Multi-Objective Reinforcement Learning Algorithm Updating Actor by Using Maximum TD-error Extracted from Some Critics

Abstract

Multi-agent system (MAS) is constructed by many autonomous agents. Conflicts occur in MAS because of complex interactions among many agents. An agent needs to carry out a task and to avoid conflicts at same time. That is, each agent has to achieve the contradicting purposes. Therefore, this paper proposes a new approach by using multi-objective reinforcement learning as decision making system of an agents. We investigate the efficiency of the proposed approach through a simulation experiment that two agents pass each other in the narrow path.