基于一般和随机对策论框架下的多智能体学习

将Q-learning从单智能体框架上扩展到非合作的多智能体框架上,建立了在一般和随机对策框架下的多智能体理论框架和学习算法,提出了以Nash平衡点作为学习目标.给出了对策结构的约束条件,并证明了在此约束条件下算法的收敛性,对多智能体系统的研究与应用有重要意义.

MULTI-AGENT LEARNING BASED ON GENERAL-SUM STOCHASTIC GAMES

Q -learning from original single-agent framework is extended to non-cooperative multi-agent framework, and the theoretic framework of multi-agent learning is proposed under general-sum stochastic games with Nash equilibrium point as learning objective. We introduce a multi-agent Q -learning algorithm and prove its convergence under certain restriction, which is very important for the study and application of multi-agent system.