The Study And Application Of Agent's Individual Learning And Multi-Agent Collaboration

Multi-Agents and distributed artificial intelligence have got more and more interests in the field of artificial intelligence (AI) in recent years. Multi-Agents system is to solve the problem of intelligent action among all members of a group, which meeting the need of handling real world complexities.Robot World Cup (RoboCup) is an attempt to foster AI and intelligent robotics research by providing a standard problem so that various theories algorithms and architectures can be evaluated, it includs the design principles of autonomous agents, Multi-Agent collaboration, strategy acquisition, realtime reasoning, robotics, and sensor fusion and many more.There are two problems in building a robot soccer team: how to get individual action and how to cooperate all the players.Individual action is action of one player including shooting, intercepting ball and passing ball, which is completed according to some commands that server provides. This thesis studies the problem by using two algorithms: BP algorithm and RBF algorithmMulti-Agent collaboration is to solve the problem of how to organize all the players to win a game. Coordination graph is a new approach for the Multi-Agent collaboration, which decomposes full cooperation problem into many sub-problems in order to decrease the action space of Agent. However, because coordination graph needs discrete state variant ,it can't be applied in continuous state space such as RoboCup which communication condition is limited. To overcome the shortcoming of cooperation graph, a Role based Cooperation Graph(RCG) is presented and the variable elimination algorithm to compute combined action in cooperation graph is improved. The RCG is used by players for selecting action in forbidden field of the opponent team.