Multi-Agent Reinforcement Learning Method for Disassembly Sequential Task Optimization Based on Human–Robot Collaborative Disassembly in Electric Vehicle Battery Recycling

Abstract With the wide application of new Electric Vehicle (EV) batteries in various industrial fields, it is important to establish a systematic intelligent battery recycling system that can be used to find out the resource wastes and environmental impacts of the retired EV battery. By combining the uncertain and dynamic disassembly and echelon utilization of EV battery recycling in the remanufacturing fields, human–robot collaboration (HRC) disassembly method can be used to solve huge challenges about the efficiency of retired EV battery recycling. In order to find out the disassembly task planning based on HRC disassembly process for retired EV battery recycling, a dynamic disassembly sequential task optimization method algorithm is proposed by Multi-Agent Reinforcement Learning (MARL). Furthermore, it is necessary to disassemble the retired EV battery disassembly trajectory based on the HRC disassembly task in 2D planar, which can be used to acquire the optimal disassembly paths in the same disassembly planar by combining the Q-learning algorithm. The disassembly task sequence can be completed through standard trajectory matching. Finally, the feasibility of the proposed method is verified by disassembly operations for a specific battery module case.