Using Multi-Agent Deep Reinforcement Learning For Flexible Job Shop Scheduling Problems

Abstract

The flexibilization and increase of new production concepts such as matrix manufacturing with the help of autonomous logistics robots (AGVs) pose new challenges to production scheduling. To solve these flexible job shop scheduling problems (FJSSP) for arbitrary production arrangements, a concept for a multi-agent system based on Deep Reinforcement Learning (MARL) is proposed. The focus is on speed and quality of scheduling, easy creation of new manufacturing setups and extensibility to other scheduling problems such as logistics. An algorithm to solve these problems is given and evaluated on an exemplary job shop. Future research questions and extensions are then discussed.