An incremental fuzzy controller for large dec-POMDPs

This paper proposes an incremental fuzzy controller to find a sub-optimal policy for large multi-agent systems modeled as DEC-POMDPs. This algorithm employs a compact fuzzy model to overcome the high computational complexity. In our method, each agent uses an individual fuzzy decision maker to interact with the environment. An incremental method is utilized to tune the rule-base of each agent. Reinforcement learning is used to tune the behavior of the agents to achieved maximum global reward. Moreover, we propose an elegant way to create initial rule-base according to the solution of the underlying MDP to increase the performance of the algorithm. We evaluate our proposed approach on several standard benchmark problems and compare it to the state-of-the-art methods. Experimental results show that the proposed incremental fuzzy method can achieve better results compared to the previous methods. Using compact fuzzy rule-base not only decreases the amount of memory used but also significantly speeds up the learning phase and improves interpretability.