Utilisation d'un compas visuel pour la navigation d'un robot mobile

Abstract

In this thesis, we consider navigation and orientation of a robot using measurements of angles, without any distance information or reconstruction of the environment. Previous models derived from experiments involving bees showed that a minimalistic approach suffices to perform homing if the orientation is known. We analyse the consequence of an error in orientation and the degraded functionality that remains possible, but is unsatisfactory. We elaborate an original and minimalistic reorientation method, called ``visual compass''. The observation of three panoramas of known orientation and a first-order approximation is sufficient stored information for reorientation. We compare the visual compass with another method, simpler and more approximate. We show formally and experimentally that the precision of each is sufficient to navigate and that their range can be arbitrarily extended via the building of a cognitive map.