Design And Implementation Of The Decision System In Small Size Robot Soccer

Robots' playing soccer games has been dramatically developed as a technical competition in recent years. It is a interdisciplinary field of different subjects, such as exact-mechanism, robotics, automated control, sensor and fusion, communication, computer vision and image processing, multi agent, reasoning and decision, machine learning, etc. It is an ideal test bed of robotics and artificial intelligence.This thesis focuses on RoboCup Small Size League to design and implement the decision system of a small-size robot soccer team based on multi agent and reinforcement learning, which involves network communication, serial communication, wireless communication, etc. The decision system works out some problems in vision information transmission, world model processing and defense strategy. Detail content as follows:First, it introduces the RoboCup soccer games. A robot soccer team can be treated as a multi agent system in which the cooperation and adversarial problems are hotspots and the first part introduces multi agent system (MAS) and reinforcement learning in MAS.Second, it describes the parts that a small-size robot soccer team composed of in detail and the function of every part. Also, it discusses some solution of problems in the parts.Then, it designs and implements the decision system of a small-size robot soccer team. In the implementation, it presents and implements a double-channel real time network communication model of vision information and an efficient decision mechanism with event trigger. Meanwhile, it discusses some questions in world model and strategy.In the end, it presents a solution of resolving one-vs-one defense problem and implements a defense strategy using reinforcement learning based on Markov games. The strategy is proved effective in some experiments and games..The system described in the thesis was integrated in the small-size robot soccer team DUT Fantasia SmallSize which participated RoboCup China Open 2006. In the games, the system worked well.