Intelligent path control of autonomous AGVs in flexible manufacturing systems: a reinforcement learning approach

Automated Guided Vehicles (AGVs) play a crucial role in material handling and transportation in automated manufacturing, enhancing efficiency and accuracy. However, traditional AGVs follow predefined paths, limiting flexibility in dynamic industrial environments with unpredictable material movement and human activity. This paper introduces a hybrid navigation strategy for autonomous AGVs using a dynamic programming approach integrated with artificial intelligence techniques to enhance autonomy and flexibility. The approach utilizes a Markov Decision Process (MDP) to determine optimal, collision-free paths by considering predefined path costs based on distance and dynamically updating travel times. The AGV selects the best route using this updated network. Once an optimal path is identified, the Timed Elastic Band (TEB) planner guides the AGV to intermediate goal points. For obstacle avoidance, the AGV classifies detected objects as static or dynamic using LiDAR scan data, enabling real-time reactive navigation. Extensive simulations and hardware experiments validate the effectiveness of the proposed method in single AGV path planning. Additionally, in multi-robot material handling, updating the network improves the coordination of multiple AGVs, preventing collisions and optimizing shop floor efficiency. The results demonstrate that this dynamic programming-based approach significantly enhances autonomous AGV navigation in industrial settings.