A simple neuro-fuzzy controller for car-like robot navigation avoiding obstacles

Abstract

This paper describes how the combination of neuro-fuzzy techniques with geometric analysis offers a good trade-off between purely heuristics and purely physical approaches when solving the problem of car-like robot navigation. The controller described, which follows a reactive technique, generates trajectories of near-minimal lengths when no obstacles are detected and, in presence of obstacles, generates minimum deviations from them. All these reference paths meet the kinematic constraints of car-like robots and take into account dynamic issues. Besides its efficiency, the proposed controller is very simple and linguistically interpretable. The whole controller has been designed and verified by using the CAD tools of the Xfuzzy environment.