Hierarchical reinforcement learning based on multi-agent cooperation game theory

Abstract

Reinforcement learning is a good way for multi-robot systems to handle tasks in unknown or fuzzy models. MAXQ is a layered reinforcement learning algorithm, but some of its parameters are limited. In order to improve the ability of processing tasks, in this paper, the improved MAXQ method is employed to adjust the parameters of the collaborative rules. In addition, in order to improve the task allocation speed, a cooperative game method is used among multiple robots. Finally, to test our approach, we conducted a series of experiments. The experimental results demonstrated that our model achieved better as compared with other detection models. The convergence speed is increased, by about 26%. It shows that our model is effective and efficient.