Fusion of fuzzy agents for the reactive navigation of a mobile robot

In this paper we propose a sensor-based navigation algorithm based on the fusion of elementary behaviors. The proposed navigator combines two types of obstacle avoidance behaviors, one for the convex obstacles and one for the concave ones. To avoid the convex obstacles the navigator uses a fuzzy tuned artificial potential field method. The concave obstacle avoidance behavior results of the mixture of a "wall-following" behavior with a creation of transition subgoals. An automatically on-line tuned fuzzy wall-following system using a neuro-fuzzy structure is designed. The incorporation in the learning cost function of a weight decay term prevents from an excessive growth of the weights and allows a permanent learning. This method leads to a very quick and efficient learning giving a robust controller optimized with respect to the actual physical characteristics of the robot. The effectiveness of the proposed method is verified by carrying experiments on the miniature mobile robot Khepera(R).