Application Of Machine Learning Method In Soccer Robot System

Soccer Robot is one of the research hotspots in the area of Artificial Intelligence. It brings up a standard task to stimulate the interest of the public for robotics and AI technology. And now it has been used as a research platform in parallel with a usage for educational purposes, and to advance the overall technological level of society.In Soccer Robot System, all the robots cooperate to achieve the joint task. In other words, every robot receives environment information such as the position and velocity of ball and other robots from its sensors, then decide the action for next step according to the situation. The ability of decision-making and also the algorithm of path planning mainly determine the performance of the system. So machine learning is the main search direction of multi-agent systems. The specific application and experimental results in Soccer Robot System prove the feasibility and validity of the learning technique in complex adversarial systems. Robots' intelligence level can be unproved through the use of this intelligent learning method.In challenging environment, it is important for an agent to track and keep watch on others, and to infer their high-level objectives and intentions. It is desired not only to learn every agent's intention, but also to infer the group's intention. This paper proposes a new method of modeling opponents in challenging environment, based on the model of BDI(Belief, Desire, Intention) and combined with transposition principle and also probabilistic belief logic. The method can also be served for a robot to cooperate with his teammate for the smartest agents will learn to be team players. On the other hand, the process of modeling the opponent is also the process of learning from opponent. Then a robot can enrich its tactic and improve its performance. The method has been applied to the soccer robot system and the experiment results show that this method can well enhance the system's performance and intelligence degree.The Soccer Robot System is a dynamic environment with multiple obstacles. It is a problem of high complexity to perform path planning in such environments. The traditional methods are not efficient in such complex environments.In this paper, a self-learning method of robot navigation is proposed based on the reinforcement learning method and artificial potential field method. By establishing the direct mapping from the state set to the action set, the planning efficiency is improved. The learning converges to the optimal control strategy. The simulation result shows that the control strategy is optimized and the control performance is improved.