The Design And Realization Of Multi-Agent System Based On Robocup

Abstract The research on Multi－Agent System (MAS) and distributed artificialintelligence has become the focus of computer science and artificialintelligence．MAS consists of many agents． It mainly finds out the solution forthe real-time process of Information and harmony problems under complex anddynamic environments．RoboCup(The Robot World Cup Soccer Games andConferences) simulation game system is a good sample of the research ofartificial Intelligence and robot, and it is a good research platform for theresearch of MAS. The soccer server provides a fully distributed real－timemulti-agent environment with noise, collaboration and adversarial, in whichteammates have no cooperate to achieve their common goal of winning thegame, The simulation mode1s many real－world complexities such as noise inobject movement, noisy sensors and actuators, limited physical abilities andrestricted communication. The main contributions of this thesis include a multi-threaded three layeragent architecture, developing seven modules composing agent architecture byusing C＋＋．The accurate method of one line and one flag is used to completeagent localization quickly. A flexible agent-environment synchronizationscheme is constructed. Based on layered learning, actions set is developedaccording to intelligent levels from low to high by using artificial intelligencemethod. The application of Back Propagation(BP) algorithm, fuzzy neuralnetwork and reinforcement learning achieve the low actions (dash, kick) andhigh actions (freezing ball, intercept). As such the research work also includesthe implementation of scoring policy for simulated soccer agent. The statisticmethod and BP algorithm is used to three shooting conditions : a forward - II -摘 要without goalkeeper, a forward and a goalkeeper, two forwards and a goalkeeper. Finally, the thesis also contains a brief introduction to RoboCupsimulation environment in general. At the same time it provides a solidframework,which can serve as a basis for future research in the field ofmulti-agent system and RoboCup game...