Reinforcement Learning for Soccer Multi-agents System

摘要

Recently the reinforcement learning method is actively used in multi-agent systems. Because of this method played a significant role by handling the inherent complexity of such systems. Robotic soccer is a multi-agent system in which agents play in real-time, dynamic, complex and unknown environment. Since the main purpose of a soccer game is to score goals, it is important for a robotic soccer agent to have a clear policy about whether it should attempt to score in a given situation. Therefore we use reinforcement learning for optimizing policy. In the proposed method, the state spaces include two important parameters for shooting toward the goal; the distance between the ball and the goalkeeper and the probability which is obtained from the research of the UvA team. Of course, we select these parameters for effective features of scoring. Because they are more effective learning algorithm in real-time simulated soccer agent. Experimental results have shown that policy achieved from reinforcement learning lead to more effective shoots toward the goal in simulated soccer agent.