Research of RoboCup Pass Strategy Based on Improved Q-Learning

Abstract

As the ideal experimental platform of multi-agent system,RoboCup(Robot World Cup) has become the research center of artificial intelligence.Traditional Q-learning dispersed sequential state and action simply on resolving the problem about pass strategy in RoboCup environment.Puts forward a method that neural network is applied to Q-learning,system would output sequential state-action by learning Q-value based on partial state-action and improve generalization ability effectively.The method based on improved Q-learning is proposed to optimize the pass strategy and test the algorithm in RoboCup environment.The experiment shows that improved Q-learning can improve pass efficiency effectively in RoboCup environment.