The Application Of Multi-agent In RoboCup

Recently, the research on Multi-agent System (MAS) becomes a main subfield of Artificial Intelligence. In complex, real-time and unpredictable system, the agents need to act effectively both autonomously and as part of a team, to cooperate with the teammates and defense again the opponents to achieve the final goal of the whole team.RoboCup (Robot World Cup) is a typical Multi-agent System. It is a domain that fits all the characteristics of MAS. The robotic soccer is enjoyable and exciting too. It can demonstrate the research achievement of MAS. Because of the characteristics of robotic soccer, the International Joint Committee of Artificial Intelligence chooses robotic soccer as a standard problem of MAS. In this thesis, we use the RoboCup Soccer Simulation League 2D environment as the test bed of our research.First, the knowledge about the multi-agent and RoboCup is introduced simply. Then, the RoboCup Soccer Simulation League 2D environment is described. Last, some typical model of soccer teams is introduced, and the implementation process and method of Oryx is given too.The oryx agent is designed as a three layer agent. It includes communication layer, action layer and high strategy layer. The lower layer should serve for the higher layer and the higher layer can invoke the lower layer. The layer architecture of agent makes the whole program more clearly. The whole system is designed in several modules by object oriented method. It's developed in C++ under Linux platform. The thread technology accord with POSIX specification is adopted in this agent. The threads communicate each other by mutex.The agent's individual intelligence, such as intercepting, passing and shooting, is implemented through machine learning. The main algorithm used in agent is back-propagation neural network learning algorithm. Through this algorithm, the agent achieves an individual ability. Finally, a typical cooperation strategy SBSP, which means Situation Based Strategic Positioning, is used as the high strategy layer strategy method in oryx agent according the field knowledge of soccer. The agent takes account of noise and real-time environment. It has a reasonable architecture andintelligence, and it can cooperate with other agents under the same union strategy frame to reach the common goal: to be a winner in the game.In order to design a simulation agent of soccer, many technologies should be taken into account, such as design principle of autonomous agent, cooperation between multi-agent, acquest of strategy and real-time reasoning. Presently, the key points focus on research area of online learning, real-time strategy under counteractant environment, and the cooperation and collaboration among agents, including the research of dynamic cooperation mechanism, collaboration strategy and advanced offensive and defensive strategy.