Study Of Agent's Machine Learning Strategy In RoboCup Simulation Tournament

RoboCup (The robot World Cup Soccer Games and Conferences) is the most important Robot Soccer game in the world. The Robot Soccer has been become the research center of many artificial intelligence researchers. As the ideal testing plant of Machine-Learning and Multi-Agent Systems, it is involved in many technical domains. Many methods of machine learning can be checked out in the simulation tournament. As a standard problem of artificial intelligence and robot, RoboCup has come to more and more peoples' attention.Based the theory of Agent and machine learning, this paper mainly resolves two problems: the basic action of agent and decision-making of agent's high-level action. The research results are as follows:1. Reinforcement learning method is introduced into the kick action of player, and accelerates the ball to the aimed velocity. Back Propagation (BP) network is used here to imitate low action of kick, and then give the relationship of kick-power and ball-velocity.2. Application of BP algorithm achieves the intercept action of player, according to which, we give a relative passing method. And then, the paper studies on shooting problem by using the RBF algorithm. By combining of Q learning and BP network, we bring a new method of shooting, and get a good effect.