The Research On The Improvement Efifciency Of Robocup In Reinforcement Learning Algorithm

At present,the robot soccer game has become one of hotspots in ArtificialIntelligence.It contains a lot of knowledge at many regions,such as artificialintelligence,intelligence control,real time vision and so on.Robot soccer itself is atypical intelligence system. At the same time it offers a appropriate platform forintelligent system both in theory research and practice testing.As all we known, thehardware requirements of robot soccer game are relatively high, which has greatlyhampered the development of robot soccer game. In order to test various intelligentalgorithms whether can be applied well in controlling of soccer robots, it is nessary todevelop soccer robot simulation platform. The simulation platform provides astandard and promotes the multi-agent systems research on theories and algorithms inreal-time confrontational environment.The robot soccer simulation game has becomeone of the major research field in the intelligent robots.The contents of this paper is asfollows:Firstly, the fundamental job to establish the robot soccer team is to do a deepresearch on the robot soccer game simulation platform model and its application.Secondly,this paper will analysis Microsoft Robotics Studio MSRS 5vs5 androbot car correlation technique and propose the specific use method of this platform,match rules and robot car dynamics knowledge in detail.Thirdly, I will probe into analyzing and arranging the reinforcement learningtheory structure and latest progress comprehensively. I will put the current applicationof Swarm reinforcement learning algorithm into the shooting,dribbling andinterception of robot soccer simulation game in order to achieve the purpose ofconnecting artificial experience and online learning.Finally, the focus would be put on applying Swarm learning algorithm into robotsoccer game simulation platform MSRS 5VS5 for experiment and analysis.After the comparison of Sarsa algorithm in scoring goals ability, interceptionability and learning ability, the result verify the superiority of the Swarm algorithm.