Application of Artificial Evolution to Obstacle Detection and Mobile Robot Control

The Fly Algorithm is an evolutionary algorithm used for stereoscopic reconstruction. In the classical approach, a pair of stereo images is processed in order to extract 3-D information and to infer a representation of the scene. Conversely, the Fly Algorithm builds potential 3D models of the scene and tests their consistency with the two stereo images. This chapter will present some recent improvements of the algorithm, and its concrete application to obstacle detection and avoidance in mobile robotics. Section 2 presents the notion of individual approach in evolutionary algorithms and the principle of the Fly Algorithm. New genetic operators are introduced. Several internal parameters can drastically change the algorithm behaviour: this issue is the topic of section 3, where a Pareto multi-objective optimisation leads to the obtention of an efficient set of parameters. Section 4 describes a real time application to automatic driving on an electrical vehicle of the IMARA Team (INRIA). We focus on stop/go control and on direction control in order to avoid obstacles in front of the vehicle. The methods used are explained and results are shown. Finally, section 5 concludes the chapter and gives some ideas of future work to further improve the algorithm.