Adaptive algorithm for multi-agent learning optimal cooperative pursuit strategy based on markov game

Abstract

There are two problems in the process of learning optimal cooperative pursuit strategy for multiple agents in MAS (multi-agent system). One is there is usual circulation among the actions chosen by agents which make the learning process not converging, the other is there are many conflicts among the actions chosen by agents which make the learned pursuit strategy not optimal. In this paper, the procedure of learning optimal pursuit strategy for multi-agent is regard as a Markov game, and the best equilibrium of the game is regard as the converging, stable state of cooperative pursuit learning. An adaptive algorithm for multiple agents to select and attain an optimal, consistent equilibrium is proposed based on fictitious play. The simulation verifies the effectiveness of the algorithm.