Fuzzy Multi-Agent Cooperative Q-learning

Abstract- This paper presents a cooperative reinforcement learning algorithm of multi-agent systems. The cooperative behaviour is established within a leader-following framework. Specifically, the cooperative dynamics is modelled as a Stackelberg game. Based on the equilibrium definition of the Stackelberg game, a leader-following Q-learning algorithm is developed. The algorithm is generalised over continuous state space by using fuzzy logic. Index Terms- Reinforcement learning, multi-agent systems, multi-agent reinforcement learning, fuzzy Q-learning