Heuristic and learning method for obstacle avoidance with mobile robot

Abstract

In this paper, a fuzzy controller obstacle avoidance of the mobile robot Pioneer II is proposed. The fuzzy inference system FIS of this controller is performed by two methods: heuristic and reinforcement learning. the manual tuning of the fuzzy control system can be long and difficult. In contrast, reinforcement learning has proven theoretically and practically its ability to automatically optimize some parameters of the FIS. For that, the Fuzzy Actor-Critic Learning algorithm allows the determination of the parameters of the conclusions among of an available set fixed by the operator. The proposed algorithm allows the automatic determination of the parameters of the conclusions of the fuzzy rules. The simulations show that the two controllers (heuristic, RL controller) are able to avoid the different shapes of obstacles contained in known environments, and they show exceptionally good robustness when changing the environment (shape of obstacles, location of obstacles in the environment