Multi-agent reinforcement learning method for Markov games: an approach based on the estimation of the environmental model

摘要

In this article, we propose a multi-agent reinforcement learning method for Markov games. In our multi-agent reinforcement learning method, each agent infers the environmental model which consists of the other agents' policies and the state transition function, and estimates the future states by using the inferred environmental model. Each agent conducts its reinforcement learning based on the estimated future states. In order to evaluate our multi-agent reinforcement learning method, we employ the variant of the pursuit problem as a task. Through experiments, we demonstrate that our multi-agent reinforcement learning method is effective.