Collaborative Planning In Simulation Soccer Robots

At this stage, constructing intelligent agents which can produce intelligent behavior can be treated as main aim of Artificial Intelligence. Many decision-making algorithms can make good behaviors for agents which are approximated to what human do. Markov Decision Precess(MDP) provides basic model for agents’ decision-making problem in uncertain environment.RoboCup was originally designed to promote the development of research and technology of Artificial Intelligence, Robotics and other related areas. Simulation2D League was found as the earliest league of RoboCup, and now it is one of the projects which focus on agents’decision-making. In this paper, our work is to process a large scale of multi-agent planning problem, with which we use Markov decision processes to describe the uncertainty of environment.In this paper, Soccer Simulation2D is used as experiment platform, and MDP is used to describe and handle multi-agent cooperative decision-making problem in large-scale environment of uncertainty. The work mentioned in the paper contains three parts as following:●We realize and improve Trainer and rcsslogplayer, which can reappear scenes in Simulation2D competition. They make it possible to test a particular scene repeatedly. We use winning ratio to decide whether a new method is effective or not in the past, and now using random testing repeated from a particular scene is more persuasive.●We introduce a hierarchical decomposition for MDP, and combine it with inverse computation in WrightEagle, which has been tested in goalkeeper’s decision-making. By improving goalkeeper’s position, it affects opponents’ pass behavior, and it reduces the menace of opponents’attack.●We propose a method called MAXQ-MOP to solve the problem in multi-agent collaboration. MAXQ-MOP uses MAXQ-OP as basic framework, and by introducing the concept of belief pool, it has a better performance than traditional methods in man wall problem and pass between teammates.In this paper, all the work is achieved on WrightEagle.