Conflict-Free Path Planning for Automated Guided Vehicles in U-Shaped Automated Container Terminals Based on Behavior Cloning and Reinforcement Learning

To address the path planning problem of Automated Guided Vehicle (AGV) in U-shaped Automated Container Terminal (U-shaped ACT), this study proposes a fusion of behavioral cloning technique, A* algorithm, and Advantage Actor-Critic (A2C) path planning method (BA-A2C). The method addresses the limitations of traditional heuristic algorithms, such as low computational efficiency, high conflict rates, and the inefficiency of reinforcement learning during the initial training phase. By incorporating the A* algorithm to collect expert experience and combining it with behavior cloning, the proposed method provides high-quality initial strategy guidance for reinforcement learning, significantly improving network training efficiency. Furthermore, the dynamic optimization capability of the A2C algorithm enables real-time path planning and conflict resolution for multiple AGV in the U-shaped ACT environment. Experimental results demonstrate that the BA-A2C algorithm outperforms other methods in terms of training efficiency, computational efficiency, and conflict handling capabilities, validating its applicability and reliability in addressing multi-AGV path planning problems in U-shaped ACT scenarios.