Research On Cooperation And Planning Of MAS And It's Application In RoboCup

Recently,the research on multi-agent system and distributed artificial intelligence becomes a main subfield of artificial intelligence.MAS and DAI are the major background of RoboCup, which task is to promote the research and education of distributed artificial intelligence and intelligence robotics.This research need agent,which one of the team, can achieve variable action and accomplish the team's aim in a real time,dynamic and unpredict enviroment.This dissertation takes the RoboCup simulation league as testbed,takes the multi-agent system as target of study,mainly has carried on the following work:Fisrtly,after studying the existing cooperation strategies in MAS,this paper chooses cooperation in MAS as task,apply the cooperation strategies based on the static and dynamic lineup respectively from static and dynamic aspects which extremely effective increased team's overall capability,and use the strategy based on the cooperative desire matrix which highly improved the attacking competence.Secondly,Q learning isn't adapt to continuous state space and action space,and with the increasing of the state space,the Q value memory as well as inquiry becomes difficult,there will lead to dimensional disaster.Otherwise,traditional reinforcement learning hasn't the ability of generalization.All of these disadvantages restrict its application in the complex and variable environment.In order to solve those problems,we use the fuzzy Q learning and CA-FCMAC Q learning algorithm to do it,and used in the RoboCup for validating it.The results of experiment illuminates the modified algorithm effectively improve the individual skill.Finaly,because of few informations are provided by outer environment in MAS,and the reinforcement learning usually is inefficient.We use the algorithm based on the experience knowledge to optimize it,so as to increasing the speed of the agent learning,and combined it with the conception of intrinsic motivation from psychology,put forward the intrinsic reinforcement learning based on the exerience konwledge.At last,this algorithm is used for cooperating skill in MAS,the results of experiment show that the modified algorithm has faster speed to converge and better performance.