Statistics Based Q-learning Algorithm for Multi-Agent System and Application in RoboCup

This paper proposes statistic learning based Q-learning algorithm for Multi-Agent System, the agent can learn other agents’ action policies through observing and counting the joint action, a concise but useful hypothesis is adopted to denote the optimal policies of other agents, the full joint probability of policies distribution guarantees the optimal action choice to the learning agent. The algorithm can also improve the learning speed because the conventional Q-learning space is cut from exponential one to linear one. The convergence of the algorithm has been proved; the successful application of this algorithm in the RoboCup shows its good learning performance.