaneous mobility demand. These methods aim prefigure the arrival of autonomous vehicles. Shared to respond to a demand for global mobility; the lines mobility consists in the shared use of a vehicle (car, designed follow mobility corridors with a high motorcycle, scooter, bicycle, or other travel modes). concentration of departure and arrival points. Modeling of these new services is a real challenge, Generally, the demand is estimated using online especially because existing approaches are mainly questionnaires, surveys, or historical moving data. adaptation of methods devoted to classic However, these methods have some limits; the transportation services. These methods can be calculated flows do not take into account each trip's classified into two main categories: the conventional specificities but only a rough estimation of the methods and the dynamic methods. movements of a large number of users. This spatial The conventional methods are used to design rapid and temporal aggregation has led to the design of transportation lines such as subway, streetcar, or bus significant lines for which the stops are located lines [1,2,3]. These approaches can be qualified as relatively far from the real desired departure and long term methods because the transportation supply arrival points for users. It has been shown in is defined according to both urban planning purpose numerous studies that this problem of the last mile and transportation demand. Consequently, the goal is one of the major brakes which prevents users of is motivated by serving an existing demand but also personal vehicles from deporting to shared modes of to modify the travel behavior at a long time scale. transport. It is to overcome this problem of the last 245 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia mile that for the past ten years, taking advantage of This paper presents a method for estimating and the emergence of smartphones and underlying servicing the potential demand for shared mobility. geolocation technologies, research has turned To do this, a similarity function is defined to towards a new approach so-called dynamic. evaluate the similarities between two different trips. Once this preprocessing has been carried out, a Contrary to conventional approaches, dynamic clustering method is used to obtain groups of trips methods aim at adapting the service supply to the with similar spatio-temporal characteristics. This real-time demand characteristics. One of the main method's variant is then used to detect groups of benefits is that such approaches provide users with similar and recurrent trips over time (meta-clusters). short-term access to a travel mode on an as-needed Finally, a vehicle route optimization method is used basis. These transport services may take different to serve our displacement clusters' departure and forms: station-based roundtrip services, station- arrival points. based one-way services, free-floating services, etc. Similarly, many economic models exist to meet The rest of the paper is organized as follows. Section diverse user needs: public or private, membership- 1 presents the data set and introduces the based, peer-to-peer (P2P), for-hire, or public transit methodology used to estimate demand patterns. system. Moreover, sharing can include either Then, Section 2 is dedicated to the design of sequential sharing (i.e., different users sharing the customized transport lines to satisfy the demand. same vehicle one after the other), or simultaneous Section 3 focuses on the analysis of the results of the sharing (i.e., sharing the same vehicle with multiple demand estimation and the planning of new lines. users for the same trip). Simultaneous sharing is a The last part is devoted to a final discussion. particular challenge that many services try to tackle: transportation network companies (TNCs) now offer 2. ESTIMATION OF THE DEMAND ride-sourcing services (including shared taxi, This section's main objective is to show how the shuttle, etc.); ridesharing (including carpooling, demand can be decomposed into spatio-temporal vanpooling, etc.) is becoming more and more areas containing a significant number of similar popular. The main objective of real-time methods trips. The study focus on the demand of shared (or highly dynamic) is to match a maximum number mobility in Midtown and Upper East Side. The of requests while minimizing objective functions, objective is to present the method used to obtain such as the users' waiting time or the total travel clusters of similar and recurrent trips over time time. [4, 5] provide a list of objective functions and (meta-clusters). These clusters will be used to define matching policies well known in the literature. The the instances of the optimization problem presented dynamic method is well adapted to large fleets of in Section 3. The methodology is based on three vehicles. The dynamic matching between users and steps: (i) definition of a similarity function to vehicles is efficient when the number of vehicles is estimate the likeness between two trips; (ii) significant. It allows reducing the waiting time and implementation of a clustering method to create detours to pick up or drop off a user. The dynamic clusters of similar trips; (iii) development of a approach is considered as a bridge technology that method to detect recurrent clusters over time. will be replaced by autonomous vehicles when the However, it is essential to underline the fact that the technology will be mature [6,7,8]. Despite their demand is analyzed from a transportation point of many advantages, the dynamic methods also have view even if many other aspects could be taken into certain limitations. The real-time matching between account: economic, social or behaviorial. Our travelers and vehicles is complex, and it can not be method determines an upper bound of the potential performed on large instances. The number of of shared mobility. passengers served by a vehicle is often low. That is why the dynamic approach is not adapted to design We use an open-source dataset released by the New massive customized lines of transport. York City Taxi and Limousine Commission (data source: https://www1.nyc.gov/site/tlc/index.page). This paper aims to propose a new hybrid approach Although these data are not fully representative of that allows the design of massive and robust lines of human mobility since they only correspond to taxi transport adapted to the daily demand. The main trips, such a dataset provides an attractive proxy for interest that is driven by the method, is to detect a studying the individuals' routes within a city. The large number of similar and recurrent trips to study focuses on morning peak hours from 8h to 11h estimate the potential demand of shared mobility, of June 2011. The area studied is a well known high- then to design transport lines allowing pickups and density area in terms of mobility in New-York City: deposits as close as possible to the real demand of Midtown and Upper East Side [9,10]. For each trip the users. The approach is defined as hybrid because i, the dataset gives access to the following it take into account the regularity of trips over time information: departure time 𝑡𝑃𝑈 and location 𝑝𝑃𝑈 = (as in conventional methods); however, the 𝑖 𝑖 (𝑥𝑃𝑈, 𝑦𝑃𝑈) distances to be covered and the delays observed for 𝑖 𝑖 of the pick-up of the passenger(s); arrival users remain relatively short. 246 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia time 𝑡𝐷𝑂 𝐷𝑂 𝐷𝑂 𝐷𝑂 𝑙 𝑙 𝑖 and location 𝑝 = (𝑥 , 𝑦 ) of the drop- are penalized. Thus, penalties 𝜃 and 𝜃 are added 𝑖 𝑖 𝑖 𝑥 𝑡 off. when, respectively, the distances between origin (or destination) locations and departure (or arrival) First, it has been shown that the similarity function is used to quantify the likeness between two trips. times of trips 𝑖 and 𝑗 exceed, respectively, specific 𝑙 𝑙 Then the method used to detect groups of similar thresholds, 𝛿𝑥 and 𝛿𝑡 : trips in different spatio-temporal areas is presented. 𝑙 𝑙 𝑙 𝜃𝑙 = 𝑒𝑑(𝑝𝑖,𝑝𝑗)−𝛿𝑥 ∀𝑙/𝑑(𝑝𝑙, 𝑝𝑙) > 𝛿𝑙 (3) Finally, we investigate if commonalities exist 𝑥 𝑖 𝑗 𝑥 between the clusters of successive studied days. 𝛿𝑙 |𝑡𝑙−𝑡𝑙|. 𝑥−𝛿𝑙 𝑙 𝑖 𝑗 𝑡 𝛿𝑙 𝑙 𝑙 𝑙 2.1. Modeling similarity between 𝜃𝑡 = 𝑒 𝑡 ∀𝑙/|𝑡𝑖 , 𝑡𝑗 | > 𝛿𝑡 (4) individual trips Otherwise, these penalties are null. In this manner, Firstly it is essential to define a similarity function 𝑆(𝑖, 𝑗) = 𝑆(𝑖, 𝑗) + 𝜃𝑙 𝑙 𝑥 + 𝜃𝑡 defines a sharp function to estimate the likeness between trips. The goal of that enhances the differences between trips and such a function is to quantify the similarity between facilitates identification of similar travelers in the two trips. We use the similarity function presented dataset. Notice that 𝑆 is minimal (and equal to 1) in [11] because it was shown that it provides when the two trips are exactly identical. excellent results to estimate the similarity for the trips defined by a pair origin-destination. The 2.2. Detection of similar trips for similarity is calculated according to the spatio- individual days temporal commonalities between the trips. Let The method presented here allows us to detect 𝑆(𝑖, 𝑗) the similarity function between trips i and j. spatio-temporal areas where there are a significant From a physical point of view, the intuition is that number of similar trips. The function of similarity two (or more) travelers may have an interest to share exposed above only estimates the likeness between their trip if they start in the same neighborhood and two trips but does not detect clusters of similar trips. at the same moment, and want to go to the same In order to detect such groups, a clustering algorithm destination. The function S must encompass these is used. The function of similarity allows us to different spatio-temporal attributes of the trips. We compute the similarity matrix requested by a proposed the following function: clustering algorithm. A variant of a well-known 𝑆(𝑖, 𝑗) = ∑ 𝛼 clustering density-based method DB-SCAN [12] is 𝑙𝑒|𝑓𝑙(𝑖,𝑗)| (1) 𝑙∈[𝑃𝑈,𝐷𝑂] applied for each day to detect groups of similar trips. The classic DB-SCAN requires only two where 𝑓𝑙(𝑖, 𝑗) is the feasibility function and 𝛼𝑙 is a parameters: a threshold 𝜖 and a minimum number of coefficient. Function f describes the service's points 𝑀𝑖𝑛𝑃𝑡𝑠, which have to be in a radius 𝜖 so that potential to operate the shared trips, i.e. the ability to the studied point is considered as an element of the pick up (or drop off) the two travelers before both of their desired departure times: cluster, see [13] for more details. The parameter 𝜖 is the maximal distance between trips, i.e., the 𝑓𝑙(𝑖, 𝑗) = |𝑡𝑙 𝑙 𝑙 𝑙 𝑖 − 𝑡𝑗 | − 𝛾𝑑(𝑝𝑖 , 𝑝𝑗) (2) maximal value of 𝑆, allowed to consider them as similar and group them into the same cluster. where 𝛾 is the average pace to connect travelers that However, this method must be slightly adapted to want to share a trip. This parameter is a general and detect groups of different density. Thus, a successive synthetic formula to describe the operation of the DB-SCAN clustering is performed, i.e. Algorithm service and the way in which this service gathers two demand requests into the same vehicle: defining a meeting point, successive pick-ups, etc. For example, if the first traveler must walk to the second traveler's pick-up point, then 𝛾 is equal to the inverse of the walking speed. If this distance is traveled by car, meaning that the service offers door-to-door service, then 𝛾 is equal to the inverse of the vehicle speed. Consequently, f is positive if the match can be realized before the two desired departure times 𝑡𝑙𝑖 and 𝑡𝑙𝑗, whereas f is negative if travelers have to experience delay to make the match possible. Moreover, 𝛼𝑙 is equal to 1 if 𝑓𝑙(𝑖, 𝑗) > 0 and to 3 2 2 otherwise because it is more disadvantageous to be delayed. In addition to this first index of similarity 𝑆(𝑖, 𝑗), excessive distances/durations for rendezvous 247 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia 1 itdbscan ([14] for more information), using the similarity function 𝑆 as the distance, while updating iteratively the values of the parameters. Starting with a large value of 𝑀𝑖𝑛𝑃𝑡𝑠 (equals to 𝑚𝑖𝑛𝑃𝑡𝑠𝑀𝑎𝑥) and a small value of 𝜖 (equals to 𝑒𝑝𝑠𝑀𝑖𝑛), it makes it possible to identify large groups of travelers in the initial data set of trip. In other terms, we first detect large and high-density clusters. Then, the DB-SCAN method is applied on the remaining non-clustered trips to detect groups of size 𝑀𝑖𝑛𝑃𝑡𝑠 − 1. This process is repeated until 𝑀𝑖𝑛𝑃𝑡𝑠 = 𝑚𝑖𝑛𝑃𝑡𝑠𝑀𝑖𝑛 and 𝜖 = 𝑒𝑝𝑠𝑀𝑎𝑥. A cluster can be considered satisfactory if locations of origin / destination and arrival / departure times of the trips within the cluster are relatively close. To this end, the function 𝑆 is extended to consider sets of trips. In other terms, |𝑡𝑙 𝑙 𝑖 − 𝑡𝑗 |and are respectively 1 replaced by the mean distances, i.e. ∑𝑛𝑘 |𝑡𝑙 − 𝑛 𝑖=1 𝑖 𝑘 1 𝑡𝑙 𝑛𝑘 𝑙 𝑙 𝑗 |and ∑ 𝑑(𝑝 , 𝑝 ) where 𝑛 𝑛 𝑖=1 𝑖 𝑗 𝑘is the number of trips 𝑘 inside the cluster 𝑘. We then normalized these values because the acceptable delays are strongly related to the length of the trips. Consequently, the Figure 1: (a),(b),(c),(d) clusters with different quality index of cluster 𝑘, 𝑄(𝑘), is the function 𝑆 characteristics, the pick-up are depicted in green and applied to the set of clustered trips divided by the drop off in red. 𝒏𝒌 denotes the number of trips in the average length of the trips within cluster 𝑘.Notice cluster 𝒌, 𝒍 that we aim at minimizing the quality index, i.e. best 𝒌 denotes the average length of trips in 𝒌and 𝝉𝒌 denotes the average duration of trips in 𝒌. (e) clusters present values of 𝑄 close to zero. Indeed, Ratio of clustered trips per day in Midtown and 𝑄(𝑘) is low when (i) the spatio-temporal distance Upper East Side from 8h to 11h. (f) Example of between origins is low, (ii) the spatio-temporal demand graph for a randomly selected meta-cluster. distance between destinations is low, and (iii) the Clusters detected have different sizes, from 2 trips mean travel distance is large. It comes that a cluster to 74 trips gathered into the same group. It brings to 𝑘 is selected if and only if 𝑄(𝑘) is below a specific light that the shared mobility demand may take threshold 𝑄𝑚𝑎𝑥. In the remaining of the study, we many aspects requiring different forms of set 𝑄𝑚𝑎𝑥 at 3. It corresponds to a restrictive transportation services to be optimally satisfied. matching policy. Figure 1 depicts four clusters with different sizes and characteristics. The average travel length 𝑙𝑘 is directly the arithmetic average of the length of 𝑛𝑘 trips within the cluster 𝑘, whereas the average travel time 𝜏𝑘 is the arithmetic average of the duration of the 𝑛𝑘 trips. Figure 1.e shows the number of clustered trips and the total number of trips per day. The developed method detects almost 85%of similar trips per day on average in the studied zone. 2.3. Identification of regular demand pattern for multiple days Once this daily analysis is done, we investigate if commonalities in the clusters can be identified. Many approaches exist to derive the most representative partition from a group of partitions, such as meta-clustering or consensus learning [15]. Here, we use the same clustering method to maintain consistency when scaling-up. In the following, to reduce the computational time, we focus the study on the 14 days of the dataset for which the ratio of clustered trips is the highest: June 6 to 19, 2011. The 248 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia objective is now to find out if there are similar trips of clusters in the same spatial area. On average, each (relatively close departure and arrival locations and cluster contains 6.22 similar trips. Moreover, a times) made several times during the studied period. meta-cluster contains, on average 8.64 clusters. In These recurrent spatio-temporal areas are called other words, each meta-clusters contains on average meta-clusters. For that purpose, each cluster 53trips with very similar characteristics (see Section previously found is considered as a new trip, formed 3). The method aims to design a line of transport by the centroïd of its pick-up and the centroïd of its serving a set of meta-clusters with characteristics drop off. Centroïds correspond to the mean compatible (size, time-windows, etc.). Depending origin/destination locations and mean on the meta-clusters chosen, the number and size of departure/arrival times of the clustered trips. This vehicles required will be different. To quantify the information can be useful to design the potential demand of a tour, we plot the total number transportation supply because centroïds can be the of trips per day served by a tour going through these locations of common meeting points of the standby meta-clusters. Figure 2.c depicts the total number of areas of shared vehicles. A second clustering is then trips served per day for the set of meta-clusters performed, it returns clusters with similar depicted in Figure 2.a characteristics (without taking into account the initial day when the trips were made). In other words, two clusters are in the same meta-cluster, if their centroïds have close departure and arrival locations and times. Interestingly, we observe that more than 94% of the daily clusters are recurrent from one day to another. The representation of a meta-cluster on a 2D map is difficult to analyze because the time dimension is not accounted for. Consequently, we prefer to focus on the evolution of the daily clusters' size and the localization of the related origin/destination whereabouts. Each meta-cluster can be depicted as a graph of the demand. Figure 1.f shows the graph of the demand for a randomly selected meta-cluster. This figure shows that in the same spatio-temporal area, similar trips can be seen every day, except on weekends. Each meta-cluster provides precise information about its location, its estimated departure and arrival times and the total number of Then, the selection of a set of meta-clusters to find a trips performed per day. It is important to note that potential tour of vehicles is performed. We use the different individuals perform these trips from one median number of trips per day in a meta-cluster as day to another. However, global human mobility is an indicator of its size. For each studied period, a remarkably regular; this is a valuable insight to tune minimal median value required is defined; we filter transportation services and favor shared mobility the meta-clusters with a median inferior to this efficiently. value. This method allows us to obtain a reasonable number of meta-clusters to solve a relatively small 3. CUSTOMIZED SUPPLY DESIGN instance of the optimization problem. However, it As mentioned in the previous section, the spatio- should be noted that the solution is optimal for each temporal areas containing similar and regular trips period on which we solve the problem, but an (meta-clusters) are detected. The study's next optimal result is not guaranteed for a set of periods. objective is to find a way to serve the pick-up and The meta clusters found are used to define a Dial-a- drop off points in each of these meta-cluster while Ride Problem (DARP) instance. Given the large respecting a set of constraints: time windows, number of meta-clusters, it is impossible to directly vehicle capacity, size of the fleet, etc. A minimalist model vehicle tours from the whole set of meta- example of the developed method is depicted in clusters. That is why we focus this study only on high capacity vehicle tours. This method allows us Figure 2. Figure 2.a shows a set of 3 meta-clusters; to serve a large number of passengers by solving each of them contains several clusters of similar smaller problem instances. The main advantage of trips. A green marker and a red marker linked by a this method is that the calculation time depends on blue line depict a cluster containing several similar the number of meta-clusters served and not directly trips. The green and red markers designate of the number of passengers. That is why it is crucial respectively the points of pick-up and drop off of a to find the largest possible meta-clusters. If we solve cluster. A meta-cluster is depicted by an aggregation the problem for a set of large meta-clusters, the 249 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia number of passengers effectively served will be servicing a vertex 𝑖 the load of vehicle 𝑘 leaving significantly higher. vertex 𝑖, and 𝑟𝑘 𝑖 the ride time of user 𝑖. This model presents several interesting aspects: multiple vehicles, time-windows for Pick-up, or Drop Off. The main objective of this method is minimizing the total route length. However, several other constraints can be added, such as vehicle capacity, maximum route duration, or maximum ride time for users. Nevertheless, it is essential to note that the meta-clusters previously found are independent of the method chosen to serve them and vice versa. Indeed depending on the objective searched, an approach may be more interesting that another. For example, it could be interesting to use a method to minimize the total route length for a Transportation Network Company. From a user point of view, using a technique to reduce the waiting time could be more interesting than another. Several methods aim to satisfy an objective function depicted as a combination of constraints such as transportation time, ride time, excess of maximum Figure 2: Green markers depict the pick-up and red ride time, waiting time, time windows violations, markers the drop off (a) 3 meta-clusters randomly etc. [19]. A comparison with these sophisticated chosen between 08h15 and 08h35. (b) example of line methods will be studied in a future study. designed, serving the centroids of pick-up and drop off of each meta-cluster. (c) Total number of trips per day served on the set of 3 meta-clusters. (d) Shows the 4. RESULTS number of meta-clusters for each time slot in function This Section is devoted to the results of the proposed of the minimal median value of trips per day method for the case of NYC. First, the meta-clusters required. are presented and analyzed. Secondly, based on this There are many variants for the DARP problem, [16, demand decomposition, the optimization method is 17] give a list of them based on different objective tested and evaluated. functions. In our case, we use a variant presented in 4.1. Selection of the spatio-temporal areas [18]. This model is depicted below Eq.5 to Eq.17. The model is based on a three index formulation. Let First of all, it is interesting to analyze the 𝐺 = (𝑉, 𝐴) a directed graph. The set of vertices 𝑉 is characteristics of the meta-clusters found. As partitioned as follow : the first and the last element mentioned in Section Method - Estimation of the are two copies of the depot, elements from index 1 demand, in the studied area between 08h00 and to 𝑛 are pick-up and elements from index 𝑛 + 1 to 11h00, almost 85% of trips can be considered 2𝑛 are drop off. 𝑃 denotes the set of pick-up and 𝐷 similar. Moreover, more than 94% of trips are recurrent, i.e., these trips can be observed almost the set of drop off. A request is a couple (𝑖, 𝑛 + 𝑖), every day. 2136 spatio-temporal areas are detected where 𝑖 ∈ 𝑃 and 𝑛 + 𝑖 ∈ 𝐷. The load of each vertex as zones where there is a recurrent potential demand is defined as 𝑞 , with 𝑞 ≥ 0 for 𝑖 𝑖 0 = 𝑞2𝑛+1 = 0, 𝑞𝑖 of shared mobility. On average, each meta-cluster in {1, . . . , 𝑛} and 𝑞 contains 53 trips. Once again, it is important to 𝑖 = −𝑞𝑖−𝑛 for 𝑖 in {𝑛 + 1, . . . ,2𝑛}. A service duration 𝑑 notice that different users surely perform these trips. 𝑖 ≥ 0 with 𝑑0 = 𝑑 In the following, it is considered that the users' 2𝑛+1 = 0. 𝐾 denotes the set of vehicles. The meeting point is defined as the centroïds of pick-up capacity of a vehicle 𝑘 ∈ 𝐾 is 𝑣, and 𝑣 denotes the (respectively drop off) of a meta cluster. Thereby, it maximal duration of a route for a vehicle 𝑘. The arc is interesting to know the spatial and temporal set is defined as: 𝐴 = {(𝑖, 𝑗)|𝑖 = 0, 𝑗 ∈ 𝑃 or 𝑖, 𝑗 ∈ difference between the centroïds and the points of 𝑃 ∪ 𝐷, 𝑖 ≠ 𝑗 and 𝑖 ≠ 𝑛 + 𝑗, or 𝑖 ∈ 𝐷, 𝑗 = 2𝑛 + 1} pick up and drop off. Table 1 shows that the spatial the cost of traversing an arc (𝑖, 𝑗) with a vehicle 𝑘 is distances are close to 200m. The average temporal 𝑐𝑘𝑖𝑗, and the travel time between two nodes 𝑖 and 𝑗 is shifts are nearly 6 minutes which is entirely 𝑡𝑖𝑗. 𝐿 denotes the maximal ride time and the time acceptable. It shows that the meta-clusters found are relatively close to the initial clusters estimated from window of a vertex 𝑖 is [𝑒 𝑘 𝑖, 𝑙𝑖 ]. 𝑥𝑖𝑗 is a binary variable the real rides of users. The average travel distance equal to 1 if and only if (𝑖, 𝑗) is traversed by a vehicle and time in the meta-clusters are respectively 𝑘 ∈ 𝐾. Let 𝑢𝑘𝑖 the time at which a vehicle 𝑘 starts 1.71km and 11.1min. Although these data are not 250 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia fully representative of human mobility since they users to serve. The load of each drop off is defined only correspond to taxi trips, such a dataset provides as −𝑞𝑖. A time window of 20 minutes is defined to an attractive proxy for studying the individuals' serve the different points. This value is not routes within a city. representative of the real difference between the desired service times and the effective times of Table 1: Average spatial and temporal distances service, but it provides an upper and lower limit that between pick-up, drop off and the centroïds of the meta-clusters. should not be exceeded. If this value is not enough, the constraint will often be violated then; no solution will be found. The travel time between two nodes 𝑣 and 𝑗 is estimated according to the results presented is [10]. We set the average speed for a vehicle to 9.65𝑘𝑚/ℎ. Figure 3.a .b and .c show for each period the a map of the designed routes. Each color designates a specific transport line. Table 2 indicates In a first time, it is necessary to select a reasonable the result of the DARP. For the three periods, routes number of meta-clusters to solve a relatively low allowing to serve all the selected meta-clusters in instance of the optimization problem. As said in less than 9 seconds are found. This result proves that Section 2, the median number of trips per day in a the method is a good way to design lines serving meta-cluster is used as an indicator of its size. Figure many users (more than 600 trips per day on average). 2.d shows us for each one hour period the number of Besides, for each period, we calculate the average meta-clusters depending on the chosen minimal delays and time advances for each point served by median value. In other words, a meta-clusters is the optimized tour. This value is estimated by the counted if and only if its median value of trips per difference between the wished times of departure day is greater or equal to the chosen value. It is and arrival (given by the centroïds of the meta-important to note that the number of points clusters) and the hour of service given by the solving effectively treated in the DARP will be for each of DARP. These values show that the developed period 2 ∗ 𝑛𝑢𝑚𝑏𝑒𝑟𝑜𝑓𝑐𝑙𝑢𝑠𝑡𝑒𝑟𝑠, because a vehicle method relatively little impact on demand. Indeed, serve a pick-up and a drop-off for each meta-cluster. on average, the delay is 12 minutes and the advance According to Figure 2.d, the minimal median value is 10.6 minutes, which is acceptable since the 24 has been selected to find potential routes with a number of users served is high. large number of users with very short execution times. To the best of our knowledge, there are no classical optimization methods to find round serving such a 4.2. Demand-driven route optimization quantity of similar and recurrent trips in such a tight timeframe. The theoretical studies on DARP [17] According to the results showed in Section 3.1, three show that the exact method used in this paper can periods of one hour for which the meta-clusters solve instance until 36 points. It would not be contain at least a median of 24 trips per day are possible to solve instances with so many passengers selected. Figure 3.a shows the potential number of without using an aggregation method in meta- trips per day that can be served on the period 08h00- clusters. 11h00. This result illustrates one of the method's interests: the 18 meta-clusters selected represent The existing dynamic methods such as [20,21,22] actually 614 trips per day on average. The median obtain trips delay between 2 and 6 minutes; number of trips served per day for this set of meta- however, these services work with large fleets of clusters is 756. Also, we note that the demand is vehicles with limited capacities (between 2 and 10). extremely regular every day of the week (except Moreover, these methods work only on networks weekends) for the two weeks of analysis. In the case with a limited number of nodes. of an effective implementation of optimized lines, it would be interesting for the service to be operated from Monday to Friday. The parameters used for DARP are adjusted for each period of one hour. Each centroïd of pick-up and drop-off of the meta-clusters are inserted in the sets 𝑃 and 𝐷. The number of vehicles 𝑉 for each period is depicted in Table 2. For each vehicle, its capacity 𝑄𝑘 = 80, which corresponds to the average capacity of a bus. For each arc (𝑖, 𝑗), the cost 𝑐𝑖𝑗 is defined as the spatial distance between 𝑖 and 𝑗. For each node 𝑖, we set the service duration 𝑑𝑖 = 2 minutes. The load of each pick-up 𝑞𝑖 is defined as the number of 251 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia Table 2 : Result of the search of rounds for the 3 time Making it possible to serve on average more than periods from 08h00 to 11h00. 600 trips per day with calculation times lower than 9 seconds. This hybrid method between classical and dynamic approaches allows to design high capacity lines based on the real demand of mobility. Moreover, it allows to obtain lines with restricted spatio-temporal deviations from the demand described by the centroïds of the meta-clusters. Setting up massive lines close to the initial demand of users provides a partial response to the last mile problem, which is one of the main obstacle to shifting users of private vehicles to shared modes of transport. As we mentioned in the introduction, this study is part of the adaptation of current methods to new requirements for the deployment of autonomous vehicles. The method presented makes it possible to overcome the incompatibilities between current methods and new approaches adapted to the needs of services based on the use of autonomous vehicles. Several ways are studied in order to complete the current method. The first is to take real-time aspects into account in the method. This can be done in several ways, either with instant classification or by taking into account the results obtained to anticipate future demand. Another interesting aspect is the design of more or less dynamic lines according to the demand in a studied area. For example, Figure 3: (a) depicts the total number of users depending on the number of users and the required effectively served in the set of meta-clusters selected responsiveness of the service, different solutions can in Section 3.1. (b),(c),(d) shows for each period be implemented: classic or dynamic bus lines, taxi, presented in Table 2, the customized lines found. etc. Finally, taking into account the dynamic aspects of the network to choose routes according to the 5. CONCLUSION network's particular events: congestion, roadworks, This article presents an optimization method based etc. seems to be an excellent way to improve the on a decomposition of the demand and a resolution current method. Finally, the method's scalability will of DARP on a reduced instance. This data-driven be widely studied to maximize the number of data method allows in a first time to identify clusters of processed and thus the veracity of the results similar and regular trips over time (meta-clusters). obtained. Then these meta-clusters are considered as points to serve in an instance of DARP. This method's main REFERENCES interest is to design tours of vehicles to serve a large [1] Vuchic, V. R. and A. Musso, Theory and number of potential users. As shown in Section 3, practice of metro network design. Public the main advantage of this method is that the transport international, Vol. 3, 1991, p. 298. execution time of the optimization problem does not depend on the number of users served, but only on [2] Carrese, S. and S. Gori, An urban bus network design procedure. In Transportation planning, the number of meta-clusters served. Springer, 2002, pp. 177–195. As part of this study, the analysis of pattern [3] Kocur, G. and C. Hendrickson, Design of local recurrence was carried out over two weeks. bus service with demand However, the proposed method makes it possible to equilibration.Transportation Science, Vol. 16, detect regular patterns over much more extended No. 2, 1982, pp. 149–170. periods. This can be particularly useful in the case of [4] Agatz, N., A. Erera, M. Savelsbergh, and X. effective implementation of transport lines. Besides, Wang, Optimization for dynamic ride-sharing: it is possible to integrate many other parameters into A review. European Journal of Operational the objective function of DARP to design lines as Research, Vol. 223, No. 2, 2012, pp. 295–303. close as possible to actual user demand. The results obtained show us that rounds of large- capacity vehicles (80 peoples) can be identified. 252 YRS 2021 Cyril Veve, Nicolas Chiabaut 15. -17. September DEMAND-DRIVEN OPTIMIZATION METHOD FOR SHARED MOBILITY SERVICES Portoroz, Slovenia [5] Najmi, A., D. Rey, and T. H. Rashidi, Novel [14] Veve, C. and N. Chiabaut, Estimation of the dynamic formulations for real-time ride-sharing potential demand of shared mobility services systems. Transportation Research Part E: through clustering of similar individual trips. Logistics and Transportation Review, Vol. 108, Transportation Research Board 99th Annual 2017, pp. 122–140. meeting. No. 20-00168, 2020. [6] Zmud, J. P. and I. N. Sener, Towards an [15] Lopez, C., L. Leclercq, P. Krishnakumari, N. understanding of the travel behavior impact of Chiabaut, and H. Van Lint, Revealing the day- autonomous vehicles. Transportation research to-day regularity of urban congestion patterns procedia, Vol. 25, 2017, pp. 2500–2519. with 3D speed maps. Scientific Reports, Vol. 7, No. 1, 2017, p. 14029. [7] Fagnant, D. J. and K. M. Kockelman, Dynamic ride-sharing and fleet sizing for a system of [16] Ho, S. C., W. Szeto, Y.-H. Kuo, J. M. Leung, M. shared autonomous vehicles in Austin, Texas. Petering, and T. W. Tou, A survey of dial-a-ride Transportation, Vol. 45, No. 1, 2018, pp.143– problems: Literature review and recent 158. developments. Transportation Research Part B: Methodological, Vol. 111, 2018, pp. 395–421. [8] Bongiovanni, C., M. Kaspi, and N. Geroliminis, The electric autonomous dial-a-ride problem. [17] Cordeau, J.-F. and G. Laporte, The dial-a-ride Transportation Research Part B: problem: models and algorithms. Annals of Methodological, Vol. 122, 2019, pp. 436–456. operations research, Vol. 153, No. 1, 2007, pp. 29–46. [9] Ferreira, N., J. Poco, H. T. Vo, J. Freire, and C. T. Silva, Visual exploration of big spatio- [18] Cordeau, J.-F., A branch-and-cut algorithm for temporal urban data: A study of new york city the dial-a-ride problem. Operations Research, taxi trips. IEEE transactions on visualization and Vol. 54, No. 3, 2006, pp. 573–586. computer graphics, Vol. 19, No. 12, 2013, pp. [19] Jorgensen, R. M., J. Larsen, and K. B. 2149–2158. Bergvinsdottir, Solving the dial-a-ride problem [10] NYC Department of Transportation, New using genetic algorithms. Journal of the York City Mobility Report. operational research society, Vol. 58, No. 10, https://www1.nyc.gov/html/dot/downloads/pdf/ 2007, pp.1321–1331. mobility-report-singlepage-2019.pdf, 2019. [20] Santi, P., G. Resta, M. Szell, S. Sobolevsky, S. [11] Chiabaut, N. and C. Veve, Identifying Twin H. Strogatz, and C. Ratti, Quantifying the Travelers Using Ridesourcing Trip Data. benefits of vehicle pooling with shareability Transport Findings, 2019. networks. Proceedings of the National Academy of Sciences, Vol. 111, No. 37, 2014, pp. 13290– [12] Ester, M., H.-P. Kriegel, and X. Xu, A Density- 13294. Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise, 1996, p. 6. [21] van Engelen, M., O. Cats, H. Post, and K. Aardal, Enhancing flexible transport services [13] Ester, M., H.-P. Kriegel, J. Sander, X. Xu, et al., with demand-anticipatory insertion heuristics. A density-based algorithm for discovering Transportation Research Part E: Logistics and clusters in large spatial databases with noise. In Transportation Review, Vol. 110, 2018, pp. Kdd, 1996, Vol. 96, pp. 226–231. 110–121. [22] Alonso-Mora, J., S. Samaranayake, A. Wallar, E. Frazzoli, and D. Rus, On-demand high- capacity ride-sharing via dynamic trip-vehicle assignment. Proceedings of the National Academy of Sciences, Vol. 114, No. 3, 2017, pp. 462–467. 253 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Laura Vovčko, MSc (Geol.) Assist. Prof. Dr. Stanislav Lenart (Civ. Eng.) Slovenian National Building and Civil Engineering Institute Dimičeva ulica 12, Ljubljana, Slovenia laura.vovcko@zag.si ABSTRACT To establish a sustainable mobility, railway transport infrastructure plays very important role. By implementing the environmental management system, railways are for the upholding the highest standards in environmental care. The demands for a clean and healthy environment are included in all aspects of different railway operations as well as in construction and maintenance activities at railway infrastructure. Durability of the railways is highly related to the type of railway track and the fragility of used material. This encourages us to find the optimum railway sleeper material which suits to technical requirements and provide the highest resistance to abrasion of ballast material in contact with railway sleeper. This research presents the effect of material type of railway sleepers upon the degradation of ballast material during cyclic loading. Seven different compositions of railway sleeper and one type of ballast material were used in the research. Sleeper materials include wood and concrete, while the latter was improved by replacing coarse aggregates with rubber aggregates. Furthermore, under sleeper pads of geotextile and rubber layers were added. Each type of railway sleeper resting on ballast layer was loaded by one million load cycles of a maximum amplitude equal to the maximum rail traffic load. Besides deterioration of ballast material and damage of interected sleeper material, also permanent and resilient vertical displacements were observed. Structural (hysteresis) damping and stiffness characteristics vs. number of load cycles were evaluated. Their changes during different life-cycles of ballast track were observed. Keywords: Ballasted railway track, Railway sleeper, Cyclic loading, Ballast material, Deformation characteristics 1. INTRODUCTION and the type of railway sleeper based on their behaviour under cyclic loading. Durability of the railway track is highly related to the fragility of ballast material which is placed beneath The most appropriate ballast materials generally the railway sleepers to keep them in place during the consist of coarse-grained hard minerals (Raymond, train passage. Ballast layer presents a significant part 1985). Although the sedimentary and metamorphic of the railway structure, its primary objective is to rock types are in general weaker than igneous rock distribute the load through the sleepers to the ground type, the properties cannot be unambiguously beneath and to prevent their over stressing. To determined based only on the rock type. The achieve a sufficient load capacity and durability of formation mechanism of rocks namely presents a the railway structure, changes in ballast material significant impact on the rock, respectively ballast properties during cyclic loading must be anticipated. material properties. According to Watters et al. Cyclic loading can namely cause deformation and (1987) the hardness of a rock is not depended only on degradation of ballast material which can lead to the hardness of the component minerals but also on premature destruction of railway track quality and mineral cleavage fracture, textural attributes and in consequently more frequent maintenance case of clastic sedimentary rocks on the degree of performance to keep track functional. The predicitve cementation or consolidation. Furthermore, since maintenance can be significantly reduced with rocks are aggregates of many mineral grains or previous choosing the appropriate ballast material crystals, their properties depend of their various contained minerals. These properties are in general 254 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia determined by averaging the relative properties and composite railway sleepers consist of three sometimes also orientations of the various grains or classifications (Ferdous et al., 2015). Researchers crystals. Many properties also depend on grain or have shown that increase in rubber content in crystal size, shape, packing arrangement, the concrete railway sleeper leads to reduction in temperature and pressure, etc. (Klein & Carmichael, compressive strength. In general, the concrete 2020). sleepers have low tolerance for impact loading conditions (Remennikov & Kaewunruen, 2007). The Because of the large range of factors that affect the reduction of compressive strength appeared at even mechanical properties of rock, the mechanical tests very small consumption of rubber content in are considered as the most appropriate indicators for concrete. By adding the rubber into the concrete the evaluating the individual ballast material properties ability to absorb the vibration energy of concrete can and their deformation and degradation behaviour. In be improved (Meesit, 2017). most countries the requirments for ballast materials are defined by limit values for the Los Angeles By increasing the demand to heavier traffic load and abrasion test and the micro-Deval test. Several running speed a better insigne into the dynamic researchers (Wright 1983; Selig & Boucher 1990) characteristic of railway sleepers and ballast material have shown that these conventional mechanical tests is crucial. The typical loading frequency of traffic for ballast material do not reflect the actual field loading in the track is normally around 8-10 Hz for a conditions during train loading. In Europe some normal train and may reach 30 Hz for a high-speed studies of mechanical properties of ballast material train (Aursudkij et al., 2009). In this study, the effect and its behaviour in railway track under cyclic of material composition of railway sleeper upon the loading have been executed from 1970 to 1975. Since degradation of the ballast material has been then, various types of experiments have been investigated under cyclic loading. Each type of developed to evaluate the ballast properties. railway sleeper installed on ballast layer was loaded Regarding to Sadeghi et al. (2016) the characteristic by one million load cycles of maximum amplitude of ballast material to insure better overall equal to the maximum railway traffic load. Beside performance of the railway structure can be divided deterioration of ballast material and damage of into mechanical, environmental, physical and interacted ballast particles, also permanent and geometry of ballast profile category. These categories resilient vertical displacements were observed. are defined with establishing an estimation of the Structural (hysteresis) damping and stiffness resistance to in track instability and degradation characteristics vs. number of load cycles were under loading. During loading the voids between evaluated and their changes during different life- larger aggregates can fill with finer particles resulting cycles of ballast track were observed. from the degradation of ballast grain fouling. The foulding process causes various changes in the 2. RAILWAY STRUCTURE gradation of ballast. According to Selig & Walters A railway structure (Fig. 1) consists of rails resting (1994) and Meeker (1990), the fouling can be caused on rail-pads which are laid on railway sleepers that by mechanical particle degradation during are placed on top of a railway bed – ballast layer. The construction and maintenance work and traffic loads from the trains are applied by the vehicles onto loading, with the chemical weathering related from the rails and the rail-pads are reducing their environmental changes, migration of fine particles vibrations. Railway sleepers ensure certain amount of from the surface and the underlaying layers, the flexibility and elasticity to quickly return back to its spillage from freight traffic and degradation of original shape after the train passing. During the train sleepers. In general, the awareness of changes that passage the railway sleepers transfer the pressure can lead to degradation of ballast material must be from rails to railway bed which consists of ballast and taken into consideration to avoid further maintenance sub-ballast layer above a subgrade. difficulties. While the cyclic loading can lead into the damage of railway sleepers or even the breakage of ballast layer, also the vibration response of sleepers under loading must be taken into the consideration in analysis and the railway structure design procedure. The vibration response to cyclic loading depends of a type of railway sleeper. Nowadays, composite railway sleepers present a significant role in the railway development. Several composite sleeper technologies have been developed to provide the Source: (Kumawat et al., 2019). alternative types of sleepers with ensuring Figure 1: Railway structure. sustainability from an environmental perspective. Based on the amount, length and orientation of fibres 255 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia While the ballast layer is subject to deformation and weathering, rough surface and minimum hairline degradation due to train passing, the ballast particles cracks (Indraratna et al., 2007). Particle should tend to breakdown or to develop internal cracking consist of a hard and durable rock. The particles must which leads to weakening of the railway bed. sustain durable at external impact (e.g., atmospheric According to Selig and Waters (1994) and Ionescu conditions, loading, …) which can cause their (2004) one of the responsible factors for decrease of breakage. The characteristics of durability can be ballast efficiency is the contamination of ballast with obtaining with aggregate crushing and wet attrition, finer particles. as well as with wet strength and wet/dry strength variation and Los Angeles test. The ballast particle Therefore, beside an appropriate design of railway size distributions are very uniform, some well-graded structure (e.g., thickness of layers, size and spacing of the sleepers …), the properties of ballast material distributions could namely reduce drainage capacity and increased fouling risk (Indraratna et al., 2007). and the effect of pressure transfer from railway Ballast material with well-graded distributions can be sleepers must be taken into consideration. Generally, compacted to higher densities than uniform material. physical properties of ballast material can be divided Indraratna et al. (2003) has shown the effect of into physical properties of individual particle and particle size distribution on axial and volumetric physical properties of the whole ballast layer. strain behaviour during cyclic loading. They found Regarding to Indraratna et al. (2006) the first group out that the very uniform and uniform material includes petrological test, particle shape and surface exhibit higher axial and volumetric strain. Most of examinations and durability tests, while the second ballast initial properties changes under dynamic group considers the permeability, void ratio, bulk loading due to breakage, deformation and fouling. density and specific gravity. According to Indraratna et al. (2005) the aggregate 2.1. Ballast layer degradation can be represented by the ballast breakage index (BBI) which employs the change in 2.1.1. Ballast material the fraction passing a range of sieve size and it is estimated from incorporating a linear particle size Generally, ballast material is a free-draining granular axis. The BBI can be define with equation 1, where material (Fig. 2) used as load-bearing material. It p.s.d.c presents the particle size distribution curve needs to be tough enough to resist breakage under and a.b. the arbitrary boundary. external impact, hard enough to resist abrasion due to inter particle contact and dense enough to resist their 𝐴𝑟𝑒𝑎 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡ℎ𝑒 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 𝑎𝑛𝑑 𝑓𝑖𝑛𝑎𝑙 𝑝. 𝑠. 𝑑. 𝑐. lateral deformation. Furthermore, the ballast material 𝐵𝐼 = 𝐴𝑟𝑒𝑎 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡ℎ𝑒 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 𝑝. 𝑠. 𝑑. 𝑐. 𝑎𝑛𝑑 𝑡ℎ𝑒 𝑎. 𝑏. must be also freeze-thaw resistant; therefore, voids which could be filled with water must be avoided. (1) Voids filled with water can namely lead to inter particle volume change during freeze-thaw process. 2.2. Railway sleeper The mechanical degradation of ballast material Railway sleepers (Fig. 3) were first made form produces finer particles that reduces the draining timber; later the steel and concrete types of sleepers capacity of the railway track and can lead in internal became more and more popular (Remennikov & pore water pressure development which can cause Kaewunruen, 2006). During the last few decades, the significant deterioration of railway tracks (Chanda & railway industry has focused on a cement-based Krishna, 2003; Indraratna & Khabbaz et al. 2003, concrete rather than timber and steel sleepers. Mono- Ionescu, 2004). block prestressed concrete sleepers were first applied in 1943 and are now used in heavy haul and high- speed railway track constructions (Kaewunruen, 2010). In recent years, composite sleepers have been produced using new technologies. Different kinds of Source: (Sang Yeob et al., 2021). railway sleeper can be applied in different area and every type of railway sleeper has its own advantages Figure 2: Freeze-thaw process. and disadvantages. The draining and compaction ability of ballast layer Wooden railway sleeper is a traditional type of depends of their void ratio which is related by the railway sleeper suitable for low-speed railway tracks. particle size and gradation (Sadeghi et al., 2016). To According to the types of wood, they can be divided reinsure appropriate void ratio, ballast material is into hardwood sleepers (oak, jarrah, etc.) and generally composed of medium to coarse-sized softwood sleepers (douglas, etc.). Wooden sleeper is particles (10-60 mm) with only small percentage of light and therefore easy to transport, install and smaller particles (particles less than 20 mm). Ballast maintain. On the other hand, it can be easy affected material should contain angular particles with high by humidity and it is therefore is not durable. specific gravity, high shear strength characteristics, high toughness and hardness, high resistance to 256 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia Concrete railway sleeper is mostly made from pre- present an issue when the sleepers are installed in stressed concrete and usually used in the high-speed bridge. They have excellent design flexibility, good railway. Due to the prestressing, the damage form flexural and shear strength, easy drilling and good external pressure is reduced and therefore the fire performance. The production of these sleepers is prestressed sleeper has higher barring capacity. In quite slow and can increase manufacturing cost. comparison to wooden sleeper is has longer service life and it needs less maintenance. It also offers better stability due to its weight, but on the other hand the later presents also some difficulties regarding handling it. Due to its weight, it has also some limitation in application (e.g., bridges and crossing). Concrete sleepers are defined as mono-block or twin- block. In general, the concrete sleepers have damping ratio around 0,1 – 2,0 %, depending of support conditions and level of prestressing (Remennikov & Kaewunruen, 2006). Steel railway sleeper is formed by pressed steel and with a trough-shaped section. It is stronger than wood Source: (http://www.railway-fasteners.com/news/comparison-of- railway-sleepers.html) and cheaper than concrete. Steel sleeper is easy to install and it is also easy to handle. It has a long Figure 3: Different types of railway sleepers; wooden service life and it is recyclable. The problem presents railway sleeper (a), concrete railway sleeper (b), steel a chemical corrosion and high maintenance cost. railway sleeper and composite railway sleeper (d). Composite railway sleeper is known also as plastic 3. TESTING METHOD sleeper and it presents a mixture of plastic and used waste rubber. These type of sleepers combines the 3.1. Ballast material pliability of wooden and durability of concrete The ballast material used in this study was limestone, sleepers. Composite sleepers are good in reducing the a prevalent Slovenian sediment rock. Material vibration. They are cuttalbe and therefore easy to obtained from the quarry was firstly cleaned with handle. Due their recyclable ability they can be water to remove dust and fine-grains adhering to the reused into new sleepers. They have a long service aggregates and then dried before the sieve analyse. life; the disadvantage presents their cost. According Initial ballast particle distribution according to SIST to Ferdous et al. (2015) the composite sleepers can be EN 933-1:2012 and SIST EN ISO 17892-4:2017 has divided into following types: shown a uniform coarse-grained material Type-1 Composite railway sleepers can consist of composition. Particles had sharp angular edges. recycled plastics (plastic bags, etc.) or also bitumen 3.2. Railway sleepers with fillers (gravel, sand, etc.). Their structure behaviour is mainly polymer driven and they offer a In study, seven different types of railway sleeper range of benefits including ease of drill and cut, good were used (Table 1). Sleeper material included wood durability, consumption of waste materials, etc. and concrete, while the latter was also improved by However due low strength and stiffness they suffer replacing coarse aggregates with rubber aggregates. from limited designs flexibility. Furthermore, under sleeper pads of geotextile and rubber layers were added (Fig. 4). Type-2 Sleepers with long fibre reinforcement in the longitudinal direction and no or very short random fibre in the transverse direction. These sleepers are suitable for ballast railway track where the stresses in sleepers are governed by flexural and shear forces. These types of sleepers are easy to drill and cut, they have good durability, and flexural strength and modulus of elasticity are the advantages. Low shear strength and shear modules, limited design flexibility, marginal fire resistance and high price are some of the challenging issues associated with this type of category. Type-3 Sleepers with fibre reinforcement in longitudinal and transverse directions. The disadvantages can present a non-ductile behaviour of glass fibre reinforced polymer sleeper and can 257 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia Table 1: Types of railway sleepers used in study. characteristics vs. number of load cycles were Test Test evaluated and their changes during different life- Type of railway sleeper Type of railway sleeper cycles of ballast track were evaluated. After the no. no. cyclic loading, the ballast was removed and the ballast particles were sieved again according to SIST EN 933-1:2012 and SIST EN ISO 17892-4:2017. For each individual type of sleeper three tests were 1 Wood performed. Concrete (10 2 Concrete 5 % rubber) Concrete (10 Concrete % rubber) 3 with 6 with rubber rubber pad pad Concrete Concrete (10 % with 4 7 rubber) with geotextile geotextile pad pad Figure 5: Ballast material during cyclic loading. 5. RESULTS 5.1. Grain size distribution Figure 4: Types of railway sleepers used in study. According to SIST EN 933-1:2012 and SIST EN ISO 4. TEST PERFORMANCE 17892-4:2017 the sieve analyses was made for initial ballast material (IBM) before loading and for the The effect of material composition of the railway same material after the cyclic loading (BM) for all sleepers upon the ballast material during cyclic types of sleepers. Particle size distribution curves loading has been evaluated (Fig. 5). Therefore, the presents relation between cumulative percentages initial ballast material was placed into the steel passing vs. sieve size. The coefficients of uniformity chamber (50 cm x 50 cm x 30 cm) in a combination (Cu) and the ballast breakage index according to with individual type of sleeper. For the test each 10- Indraratna et al., (2005) were quantified. The cm ballast layer was at first weighted and then breakage potential is considered to be the area dropped in the chamber box. The whole box was between the particle size distribution and an arbitrary filled with 145 kg of dry ballast material; the dry reference line connecting the point between the density of 1933 kN/m3. intersection of d95 of the largest sieve size and the The cyclic load was applied to the rail-sleeper by minimum particle size of 2.40 mm. Obtained hydraulic actuator through a cylindrical steel ram. characterisitcs are summarized and presented in Each type of railway sleeper resting on ballast layer Table 2 and Fig. 6. was loaded by one million load cycles of maximum amplitude equal to the maximum railway traffic axle load of 250 kN. Beside deterioration of ballast material also vertical displacements were measured. Structural (hysteresis) damping and stiffness 258 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia Table 2: Characterisics of initial ballast material (IBM) and ballast material upon each individual sleeper type after the cyclic loading (BM). N Cu BBI o Composition . [-] [-] Initial ballast material 0 1.54 - = IBM IBM + Wood 1 1.55 0.1595 = BM Wood Figure 7: Force-time curves. IBM + Concrete 2 1.60 0.2317 = BM Concrete IBM + Concrete with rubber pad 3 1.57 0.1754 = BM Concrete + rubber pad IBM + Concrete with geotextile pad 4 1.57 0.1469 = BM Concrete + geotextile pad IBM + Concrete (10 % rubber) 5 1.56 0.1279 = BM Concrete (10 % rubber) Figure 8: Vertical displacement-time curves. IBM + Concrete (10 % rubber) with 6 rubber pad 1.56 0.1164 6. ANALYSIS OF RESULTS = BM Concrete (10 % rubber) + rubber pad According to SIST EN 933-1:2012 and SIST EN ISO IBM + Concrete (10 % rubber) with 17892-4:2017, the initial ballast material was 7 geotextile pad 1.56 0.1192 classified as uniform coarse-grained material with a = BM Concrete (10 % rubber) + geotextile pad maximum particle size of 58 mm. It is evident, that the particle distribution curves of ballast material tested upon different types of railway sleepers lies above the distribution curve of initial ballast material (Fig. 6). Cyclic loading namely causes an increase of particle breakages and thus the shift of particle size distribution line. Changes in particle size distribution are evident in coefficients of uniformity (Cu) and ballast breakage index (BBI) as presented in Fig. 9. Figure 6: Particle size distirbution curves for intact ballast material and ballast material for each individual sleeper material after the cyclic loading. 5.2. Cyclic loading Each type of railway sleeper resting on ballast layer was loaded by one million load cycles of maximum amplitude equal to the maximum railway traffic load, corresponding to axel load of 250 kN. Regarding the area of testing part of a sleeper, a sinusoidal cyclic load was applied with an amplitude of 14 kN. During Figure 9: Coefficients of uniformity (Cu) and ballast breakage index (BBI) of ballast material upon the test a vertical displacement was measured and the different type of railway sleeper after the cyclic relations between applied force and vertical loading. displacement, they are presented in Fig. 7 and 8. Structural (hysteresis) damping and stiffness characteristics vs. number of load cycles were evaluated and their changes during different stages of 259 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia service life of ballast track, namely after different material degradation when using the concrete railway number of applied load cycles were analysed. sleepers with added rubber aggregates indicated also According to Sas and Gluchowski (2012), the low BBI value in Fig. 9, while at tests with other resilient module (Mr) is evaluated by dividing the types of railway sleeper have higher BBI values. axial stress (σd) by the resilient strain (ɛa). The axial stress was obtained by applied force and cross- 7. CONCLUSION sections area of the sample, while the resilient strain The performance of ballast track under various types was obtained by resilient displacements in one cycle of sleepers was evaluated with this study. Laboratory with the initial height of the sample. The relation test results indicate that the particle breakage depends between axial stress versus resilient strain and on the type of railway sleeper. According to the obtained resilient modulus at each cyclic number are particle breakage indexes (BBI), the addition of presented in Fig. 10 and 11. rubber aggregates to the concrete railway sleeper lowers the ballast material degradation. Particle breakage index also indicates that adding pads of geotextile or rubber under the concrete railway sleeper without rubber aggregates within concrete lowers the ballast degradation, while pads under the sleeper made of concrete with added rubber aggregates has no significant impact upon the ballast degradation. The cyclic loading tests conducted in the present study could help to understand the long-term Figure 10: Cyclic loading curves. response of a railway track under different types of railway sleepers. We can conclude that the resilient modulus became stable after the certain number of loading cycles. The largest deformations are therefore evident at an early stage of cyclic load, in our study within 10 loading cycles. At this stage the granular material is probably influenced by compaction. Resilient modules are the highest by using the concrete railway sleeper and get lower by adding pads of geotextile or rubber under it. Small material degradation by using the concrete railway Figure 11: Resilient modulus-cyclic number curves. sleepers with added rubber aggregates due its ability Generally, a slight increase in modulus with an in compere with other types of railway sleepers are increase of load cycles number is observed (Fig. 11). also evident from BBI values. At the beginning of each stress level, the resilient modulus becomes stable in 10 loading cycles. REFERENCES Afterward, the cyclic loading lids to produce a [1] Aursudkij, B., McDowell, G. R., & Collop, A. resilient vibration that results in a stable deformation. (2009). Cyclic loading of railway ballast under The resilient behaviour is mainly governed by the triaxial conditions and in a railway test facility. contacts of ballast particles. The highest resilent Granular Matter, 11(6), 391-401. modules is due its low ability to absorb vibration [2] Chanda, M., & Krishna, R. (2003). Selection and energy evident at concrete railway sleeper an it testing of ballast stones for underground railway lowers by adding pads of geotextile or rubber under tracks. African Journal of Science and it. Within 100 and 100 000 loading cycles some small Technology, 4(2), 42-50. increases and decreases of resilient modules are [3] Ferdous, W., Manalo, A., Van Erp, G., noticeable at test were concrete railway sleepers with Aravinthan, T., Kaewunruen, S., & Remennikov, added rubber aggregates and concrete with pads of A. M. (2015). Composite Railway Sleepers – rubber were used. Small changes of their resilient Recent developments, challenges and future modules were probably caused by some negligible prospects. Composite Structures, 134, 158-168. reorientation and particle brakeage which led to more stable reorientation that was highly resistant even at [4] Indraratna, B. H., Khabbaz, H., & Lackenby, J. (2003). Behaviour of railway ballast under additional 1000 0000 loading cycles. At the high dynamic loads based on large-scale triaxial degree of cyclic loading the particle degradation can testing. AusRAIL PLUS (pp. 17-19). Sydney, be developed again. The latter is evident as a decrease Australia. in resilient module at 1000 0000 loading cycles at test that included concrete, concrete with added pads of [5] Indraratna, B. H., Lackenby, J., & Christie, J. (2005). Effect of confining pressure on the geotextile and at wooden railway sleepers. Small 260 Laura Vovčko, Stanislav Lenart YRS 2021 15. -17. September CYCLIC LOAD TESTS OF RAILWAY SLEEPER-BALLAST INTERACTION Portoroz, Slovenia degradation of ballast under cyclic loading. [14] Raymond, G. P. (1985) Research on Railroad Geotechnique, 55(4), 325-328. Ballast Specification and Evaluation. Transportation Research Record, 1006, 1-8. [6] Indraratna, B., Khabbaz, H., Salim, W., & Christie, D. (2006). Geotechnical properties of [15] Remennikov, A. M., & Kaewunruen, S. (2006). ballast and the role of geosynthetics in rail track Experimental Investigation on Dynamic Railway stabilisation. Ground Improvement, 10(3), 91- Sleeper/Ballast Interation. Experimental 101. Mechanics, 46, 57-66. [7] Indraratna, B., Shahin, & M. A., Salim, W. [16] Remennikov, A. M., & Kaewunruen, S. (2007). (2007). Stabillisation of granular media and A review of loading conditions for railway track formation soil using geosynthetics with special structures due to train and track vertical reference to railway engineering. Ground interaction. Structural Control and Health Improvement, 11(1), 27-43. Monitoring, 15(2), 207-234. [8] Ionescu, D. (2004). Ballast Degradation and [17] Sadeghi, J. M., Ali Zakeri, J., & Emad Motieyan Measurment of Ballast Fouldin. 7th Railway Najar, M. (2016). Developing Track Ballast Engineering Conference (pp. 169-180). London, Characteristics Guideline in Order to Evaluate Its UK: Commonwealth Institute. Performance. International Journal of Railway, 9(2), 27-35. [9] Kaewunruen, S. (2010). Sleepers and fastening. In: Track design fundamentals, Rail Corporation. [18] Sang Yeob, K., Junghee, P., Wonjun, C., Jong- Sub, L., & Santamarina, C. (2021). Soil [10] Klein, C., & Carmichael, R. S. (2020). Response during Globally Drained and Encyclopedia Britannica. Undrained Freeze-Thaw Cycles under Deviatoric https://www.britannica.com/science/rock- Loading. Journal of Geotechnical and geology. Geoenvironmental Engineering, 147(2), 14-31. [11] Kumawat, A., Raychowdhury P. & Chandra, S. [19] Selig, E. T., & Boucher, D. L. (1990). Abrasion (2019). Frequency-Dependent Analytical Model Tests for Railroad Ballast. American Society for for Ballasted Rail-Track Systems Subjected to Testing and Materials. Geotechnical Testing Moving Load. International Journal of Journal, 13(4), 301-311. Geomechanics, 19(4). [20] Selig, E. T., & Waters, J. M. (1994). Track [12] Meeker, L. E. (1990). Engineering and economic Geotechnology and Substructure Management. factors affecting the cost of railroad ballast-MSc. London: Thomas Telford Publishing. Thesis. Reno, U.S. [21] Watters, B. R., Klassen, M. J., & Clifton, A. W. [13] Meesit, R. (2017). Vibration Characteristics of (1987). Evaluation of Ballast Materials Using Micro-Engineered Crumb Rubber Concrete for Petrographic Criteria. Transportation Research Railway Sleeper Applications. Journal of Record, 1131, 45-63. Advanced Concrete Technology, 15(2), 55-66. [22] Wright, S. E. (1983). Damage caused to ballast by mechanical maintenance techniques. British Rail Research Record, 16, 174-180. 261 Panagiotis Zeibekis, Melina Karagkouni, Dimitrios Dimitriou, Maria Sartzetaki YRS 2021 VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS 15. -17. September SOCIOECONOMIC FOOTPRINT Portoroz, Slovenia VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS SOCIOECONOMIC FOOTPRINT Panagiotis Zeibekis, Researcher Melina Karagkouni, Researcher Dimitrios Dimitriou, Professor Associate Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece, pzeimpekis@yahoo.gr, melkaragouni@gmail.com, ddimitri@econ.duth.gr Maria Sartzetaki, Senior researcher/Adjusted lecturer Department of Economics, Democritus University of Thrace Panepistimioupoli, Komotini, Greece, University Neapolis, Pafos, Cyprus msartze@econ.duth.gr ABSTRACT The complexity of decision-making for investments in energy sector addressed to a variety of stakeholders with different expectations and to many business and financial risks and uncertainties. This especially applies to decisions regarding the implementation of capital-intensive energy projects where large amounts of capital are reserved. There are many cases where decision-makers do not clearly distinguish between the investment likelihood alternative scenarios and the socioeconomic footprint of the project. This paper deals with an assement framework to evaluate the socioeconomic footprint of energy projects in terms of economic, social, and environmental issues affected the decision process and the project development strategy. By a system of system approach key cost and benefits variables are highlighted and a series of key performance indicators are presented. The case study focus on natural gas facilities in Thrace Prefecture in North Greece. Keywords: Valuable investments, energy infrastructure, socioeconomic footprint 1. INTRODUCTION that began to develop their NG distribution network during the latter half of the 1900s. As the need for a more sustainable and decarbonized future is continuously growing [1] gas has become Natural gas was brought in Greece during 1997, the fastest growing primary energy source in the therefore its distribution networks are still not fully world, due to its environmental friendly nature and developed. Currently, its National Natural Gas its multiple uses across a number of sectors. In Distribution System consists of the main natural gas Europe, the demand for natural gas is rapidly distribution system, which spreads through increasing, reaching 538 bcm in 2018. While the transmission pipelines and branches across the demand is growing so fast, the necessity for more biggest part of the country, the Trans Adriatic expanded and well-developed natural gas Pipeline (TAP), an 878 kg long pipeline, that will distribution infrastructure is rising as well. transfer natural gas to Europe from the Caspian Currently, Europe has a quite large natural gas region through Greece, Albania and Italy and also distribution infrastructure, where countries such as the under construction IGB and IGI distribution Belgium, France, Germany, Norway, Italy the pipelines, which will be natural gas interconnector Netherlands and the UK have the best-developed spreading from Bulgaria to Komotini Greece and distribution network. However, Russia and Norway from Turkish – Greek boarder to Italy, respectively. remain the main natural gas suppliers to the remain This paper deals with the sustainability assessment European countries, since combined they provide and evaluation of the feasibility of Natural Gas almost 2/3 of the natural gas supplied to the EU [2] distribution facilities project. The application of the in contrast to Portugal, Greece and Northern Ireland, proposed methodology framework is the 262 Panagiotis Zeibekis, Melina Karagkouni, Dimitrios Dimitriou, Maria Sartzetaki YRS 2021 VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS 15. -17. September SOCIOECONOMIC FOOTPRINT Portoroz, Slovenia Alexandroupolis Independent Natural Gas System 2.1. Economic Assessment Framework (INGS) project. A fundamental part of evaluating the feasibility of 2. METHODOLOGY FRAMEWORK the project is conducting an economic assessment, which aims to examine the expenses and the costs Sustainability has become a key priority for every occurring during the development of the project. operational process and has mainly three pillars: After determining the different costs and expenses, economic, environmental and social as shown in Payback Time and ROI have to be calculated. figure 1. Investments in energy infractruture have an Payback time is the time it takes a project to repay enormous impact in the three sustainability pillars its initial investments and it is an effective criterion mentioned above. More specifically, natural gas is in evaluating the economic viability of a project. responsible for various environmental Therefore, the smaller the payback time is the implications.Even though natural gas emits quicker the project will generate a profit and will approximately half as much CO2 as coal when awake the interest of various investors. This payback combusted, there have been various considerations time method assumes that all the investment is made latetly regarding its upstream emissions due to in year zero and revenues begin immediately. variations in production basins [6]. However, natural Therefore, it may be assess whether a project can gas has been gaining penetration due to the lower generate a profit or not, but it can be a trusted pollutants of conventional energy sources and criterion of its profitability. Return on Investment improved efficiency.Ultimately, the environmental (ROI) is a performance measure used to evaluate the impacts of natural gas operations are depedant upon efficiency of one or several investments and the life cycle methane leakages rates, in combination evaluate the financial consequences of investments with other factors regarding subjective policy and actions. When a ROI turns out to be negative considerations. that suggests that the total cost overcomes the return and so the project will generate a loss instead of a Also, there are several social considerations profit and an investment should not be made. On the regarding natural gas distribution processes. The other hand, when a ROI is positive, then the project construction and operation of an FSRU station has the potential to bring significant advantages on a is a success and the bigger the ROI is, the more social level to the country. First of all, it is a source probable is for a company to produce a profit [4,5,7]. of producing energy with low carbon emissions, 2.2. Balance scorecard which effects positively not only the environment but also the the health of the people located near the Performance indicators support the management of plant. Moreover, it can contribute to increase infrastructures. As no single performance indicator significantly the employment rates of the can give a full picture regarding infrastructure community, since an adequite workforce is essential performance, each indicator presents a partial view in driving and running the sustainable FSRU station. from a specific viewpoint and is therefore not The availability of enough workers on site and enough to serve as a basis for management enough working time is essential for the plant to be decisions. A popular performance measurement operated properly. scheme suggested by [8] is the Balanced Scorecard (BSC) that developed to measure performance The feasibility evaluation of the Natural Gas metrics from financial, customer, internal processes distribution project in Alexandroupolis is assessed and growth perspectives. By combining these based on these three pillars, giving significant different perspectives, BSC helps decision makers to attention to social and economic and includes a understand the inter-relationships and trade-offs variety of different factors that are assessed in order between alternative performance dimensions, thus to obtain a conclusion on the sustainability of this leading to improved decision making and problem project. solving. The base point of BSC would be the identification of the unit’s strategic plan. This would involve the development of a goals, strategy, outputs, measures, targets and four different financial perspectives [8] 3. APPLICATION 3.1. Overview of the project This project comprises an offshore floating unit for the reception, storage and re-gasification of LNG and a system of a subsea and an onshore gas Figure 1: Sustainability Pillars transmission pipeline through which the natural gas is shipped into the Greek National Natural Gas 263 Panagiotis Zeibekis, Melina Karagkouni, Dimitrios Dimitriou, Maria Sartzetaki YRS 2021 VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS 15. -17. September SOCIOECONOMIC FOOTPRINT Portoroz, Slovenia System (NNGS) and onwards to the final Table 1: Capital expenditures for FSRU Station consumers. The Alexandroupolis INGS has also the construction capacity to connect with and transmit gas into other Cost (mil €) gas transmission systems such as TAP . This New FSRU Vessel Cost 223,3 assessment framework will be based on the Infrastructure cost 98,25 evaluation of the three economic, social and CAPEX 321,55 environmental Pillar [9,10]. Contigency (10% of CAPEX) 32,15 Owner’s Cost 48,25 3.2. Greece NG submission system TOTAL CAPEX 401,95 Natural Gas is, along with the petroleum products, the second biggest energy source used in the 21st The operational expenses were taken from the century while the natural gas market of Greece, the September 2016 press kit of the Toscana FSRU total consumption of natural gas for 2017 was 4,9 Station in Italy and Wartsila official website and billion cubic meters (bcm), while the prediction for were estimated at 38,58 mil € per year. The charge 2028 is to be increased in 6,2 bcm. Greece is fully rates (€/cubic meter) were taken from the study of dependent in natural gas imports, since consumes in Brian Soghurst [12] and the official website of a year 5.0 times more than its total reserves. It Toscana FSRU Station and are presented at Table 2. substains this consumption by importing 175,727 MMcf of natural gas per year (in 2017) [11]. The Table 2: Charge rates of FSRU Stations around the entrance points are three: the entrance point of world, (for all the above calculations we applied the dollar – euro exchange rate of 1st May 2019 (0,8929 Sidirokastro in Greece to Bulgaria borders, the $/€). entrance point of Kipoi in Greece to Turkey borders FSRU Station Charge rate (€/cm) and the Revithousa LNG land-based terminal. In Indonesia 0,034 2017, the percentages of natural gas supply Lithuania 0,026 participation for the three entrace points mentioned Chile 0,022 above are 58,7%, 12,4% and 28,9%, respectively. Italy 0,039 3.3. Τhe Alexandroupolis FSRU Station The Alexandroupolis FSRU Station has been The benefits of the operation of FSRU station were decided to be the 4th entrance point of Greek’s studied both at national and local level. Primarily, at natural gas market. The capacity of its storages will national level the station will contribute to the Greek be 170.000 cubic meters (cm), while the gas market security, becoming the fourth entrance regasification capability will be 6, 1 bcm per year. point of Greek national gas transmission system. The main component of the Project will be the With Turkey stopping the supply of natural gas offshore Floating Storage and Regasification Unit through the pipeline from Bulgaria at the end of LNG vessel, with 300m length, 32,5m breadth and 2019, 80% of natural gas will come to Greece from 26,5m height, which will be anchored at a fixed pipelines that cross Turkey. Therefore, the necessity location at a distance of 17,6km southwest from the of an extra entrance point is extremely high. Also, it port of Alexandroupolis. The Project will also will strengthen the geostrategic position of Greece consist of the permanent offshore installations and in the Balkan area. The FSRU station will feed the the subsea and onshore sections of the gas Bulgarian gas market through the IGB pipeline and transmission pipelines, which will be 24km and 4km feed gas to other countries as well, such as Serbia, long respectively. The pipeline of the FSRU station North Macedonia, Romania and Hungary. will be connected to the Greek natural gas On the other hand, at a local level the benefits would transmission system in Amfitriti Station. be 120 new job positions and supply of services The FSRU Station is a cornerstone for the from local companies, that are expected to cost at establishment of a gas hub in the area, since it will around 20 mil € per year [13]. Increase at the supply the Greek gas market, the Bulgarian and earnings of local companies from the professional Balkan gas market via the IGB pipeline and the tourism. The economic impact of the professional Italian market through the TAP pipeline. tourism is estimated to be from 187.500€ to 457.500€ per year. Last but not least, the FSRU 3.4. Analysis results station will have positive impact to the environment, The capital expenditures used for both calculations due to the lower carbon and sulphur emissions of were taken from the study conducted by Brian natural gas compared to diesel and coal. The project Songhurst for Oxford University: “The Outlook for shall also have zero effect on the sensitive ecosystem Floating Storage and Regasification Units (FSRUs)” of the Evros river delta and shall be located far from [12]. The capital expenditures for FSRU stations any Natura 2000 area. To ensure this, systematic construction are shown in Table 1. monitoring for marine, physical, biological and eco- toxicological surveys shall be carried out. 264 Panagiotis Zeibekis, Melina Karagkouni, Dimitrios Dimitriou, Maria Sartzetaki YRS 2021 VALUABLE INVESTMENTS IN ENERGY INFRASTRUCTURE TOWARDS 15. -17. September SOCIOECONOMIC FOOTPRINT Portoroz, Slovenia 4. CONCLUSIONS [3] Dimitriou D., Sartzetaki M. et al. (2020). Chapter 1: Mobility as A Service: concept, Access to energy resources, energy supply security, objectives, actors, models and development, insufficient investment in energy distribution Book title: Mobility as a Service, Transport infrastructure and sluggish progress in mitigating Trends and Economics 2018–2019, United CO2 emissions are wide recognized in the market. Nations, UNECE, ISBN: 978-92-1-117238-6, The evaluation of the efficiency of energy pp. 9-13. distribution project should analyses the impacts in [4] Dimitriou D. and Sartzetaki M., Chapter title: above issues and in literature there is a lack of Social dimension of aviation on sustainable publication regarding multidimensional assessment. development, Book title: Sustainable aviation - This paper present the outline of methodology and Greening the Flightpath, Palgrave Macmillan, assessment framework on evaluation the feasibility pp. 173-191 of an energy distribution system by adopting a [5] Dimitriou D., Sartzetaki M. (2020). Assessment system of system approach. framework to develop and manage regional intermodal transport network, Journal: Research While most economics accept the importance of in Transportation Business & Management, sustainable development and social justice within Elsevier, and between generations, they also point to https://doi.org/10.1016/j.rtbm.2020.100455 historical experience suggesting that decisions about [6] Draucker L., Bradbury J., Obeiter M., Wang W., new infrastructure should meet the social targets Stevens A. (2013). Clearing the air: Reducing toward welfare. On the investors side energy market upstream greenhouse gas emissions from U.S. prices are key driver for efficient energy facility natural gas systems, World Resources Institute. projects as in the concept most of them are capital intensive projects meaning some times the payback [7] Sartzetaki M. (2019). Appraisal of capital- intensive investments upon effectiveness to period is too long. Risk assessment assumptions and business ecosystem, Journal of Economics and energy market forecasting its crusinal in the Business, Vol.2, No.3, pp. 900-909. feasibility study and this paper provides some key highlights in this area. [8] Kaplan, R. and Norton, D. (1992) The Balanced Scorecard—Measures That Drive Performance. Harvard Business Review, 79. REFERENCES [9] Dimitriou D., Sartzetaki M., (2016). Sustainable [1] Dester, Mauricio & Andrade, Moacyr & Bajay, Development Variables to Assess Transport Sergio. (2012). New Renewable Energy Sources Infrastructure in Remote Destinations, for Electric Power Generation in Brazil. Energy International Journal of Civil and Sources, Part B: Economics, Planning, and Environmental Engineering, Vol.10(10), pp Policy. 7. 390-397. 1314-1321. 10.1080/15567249.2011.623502. [10] Sartzetaki Μ. (2019). Attribute of the Effects [2] Dediu, D., Czajkowski, M. and Janiszewska- Diversification to Eco-System from Large Kiewra, E., 2019. How did the European natural Investments in Supply Chain, Global Journal of gas market evolve in 2018?. [online] McKinsey Management and Business Research, Volume & Company. Available at: 19 Issue 5. [Accessed 25 April 2019]. February 15, 2021) [12] Songhurst, B. (2018) LNG Plant Cost Reduction 2014 –18. Available from https://www.oxfordenergy.org/wpcms/wp -content/uploads/2018/10/LNG-Plant-Cost- Reduction-2014–18-NG137.pdf. Accessed on 5 August 2020. [13] Toscana, 2016. https://www.oltoffshore.it/en/terminal/location/ 265