Study Of Multi-agent Learning Problem Based On Reinforcement Learning

With the development of cybernetics and computer technology, the theories of multi-agent system and relevant application study in distributed artificial intelligence already become the study focus of artificial intelligence and intellectual control. Robot soccer game is a international match and academe activity, which attempts to foster distributed artificial intelligence (DAI) and intelligent robotics research by providing a standard problem where a wide range of technologies can be integrated and examined.The robot soccer game environment is a dynamic, complicated, oppositional one. And each robot can not get all the information on the field. Then how to improve the performance of the decision-making ability by learning themselves is the key to develop the robot soccer system.This thesis mainly focuses on the implement of the reinforcement learning on the robot soccer games.First, to solve the complexity of the Robocup environment and the bulkness of the state space, the credit assignment algorithm is introduced into the CMAC-Q learning algorithm, and then the improved learning speed can be seen.Second, the option algorithm and intrinsic reinforcement learning is used to prove the kick technology, intercept technology and cooperate technologies. And the decision-making ability is proved to be effective by the algorithm, which can be seen from the match.