A Multi-Agent Deep Reinforcement Learning Approach for Multiple AGVs Scheduling in Automated Container Terminals

The equipment scheduling method of automated container terminals is gradually evolving towards intelligence and automation. In this paper, a novel multi-agent deep reinforcement learning algorithm called Coupling Deep Q-network (C-DQN) is proposed to address the automated guided vehicles (AGV) scheduling problem. Specifically, the sparse reward problem of the proposed algorithm model is solved by a mask. The convergence of the model is accelerated by changing the greedy rate with a fixed step size. The shape of samples in the experience replay buffer is also improved through the methods of the centralized training and decentralized execution. Furthermore, the scheduling conflict problem caused by the cooperative game is reduced, and the overall efficiency of the scheduling task is also improved. Finally, the scenario of multi-AGV scheduling is designed to verify the effectiveness and efficiency of the proposed C-DQN algorithm.