299 Advances in Production Engineering & Management ISSN 1854-6250 Volume 20 | Number 3 | September 2025 | pp 299–308 Journal home: apem-journal.org https://doi.org/10.14743/apem2025.3.541 Original scientific paper Improving AGV path planning efficiency using genetic algorithms with hamming distance-based initialization Breznikar, Ž. a , Gotlih, J. a , Artič, Ž. b , Brezocnik, M. a a University of Maribor, Faculty of Mechanical Engineering, Maribor, Slovenia b University of Maribor, Faculty of Electrical Engineering and Computer Science, Maribor, Slovenia A B S T R A C T A R T I C L E I N F O This paper presents a Genetic Algorithm (GA) framework for warehouse navigation as a Travelling Salesman Problem (TSP) variant for Automated Guided Vehicles (AGVs). The warehouse layout is represented as a graph, where pick-up locations serve as terminal nodes. A distance matrix, computed via Breadth-First Search (BFS) enables efficient route evaluation. To promote diversity in the initial population, a Hamming distance-based vectorized ini- tialization strategy is employed, ensuring that the chromosomes are maximal- ly distinct. The GA balances exploration and exploitation by dynamically ad- justing the fitness function. Early generations emphasize diversity, while later ones focus on solution refinement, improving convergence and avoiding premature stagnation. Our key contribution demonstrates that the Hamming distance-based approach achieves comparable or better results with signifi- cantly fewer chromosomes. This reduces computational cost and runtime, making the method well-suited for real-time AGV routing in warehouses. The framework is adaptable to structured environments and shows strong poten- tial for integration into real-world logistics and robotics applications. Future work will focus on optimizing the algorithm and integrating it into the ROS 2 environment. The simplified version of the algorithm can be accessed at: https://github.com/IntoTheVoid-61/Warehouse-Pathfinder. Keywords: Automated guided vehicles (AGV); Warehouse routing; Genetic algorithms (GA); Combinatorial optimization, Hamming distance initialization; Robot operating system 2 (ROS 2) *Corresponding author: ziga.breznikar@student.um.si (Breznikar, Ž.) Article history: Received 6 July 2025 Revised 24 October 2025 Accepted 27 October 2025 Content from this work may be used under the terms of the Creative Commons Attribution 4.0 International Licence (CC BY 4.0). Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 1. Introduction The rapid expansion of automated manufacturing has brought increasing demand for intelligent logistics and autonomous systems in warehouses. Among these, Automated Guided Vehicles (AGVs) play a crucial role in transporting goods efficiently within complex environments. A con- siderable body of research has investigated the application of AGV systems across a wide range of environments and operational contexts, supported by diverse methodological approaches. In recent years, studies have increasingly adopted metaheuristic techniques to enhance system performance and optimization [1–10]. As warehouse layouts become more intricate and dynam- ic, determining optimal routes for AGVs becomes a critical challenge, essential for reducing de- livery times and operational costs. Genetic Algorithms (GAs), inspired by the principles of natural selection and evolution, are well suited for solving complex optimization problems such as AGV routing. By evolving popula- tion of candidate solutions through operations, inspired by biological evolution, GAs can effi- ciently explore large solution spaces and efficiently convergence towards near-optimal solutions even in the presence of multiple constraints and non-linearities. Their flexibility and robustness Breznikar, Gotlih, Artič, Brezocnik 300 Advances in Production Engineering & Management 20(3) 2025 have made them a popular choice for various combinatorial problems, including the well-known Travelling Salesman Problem (TSP) [11, 12], which closely parallels AGV routing in warehouses. Traditional or uninformed search strategies, such as brute-force or blind random search, quickly become impractical as the scale of the problem increases [13]. In high-dimensional warehouse environments, the combinatorial explosion of possible routes leads to significant computational costs and sub-optimal outcomes. This underscores the need for heuristic or me- taheuristic approaches that can guide the search process intelligently. In this study, we propose a GA-based approach for warehouse routing, tailored for a top- down two-dimensional warehouse representation. Our algorithm encodes AGV routes as se- quence of terminal locations and evolves these sequences to minimize travel distance. A central focus of this work is the effect of initial population diversity on GA performance. Specifically, we compare two initialization strategies: the conventional random initialization and a Hamming distance-based approach [14] explicitly designed to maximize population diversity, thereby en- hancing exploratory capabilities in the early stages of the GA. To evaluate the efficiency of the proposed method, extensive experiments were conducted using both initialization strategies under controlled settings. Results demonstrate that Hamming distance-based initialization performs comparably or better than standard approach, even when using significantly fewer individuals per generation. This indicates that strategic population de- sign can reduce computational costs while maintaining solution quality-a critical insight for real- time AGV routing applications in operational warehouse systems. The remainder of this paper presents the implementation of our GA-based routing algorithm, the design of the Hamming-based initialization, and detailed statistical analysis of the results obtained through multiple experimental runs. 2. Related work The Travelling Salesman Problem (TSP) is a classic benchmark in combinatorial optimization and has been extensively addressed using GAs due to their ability to explore large, complex search spaces and avoid local optima through evolutionary operators. Standard GA implementations typically start with randomly generated populations, which help avoid early convergence but still frequently lead to premature stagnation in complex or highly constrained problems. In the context of warehouse logistics, the TSP is frequently adapted to model route optimiza- tion for AGVs, where efficient sequencing of pick-up and delivery tasks is critical. Several works have proposed heuristic and metaheuristic-based solutions, including Ant Colony Optimization (ACO) and GA-based frameworks [11]. However, many approaches assume idealized conditions or rely on brute-force exploration, which becomes computationally expensive as the number of tasks and constraints grow. As a general approach, the authors in [15] proposed an enhanced GA for the TSP, in which the population is initialized using the Iterative Approximate Method, signif- icantly improving solution efficiency and convergence speed. Recent advancements have focused on enhancing diversity within GA population to improve exploration and convergence stability. For example, some authors have introduced adaptive mutations rates or hybridized GAs with local search techniques to maintain population diversity. Hamming distance, a measure of dissimilarity between binary strings, has been proposed for initializing populations that are maximally distinct [16]. While it has been applied in other do- mains, its use in structured, graph-based environments like warehouse routing remains relative- ly underexplored. To our knowledge, few studies have empirically compared Hamming distance-based initiali- zation with traditional random initialization in the context of AGV routing on warehouse graphs, employing statistically rigorous analysis to evaluate performance and convergence behaviour. This paper contributes to the existing literature by integrating a Hamming distance-based popu- lation initialization within a GA specifically designed for AGV navigation. By fostering high initial diversity, the algorithm reduces reliance on large population sizes, thus enhancing computation- al efficiency without sacrificing solution quality. Improving AGV path planning efficiency using genetic algorithms with hamming distance-based initialization Advances in Production Engineering & Management 20(3) 2025 301 3. Environmental modelling and preprocessing 3.1 Matrix-based environment encoding The warehouse environment is defined by three structural parameters: • number of aisles, • number of storage locations per aisle, • number of storage blocks. These parameters provide a compact and flexible description of the warehouse layout, which is essential for scalable simulations and graph-based modelling. To facilitate algorithmic processing, the physical warehouse is first abstracted into a two- dimensional matrix. Each cell in the matrix corresponds to a discrete warehouse location and is assigned an integer label representing its functional role. Fig. 1 illustrates the top-down view of the warehouse layout and its matrix representation. Fig. 1 Top-down view of the warehouse layout and its matrix representation Table 1 Classification of colours and numerical values from Fig. 1 Value Colour code Description 0 White Free space: Traversable 1 Black Empty storage location: Untraversable 2 Red Storage location: Untraversable 3 White Pickup location: Traversable 9 Green Start and end point: Traversable Table 1 summarises the classification scheme used to assign numerical values to different ele- ments of the warehouse. Only the cells labelled 0, 3 and 9 are traversable. Among these, all nodes labelled as 3 represent mandatory pickup locations that the AGV must visit at least once. This matrix formulation allows a direct transformation into a graph, where each traversable cell becomes a node and edges rep- resent valid moves between adjacent cells. 3.2 Graph-based preprocessing To enable efficient path planning and distance evaluation, the matrix representation of the warehouse is transformed into a weighted graph [17]. In this graph, each traversable matrix cell (labelled as 0, 3 or 9) becomes a node. An undirected edge is created between every pair of adja- Breznikar, Gotlih, Artič, Brezocnik 302 Advances in Production Engineering & Management 20(3) 2025 cent traversable nodes. By default, edges are assigned a weight of 1, representing uniform movement cost. However, in real-world applications, factors such as bottlenecks, blocked or narrow passages, and other environmental constraints may increase the traversal difficulty. Such conditions are modelled by assigning higher weights to the affected edges. Once the graph is constructed, all terminal nodes, marked with the label 3 in the matrix, are identified. These represent the pickup locations that the AGV must visit at least once. The set of terminals serves as the basis for solving a TSP-like optimization task. To quantify distances between terminals, a distance matrix is computed using a BFS algo- rithm [18]. For each terminal node, BFS calculates the shortest path (in terms of total edge weight) to every other terminal. The resulting distance matrix is a symmetric square matrix where each element 𝐷𝐷 𝑖𝑖 , 𝑗𝑗 represents the shortest traversable distance between terminals 𝑖𝑖 and 𝑗𝑗 . The distance matrix serves as a critical input to the optimization algorithm. It allows for rapid evaluation of the total route length of any candidate solution, without requiring real-time path- finding through the graph [19]. This preprocessing step thus transforms the original navigation problem into a purely combinatorial optimization task, significantly reducing computational overhead during the evolutionary search. 4. Genetic algorithm optimization 4.1 Chromosome representation and problem complexity To apply a GA to the problem of AGV routing within a warehouse, we first define how a potential solution (chromosome) is represented. Each chromosome encodes a specific sequence in which the autonomous vehicle should visit all required terminal nodes, that is, the pickup locations identified during preprocessing. A chromosome consists of a permutation of all terminal nodes, where each gene represents a single terminal node, and the order of genes determines the traversal path of the AGV. This is represented in Fig. 2, which illustrates 5 different potential solutions to the routing problem, given 8 pickup locations. Fig. 2 Example of 5 different solutions to the routing problem with 8 terminal nodes. This problem closely resembles the mentioned TSP, a well-known combinatorial optimization problem. For 𝑛𝑛 terminal nodes, there are 𝑛𝑛 ! possible permutations, making exhaustive search methods computationally infeasible, even for relatively small values of 𝑛𝑛 . As the problem scales, brute-force methods become impractical due to factorial growth in complexity. Given a set of terminals 𝑇𝑇 = { 𝑝𝑝 1 , 𝑝𝑝 2 , … , 𝑝𝑝 𝑛𝑛 }, the objective is to find a permutation π, as shown in Eq. 1, that minimizes the total travel distance. 𝜋𝜋 = � 𝐷𝐷 𝜋𝜋 ( 𝑖𝑖 ), 𝜋𝜋 ( 𝑖𝑖 + 1) 𝑛𝑛 𝑖𝑖 = 1 (1) Since brute-force methods become computationally infeasible for large 𝑛𝑛 , a GA is employed to address this challenge. GA are well-suited for permutation-based combinatorial problems and enable efficient exploration of large solution spaces by evolving a population of chromosomes through genetic operations. Improving AGV path planning efficiency using genetic algorithms with hamming distance-based initialization Advances in Production Engineering & Management 20(3) 2025 303 4.2 Hamming distance-based population initialization A critical step in ensuring the effectiveness of a GA is the initialization of a diverse population. Diversity promotes broad exploration of the solution space in early generations and helps avoid premature convergence to local optima. To systematically promote diversity, we employed a strategy based on the Hamming distance. For two sequences of equal length, the Hamming distance is defined as the number of posi- tions at which the corresponding elements differ [20]. In our context, each chromosome is a permutation of terminal nodes, and the Hamming distance between two chromosomes indicates the number of differing terminal positions in the visitation sequence. The population is initialized by iteratively generating random permutations and comparing them against the already selected chromosomes. A candidate chromosome is accepted into the population only if a minimum Hamming distance from the set of existing chromosomes exceeds a decreasing threshold, starting from the maximum value 𝑛𝑛 (the number of terminals). This en- sures that the initial population is highly diverse, therefore promoting broad exploration of the solution space. If, after a predefined number of attempts, no chromosome meets the current Hamming threshold, the threshold is reduced by one and the process repeats. This adaptive mechanism balances diversity with feasibility. A high-level implementation of the described method is shown in Fig. 3. Fig. 3 Flowchart of a high-level hamming distance-based algorithm to ensure initial population diversity. Breznikar, Gotlih, Artič, Brezocnik 304 Advances in Production Engineering & Management 20(3) 2025 This Hamming distance-based algorithm ensures diversity in the starting population, ena- bling a broader and more efficient exploration of the solution space. The computational com- plexity of the initialization process increases approximately as 𝑂𝑂 ( 𝑃𝑃 2 𝐴𝐴 𝑛𝑛 ), where 𝑃𝑃 denotes the population size, 𝑛𝑛 the number of pickup locations, and 𝐴𝐴 the number of random Hamming dis- tance attempts per chromosome. 4.3 Evolutionary dynamics and evaluation This section describes the key components of the GA, including selection mechanisms, elitism strategy, mutation operators, and the dynamic fitness evaluation. These mechanisms were de- signed to balance exploration and exploitation throughout the optimization process. A dynamic tournament selection method is used to select chromosomes for reproduction or crossover. In each tournament, a subset of 𝑘𝑘 chromosomes from the current population is ran- domly selected, and the best-performing individual is chosen. To balance exploration and exploi- tation over time, the tournament size is dynamically adjusted, as shown in Eq. 2. 𝑘𝑘 = 𝑟𝑟𝑟𝑟 𝑟𝑟𝑛𝑛𝑟𝑟 (( 𝑘𝑘 𝑚𝑚 𝑚𝑚𝑚𝑚 − 𝑘𝑘 𝑚𝑚 𝑖𝑖𝑛𝑛 ) · 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 + 𝑘𝑘 𝑚𝑚 𝑖𝑖𝑛𝑛 ) (2) where 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 is defined in Eq. 3. 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 = 𝑐𝑐 𝑟𝑟𝑟𝑟𝑟𝑟𝑝𝑝 𝑛𝑛𝑐𝑐 _𝑝𝑝𝑝𝑝 𝑛𝑛 𝑝𝑝 𝑟𝑟 𝑔𝑔𝑐𝑐𝑖𝑖𝑟𝑟 𝑛𝑛 𝑚𝑚 𝑔𝑔𝑚𝑚 _𝑝𝑝𝑝𝑝 𝑛𝑛 𝑝𝑝 𝑟𝑟 𝑔𝑔𝑐𝑐𝑖𝑖𝑟𝑟 𝑛𝑛 (3) This allows smaller tournament sizes early on (encouraging diversity) and larger tourna- ments later (favouring selection pressure). Following the selection, each chromosome is either reproduced directly into the next generation with probability 𝑝𝑝 , or undergoes crossover with a different chromosome with probability 1 − 𝑝𝑝 . To preserve high quality solutions, elitism is applied by copying the top 𝑖𝑖 chromosomes unal- tered into the next generation. The number of elite chromosomes increases quadratically with the progress of generations as shown in Eq. 4. 𝑖𝑖 = 𝑟𝑟 𝑟𝑟 𝑟𝑟 𝑛𝑛 𝑟𝑟 (( 𝑖𝑖 𝑚𝑚 𝑚𝑚𝑚𝑚 − 𝑖𝑖 𝑚𝑚 𝑖𝑖𝑛𝑛 ) · 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 2 + 𝑖𝑖 𝑚𝑚 𝑖𝑖𝑛𝑛 ) (4) This strategy ensures that more optimal solutions are retained as the algorithm converges. Two types of mutation operators are employed to maintain genetic diversity: • Single-gene swap: Two genes (positions) in the chromosome are randomly swapped [21]. • 2-opt-swap: A sub-sequence of genes is reversed, a common local optimization technique in TSP-like problems [22]. The mutation probability is dynamic and is sampled from a uniform distribution U(a,b), where bounds a and b evolve as generations progress, as shown in Eq. 5 and Eq. 6. 𝑔𝑔 = ( 𝑔𝑔 𝑒𝑒𝑛𝑛 𝑒𝑒 − 𝑔𝑔 𝑠𝑠𝑠𝑠 𝑚𝑚 𝑠𝑠𝑠𝑠 ) · 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 + 𝑔𝑔 𝑠𝑠𝑠𝑠 𝑚𝑚 𝑠𝑠𝑠𝑠 , where 𝑔𝑔 𝑠𝑠 𝑠𝑠 𝑚𝑚𝑠𝑠𝑠𝑠 > 𝑔𝑔 𝑒𝑒𝑛𝑛 𝑒𝑒 (5) 𝑏𝑏 = ( 𝑏𝑏 𝑒𝑒𝑛𝑛 𝑒𝑒 − 𝑏𝑏 𝑠𝑠𝑠𝑠 𝑚𝑚 𝑠𝑠𝑠𝑠 ) · 𝑝𝑝 𝑟𝑟 𝑟𝑟 𝑝𝑝𝑟𝑟 𝑝𝑝 𝑝𝑝 𝑝𝑝 + 𝑏𝑏 𝑠𝑠𝑠𝑠 𝑚𝑚 𝑠𝑠𝑠𝑠 , where 𝑏𝑏 𝑠𝑠𝑠𝑠 𝑚𝑚 𝑠𝑠𝑠𝑠 > 𝑏𝑏 𝑒𝑒𝑛𝑛 𝑒𝑒 (6) This allows higher variability early on (favouring exploration), which gradually decreases as the algorithm approaches convergence. The fitness function transitions from the exploration phase to the exploitation phase to bal- ance global search and solution refinement. During the exploration phase, the fitness incorpo- rates both solution quality and population diversity, as shown in Eq. 7. 𝑓𝑓 = 𝛼𝛼 · 𝑛𝑛 𝑟𝑟 𝑟𝑟 𝑚𝑚𝑔𝑔 𝑛𝑛 𝑖𝑖 𝑛𝑛 𝑝𝑝 𝑟𝑟 _𝑝𝑝 𝑔𝑔𝑐𝑐 ℎ_𝑛𝑛 𝑝𝑝 𝑛𝑛 𝑝𝑝𝑐𝑐 ℎ − (1 − 𝛼𝛼 ) · 𝑛𝑛 𝑟𝑟 𝑟𝑟 𝑚𝑚𝑔𝑔 𝑛𝑛𝑖𝑖 𝑛𝑛 𝑝𝑝 𝑟𝑟 _ ℎ 𝑔𝑔 𝑚𝑚𝑚𝑚𝑖𝑖 𝑛𝑛 𝑝𝑝 _𝑟𝑟𝑖𝑖𝑝𝑝 𝑐𝑐𝑔𝑔𝑛𝑛 𝑐𝑐𝑝𝑝 (7) Normalization is necessary to account for differing numerical scales. Path lengths are normal- ized relative to population statistics, while Hamming distances are normalized relative to the maximum possible value, corresponding to the chromosome length. The coefficient 𝛼𝛼 balances the two terms and is defined by a sigmoid function [23] as shown in Eq. 8. 𝛼𝛼 = 1 1 + 𝑝𝑝 − 𝑚𝑚 (8) Improving AGV path planning efficiency using genetic algorithms with hamming distance-based initialization Advances in Production Engineering & Management 20(3) 2025 305 where 𝑚𝑚 is defined as shown in Eq. 9. 𝑚𝑚 = 20 · 𝑝𝑝𝑝𝑝 𝑛𝑛 𝑝𝑝 𝑟𝑟 𝑔𝑔𝑐𝑐𝑖𝑖𝑟𝑟 𝑛𝑛 _𝑐𝑐 𝑟𝑟 𝑟𝑟 𝑛𝑛 𝑐𝑐 𝑝𝑝 𝑟𝑟 𝑝𝑝 𝑚𝑚 𝑝𝑝𝑛𝑛 𝑟𝑟 𝑟𝑟 𝑔𝑔𝑐𝑐𝑖𝑖𝑟𝑟 𝑛𝑛 _ 𝑝𝑝 𝑛𝑛𝑟𝑟 − 10 (9) In Eq. 9, generation_counter is an integer value representing the current generation number, while exploration_end is the predefined generation at which exploitation begins. During the ex- ploitation phase, fitness is based solely on the provided path length of a chromosome. This phase focuses entirely on improving the solution quality. A high-level schematic of the proposed GA framework is illustrated in Fig. 4. Fig. 4 Algorithmic structure of the GA-based navigation framework 5. Results and discussion This section presents a comprehensive evaluation of the impact of Hamming distance-based population initialization (H) compared to random initialization (NH) in a GA designed for opti- mizing warehouse routing. The analysis spans four key performance metrics: • runtime duration, • initial population diversity, • convergence speed, • final solution quality (best tour length). Breznikar, Gotlih, Artič, Brezocnik 306 Advances in Production Engineering & Management 20(3) 2025 The experiments were performed over 20 independent runs per setting, using identical seeds and warehouse environments to ensure consistency. 5.1 Equal population size comparison In the first series of experiments, both methods were executed under identical setting with 30 individuals per generation. Each configuration was repeated 20 times using the same seeds and warehouse environment. To validate the use of parametric tests, Shapiro-Wilk and D’Agostino- Pearson normality tests were conducted for all metrics [24]. Results indicated normal distribu- tion across all groups. Table 2 summarizes the results. Table 2 Statistical analysis of GA metrics: Hamming vs. Non-Hamming Metric Hamming (Mean ± SD) Non-Hamming (Mean ± SD) 𝑝𝑝 -value Significance Runtime (s) 45.0283 ± 5.3 40.2011 ± 5.79 0.0057 significant Initial population Hamming distance 80.3567 ± 0.03 78.2347 ± 0.05 < 0.0001 significant Generations 6122.05 ± 692.09 5907.45 ± 739.28 0.2904 not significant Best tour length 400.8000 ±11.19 403.6000 ± 9.46 0.3915 not significant Key takeaways: • Population diversity: The Hamming-initialized population achieved significantly higher diversity, confirming the effectiveness. • Runtime: Hamming initialization introduced a slight computational overhead due to pair- wise distance calculations. • Final Quality & Convergence: No statistically significant improvement was observed. 5.2 Exploring population size effects Additional experiments were conducted to investigate whether Hamming distance-based initial- ization can effectively compensate for a reduced population size in GA applications. This evalua- tion was carried out on a structurally distinct warehouse layout to ensure generalizability of the findings. In this setting, we compared the performance of a GA configured with only 30 individu- als initialized using the Hamming distance strategy (H30) against a GA employing 150 individu- als initialized randomly (NH150). Despite the fivefold disparity in population size, the H30 con- figuration consistently demonstrated comparable solution quality to the NH150 setup in both final tour length and convergence behaviour. Statistical analysis of the tour length distributions supported this finding, yielding a signifi- cant 𝑝𝑝 -value [25] (p < 0.05) indicating that the performance difference favoured the H30 config- uration. These results underscore the effectiveness of diversity-promoting strategies in evolu- tionary algorithms. Instead of compensating for insufficient diversity by brute-force scaling of the population size, initializing the population with maximally dissimilar chromosomes enables more efficient exploration of the solution space. This provides a strong argument for adopting informed initialization techniques, especially in resource-constrained environments where com- putational efficiency is critical, such as real-time AGV routing in dynamic warehouse settings. 6. Conclusion This work explored the application of a GA to solve warehouse routing problem, a task critical to the efficiency of AGV system in logistics and manufacturing environments. Our results confirm that a GA-based approach is not only viable but effective for generating high quality routing so- lutions within reasonable computational budget. The simplified version of the algorithm can be accessed at: https://github.com/IntoTheVoid-61/Warehouse-Pathfinder. Future work of the algo- rithm will extend to its optimization and integration into the ROS2 environment. In a targeted comparison, a GA with just 30 chromosomes initialized using Hamming-based method achieve statistically comparable results in tour quality relative to that of a GA with 150 randomly initialized individuals. This outcome highlights the central contribution of our study: Improving AGV path planning efficiency using genetic algorithms with hamming distance-based initialization Advances in Production Engineering & Management 20(3) 2025 307 intelligent population seeding can significantly reduce the required population size without compromising performance. This finding has tangible implications for real-world deployment. Reducing population size lowers computational time and memory usage, making approach more suited for embedded or real-time systems commonly used in AGV applications. Accordingly, this study provides a practi- cal and scalable GA-based framework for warehouse routing, with added benefit of an initializa- tion method that enhances performance under constrained resources. Future work should focus on deploying the proposed system on real-world AGV platforms and evaluating its performance through experiments and field-testing. While the algorithm demonstrates strong results in two-dimensional navigation (based on top-down view of the warehouse), it does not yet account for three-dimensional considerations. Furthermore, as this study was conducted on an abstracted AGV model, specific characteristics of actual vehicle were not incorporated. Notably, the model assumed that AGV could transport an unlimited mass of cargo and did not require a return to the starting point for unloading. Future implementations on embedded system should therefore account for such practical constraints, including limited payload capacity and the need for cargo drop-off behaviour. References [1] Löffler, M., Boysen, N., Schneider, M. (2021). Picker routing in AGV-assisted order picking systems, INFORMS Journal on Computing, Vol. 34, No. 1, 440-462, doi: 10.1287/ijoc.2021.1060. [2] Agatz, N., Bouman, P., Schmidt, M. (2018). Optimization approaches for the traveling salesman problem with drone, Transportation Science, Vol. 52, No. 4, 965-981, doi: 10.1287/trsc.2017.0791. [3] Han, Z., Wang, D., Liu, F., Zhao, Z. (2017). Multi-AGV path planning with double-path constraints by using an improved genetic algorithm, Plos One, Vol. 12, No, 7, Article Id. e0181747, doi: 10.1371/journal.pone.0181747. [4] Simon, J., Trojanová, M., Hošovský, A., Sárosi, J. (2021), Neural network driven automated guided vehicle plat- form development for Industry 4.0 environment, Tehnički Vjesnik – Technical Gazette, Vol. 28, No. 6, 1936-1942, doi: 10.17559/TV-20200727095821. [5] Bostancı, B., Karaağaç, A. (2019). Investigating the shortest survey route in a GNSS traverse network, Tehnički Vjesnik – Technical Gazette, Vol. 26, No. 2, 355-362, doi: 10.17559/TV-20170924174221. [6] Xu, L., Wang, Y., Liu, L., Wang, J. (2016). Exact and heuristic algorithms for routing AGV on path with precedence constraints, Mathematical Problems in Engineering, Vol. 2016, No. 1, Article No. 5040513, doi: 10.1155/2016/ 5040513. [7] Wang, Y.J., Liu, X.Q., Leng, J.Y., Wang, J.J., Meng, Q.N., Zhou, M.J. (2022). Study on scheduling and path planning problems of multi-AGVs based on a heuristic algorithm in intelligent manufacturing workshop, Advances in Pro- duction Engineering & Management, Vol. 17, No. 4, 505-513, doi: 10.14743/apem2022.4.452. [8] Wang, Y.D., Lu, X.C., Song, Y.M., Feng, Y., Shen, J.R. (2022). Monte Carlo Tree Search improved Genetic Algorithm for unmanned vehicle routing problem with path flexibility, Advances in Production Engineering & Management, Vol. 17, No. 4, 425-438, doi: 10.14743/apem2022.4.446. [9] Boccia, M., Masone, A., Sterle, C., Murino, T. (2023). The parallel AGV scheduling problem with battery con- straints: A new formulation and a matheuristic approach, European Journal of Operational Research, Vol. 307, No. 2, 590-603, doi: 10.1016/j.ejor.2022.10.023. [10] Micieta, B., Durica, L., Binasova, V. (2018). New solution of abstract architecture for control and coordination decentralized systems, Tehnički Vjesnik – Technical Gazette, Vol. 25, Supplement 1, 135-143, doi: 10.17559/TV- 20160117100949 [11] Wang, Y.D., Lu, X.C., Shen, J.R. (2021). Improved Genetic Algorithm (VNS-GA) using polar coordinate classifica- tion for workload balanced multiple Traveling Salesman Problem (mTSP), Advances in Production Engineering & Management, Vol. 16, No. 2, 173-184, doi: 10.14743/apem2021.2.392. [12] Sui, J., Ding, S., Huang, X., Yu, Y., Liu, R., Xia, B., Ding, Z., Xu, L., Zhang, H., Yu, C., Bu, D. (2025). A survey on deep learning-based algorithms for the traveling salesman problem, Frontiers of Computer Science, Vol. 19, Article No. 196322, doi: 10.1007/s11704-024-40490-y. [13] Cormen, T.H., Leiserson, C.E., Rivest, R.L., Stein, C. (2009). Introduction to algorithms, Third edition, MIT Press, Cambridge, Massachusetts, USA. [14] Leung, Y., Gao, Y., Xu, Z.-B. (1997). Degree of population diversity - a perspective on premature convergence in genetic algorithms and its Markov chain analysis, IEEE Transactions on Neural Networks, Vol. 8, No. 5, 1165- 1176, doi: 10.1109/72.623217. [15] Alkafaween, E., Hassanat, A., Essa, E., Elmougy, S. (2024). An efficiency boost for genetic algorithms: Initializing the GA with the iterative approximate method for optimizing the Traveling Salesman Problem – Experimental insights, Applied Sciences, Vol. 14, No. 8, Article No. 3151, doi: 10.3390/app14083151. Breznikar, Gotlih, Artič, Brezocnik 308 Advances in Production Engineering & Management 20(3) 2025 [16] Cruz Chávez, M.A., Martínez-Oropeza, A. (2016). Feasible initial population with genetic diversity for a popula- tion-based algorithm applied to the vehicle routing problem with time windows, Mathematical Problems in En- gineering, Vol. 2016, No. 1, Article No. 3851520, doi: 10.1155/2016/3851520. [17] Diestel, R. (2017). Graph theory, 5 th edition, Springer, Berlin, Germany, doi: 10.1007/978-3-662-53622-3. [18] Goodrich, M.T., Tamassia, R., Goldwasser, M.H. (2014). Data structures & algorithms in Java, 6 th edition, Wiley, Hoboken, New Jersey, USA. [19] Deza, M.M., Deza, E. (2013). Encyclopedia of distances, 2 nd edition, Springer, Berlin, Germany, doi: 10.1007/978- 3-642-30958-8. [20] Black, P.E. (2006). Hamming distance, In: Black, P.E. (ed.), Dictionary of algorithms and data structures [online], Black, P.E. (ed.), from https://www.nist.gov/dads/HTML/HammingDistance.html, accessed May 13, 2025. [21] Sivanandam, S., Deepa, S. (2008). Genetic algorithms, In: Sivanandam, S., Deepa, S. (eds.), Introduction to genetic algorithms, Springer, Berlin, Germany, 15-37, doi: 10.1007/978-3-540-73190-0_2. [22] Uddin, F., Riaz, N., Manan, A., Mahmood, I., Song, O.-Y., Malik, A.J., Abbasi, A.A. (2023). An improvement to the 2- Opt heuristic algorithm for approximation of optimal TSP tour, Applied Sciences, Vol. 13, No. 12, Article No. 7339, doi: 10.3390/app13127339. [23] Science Direct. Sigmoid Function, from https://www.sciencedirect.com/topics/computer-science/sigmoid- function, accessed April 4, 2025. [24] Altman, D.G., Bland, J.M. (1995). Statistics notes: The normal distribution, The BMJ, Vol. 310, Article No. 298, doi: 10.1136/bmj.310.6975.298. [25] Gao, J. (2020). P-values – a chronic conundrum, BMC Medical Research Methodology, Vol. 20, Article No. 167, doi: 10.1186/s12874-020-01051-6.