Applying Multi-agent Reinforcement Learning to Autonomous Distributed Real-time Scheduling

Abstract

Autonomous Distributed Manufacturing Systems (ADMS) have been proposed to realize flexible control structures of manufacturing systems. In the previous researches, a real-time scheduling method based on utility values has been proposed and appliedto the ADMS. In the proposed method, all the job agents and the resource agents evaluate the utility values for the cases where the agent selects the individual candidate agents for the next machining operations. Multi-agent reinforcement learning is newly proposed and implemented to the job agents and resource agents, in order to improve their coordination processes. In the reinforcement learning method, an agent must be able to sense the status of the environment to some extent and must be able to takeactions that affect the status. The agent also must have a goal or goals relating to the status of the environment. The status, the action and the reward are defined for the individual job agents and the resource agents to evaluate the suitable utility values based on the status of the ADMS.