A multi-agent fuzzy-reinforcement learning method for continuous domains

Abstract

This paper proposes a fuzzy reinforcement learning based method for improving the learning ability of multi-agents acting in continuous domains. The previous studies in this area generally solved multi-agent learning problem by using discrete domains. However, the most of real-world problems use the continuous state spaces, Also, it is really a difficult task to handle the continuous domains for multi-agent learning systems. In this paper, proposing a novel approach, we will have two significant advantages according to the conventional multi-agent learning algorithm. One of them is that the number of state spaces of learning agents in multi-agent environment only depends on the number of fuzzy sets which were used to represent the state of an agent. Whereas, in the previous approaches, the visual area of agent or the size of domain were taken into consideration for the state space. The other advantage is that the employed environment has a continuous domain as in the real-world problems. Experimental results obtained on a well-known pursuit domain show the effectiveness of the proposed approach.