The Construction Of MAS In RoboCup Simulation

The RohoCup Simulation, which aims at fostering the research on the decision and cooperation in Multi-Agent System in a real-time and adversarial environment by providing a standard problem, has become one of the hotspots in Artificial Intelligence, Robotic, etc.This thesis focuses on the design and implement of Multi-Agent System in RoboCup (the Robot World Cup). After presenting the concept of Multi-Agent System and the background of RoboCup, the aim of this thesis, how to found a RoboCup Simulation Compition team named ZJUBase, is introduced in detail. Besides much improvement based on the UvA Trilearn, our main contributes, which have been verified in WorldCup2003 and ChinaCup2003 compitions, are described as follows.Apply the hybrid agent architecture in ZJUBase. The architecture consists of modeling, communication, actions, domain knowledge, deliberative reasoning and reactive module. Especially the research on the reactive module makes the agent a quick reaction, which ensure that the world model get the latest information and have more suitable actions. It can also adjust the parameters according with the state of network flow and the operating speed of the computer in competition.Accomplished the assignment and recognition of heterogeneous agent by an online coach. On the research of all the parameters about the heterogeneous agent, some essential guidelines are abstracted and the framework of assignment is set up with these guidelines. Meanwhile, the arithmetic on how to recognize the heterogeneous agent online has a probability of success above 98.51%.Refined the typical actions with heterogeneous agents. The interception, used by different player states separately, includes two arithmetical solutions: the way faster for the player with ball and more accurate for the player without ball, after some comparisons. The pass selection, the ball velocity in essence, which is the foundation of more cooperation, is searched all over the feasible solution space with the faster interception, from which thereinforcement learning method is applied to select the best soiution. Due to a step forward superiority over the player in adversarial environment, a shooter is developed, considering the bal1 random also, with the capability of more adaptable and more precise.