Research of Multi-Agent Cognition and Decision
Liu Dong, Yan xuefei, Xinming Li, Wang Shoubiao · 2017
Reinforcement Learning(RL) is the main cognition and decision method of Agent since its adaptation and exploration ability to the unknown environment, but it cannot ensure the convergence for the multi-Agent situation since the transition of the state space is not only influenced by the Agent itself but also influenced by the other Agent's action, so the other Agent's action has to be taken into account. In consideration of the deep theory foundation and advantages to deal with the other player's policy, this paper think that combination of the games theory and multi-Agent is a breakthrough for the cognition and decision in the status of the multi-Agent Learning. Based on the introduction of the traditional decision technology, the learning technology and the typical algorithm framework of the multi-Agent reinforcement learning with the games theory, some related algorithms such as MinMax-Q, nash-Q, FF-Q, CE-Q and Nego-Q were presented and analyzed.