A Chaotic Dynamic Potential Field‐Guided Particle Swarm Optimization Algorithm for Multi‐ <scp>AGV</scp> Path Planning in Complex Warehouse Environments

ABSTRACT In complex warehouse environments, efficient and conflict‐free collaboration among multiple automated guided vehicle (AGV) systems is key to improving the efficiency of logistics automation. However, hybrid algorithms based on artificial potential fields (APF) and particle swarm optimization (PSO) typically treat environmental perception and heuristic search as independent modules when addressing multi‐AGV problems, lacking coupling at the mechanism level. To address this issue, this paper proposes a chaotic dynamic potential field‐guided PSO (CDPF‐PSO) algorithm, which embeds environmental perception into the search process to achieve coordinated global and local optimization. First, an initial population is constructed by combining potential field information with the tent chaotic map to balance population diversity and environmental adaptability. Second, the gradient information of the APF is translated into a dynamic potential field factor and incorporated into the velocity update formula of PSO, endowing the algorithm with both global exploration and local obstacle avoidance capabilities. Finally, a dual‐layer collision avoidance strategy based on safety distance and priority is designed, and B‐spline curve path smoothing is introduced to strictly ensure the kinematic constraints of the AGV and eliminate dynamic collision risks in both the spatial and temporal domains. In the experimental section, parameter sensitivity experiments were conducted to determine the optimal configuration of key parameters and validate the algorithm's parameter stability. Comparative experimental results demonstrate that this method possesses significant comprehensive advantages in terms of path quality, convergence efficiency, and safety reliability: in a standard 20 × 20 m environment, the proposed method achieved an average path length optimization rate of 11.61%; in a complex 40 × 40 m environment, its overall performance still maintained a 9.70% improvement; and in a 100 × 100 m environment, the algorithm further demonstrated its adaptability and stability for large‐scale AGV fleets.