Cooperative Control of Multiple AGVs Based on Multi-Agent Reinforcement Learning

With the widespread application of Automated Guided Vehicles (AGVs) in manufacturing systems, warehouses, and logistics fields, achieving multi-AGV collaborative control has become a key issue for improving logistics efficiency and flexibility. This study proposed a reinforcement learning-based method for multi-AGV collaborative control, aiming to achieve coordinated motion and task coordination among multiple AGVs. The modified algorithm integrates the Gated Recurrent Unit (GRU) module and the Dual Experience Replay Buffer mechanism into the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) framework. Additionally, the reward function was designed based on the artificial potential field method. Through learning and policy optimization, AGVs can make intelligent decisions based on environmental states and task objectives, enabling obstacle avoidance, target tracking and surrounding, and formation maintenance. Experimental verification conducted in a simulation environment demonstrates that the proposed algorithm achieves faster convergence speed and higher task completion rate compared to the MATD3 and MADDPG algorithms, effectively enhancing the collaborative control performance of the multi-AGV system.