Robocup Simulation Team's Realization

RoboCup Soccer Game is a standard problem in multi-agent dynamicinpredictable environment. It is an ideal experimental platform for theresearch on multi-agent system and distributed artificial intelligence. Itinvolves many subjects to build a RoboCup simulated team, including agentarchitecture, synchronized communication, sensory information fusion,realtime reasoning, machine learning, team cooperation and counterwork,player modeling, etc.This thesis focuses on the research and realization of a simulated team.The main work and contribution are as follows:A hybrid agent architecture is designed which breaks the system intoseveral modules such as sensor, actor, world model, action model, intelligentdecision. In ordor to intercourse with server effectively, an event-drivensynchronized communication mechanism is built.World model which stores various sensor information and action predictionis designed using UML's class diagram. Partical filter is adopted to improvethe accuracy of player's self-location. Some functions which determine thesituation on soccer field are derived from basic information.Action model of four levels is built. Geometric analytical calculationmethod is adopted to realize various personal technical behaviors, such asposition, intercept, pass, dribble, shoot, defend goal, etc.Intelligent decision is divided into two parts, namely team coorperationand personal decision. In the aspect of team cooperation, formations and rolesare used to realize the distribution of team form, tactical system is designedto realize partial cooperation, communications among teammates are used toshare information and to transfer tactical intentions. In the aspect ofpersonal decision, decision tree is used to determine current technical actionfor individual player according to the situation on soccer field andcooperation suggestion. As far as the crucial technical action of ball-holderis concerned, the shoot success rate is assessed, also passing and dribblingare divided into several types for choosing. Object-oriented observationsystem is designed to enable the players to observe interested objects effectively.The above methods are proved effective by simulated experiments.