Multi-agent coordination method based on Markov game and application to robot soccer

A layered multi-agent coordination method based on Markov games is presented. According to the relationship of competition and cooperation among the multiple agents, this method adopts the zero-sum Markov game in high layer to compete with the opponent and adopts the team Markov game in low layer to accomplish cooperation in the team. With the application and experiment in Robot Soccer, it is shown that this method is better than the traditional multi-agent learning method.