A Study Of Multi-Agent Reinforcement Learning Based On Credit Assignment

Reinforcement learning has been the hotpot in the research of multi-agent systems(MAS) and machine learning ,because it doesn't need the environment model. In fact, a reinforcement-learning agent learns through its interaction with the environment. MAS is often applied into open,complex and dynamic environment,in which a single agent is insufficient to solve the faced task , in order to adapt to the environment's dynamic changes,agents must have the learning capability as well.But the traditional single agent learning theory cannot hold true in the case of MAS.The framework of multi-agent reinforcement learning(MARL) is Markov games.The writer aims at the research on the game theory and proposes a new method to solve the problem of temporal credit assignment and the structure credit assignment in MARL.The following 3 points are the main research achievements and innovations .The software that used in this paper is visual C#.net. And the platform of the research is the pursuit game problem.Pursuit game problem is considered as a dispersed dynamic game problem with two players.The credit assignment function is constructed with the distance influence,movement influence ,credit influence considered. The payoff matrix of the pursuer is got by calculating the threat value under different strategies. The optimal strategy for each movement is obtained by resolving the payoff matrix.the mathematical model used for strategy selection and a new MARL algorithm based on credit assignment is presented.●Analyzing algorithm and research actuality of MARL deeply;●Presenting a new MARL algorithm based on credit assignment combined with the Makov game environment and advanced research work in point;●Proved the validity and astringency of the new MARL algorithm by the case analysis and emulation.