Stratégies de commande référencées multi-capteurs et gestion de la perte du signal visuel pour la navigation d'un robot mobile

Abstract

The literature provides many techniques to design efficient control laws to realize robotic navigation tasks. In recent years, the sensors improvement gave rise to the sensor-based control which allows to define the robotic task in the sensor space rather than in the configuration space. In this context, as cameras provide high-rate meaningful data, visual servoing has been particularly investigated, and can be used to perform various and accurate navigation tasks. This method, which relies on the interaction between the camera and the visual features motions, consists in regulating an error in the image plane. Nonetheless, vision-based navigation tasks in cluttered environment cannot be expressed as a sole regulation of visual data. Indeed, in this case, it is necessary to preserve not only the robot safety (i.e. non-collision) but also the visual features visibility. This thesis addresses this issue and aims at developing sensor-based control laws allowing a mobile robot to perform vision-based tasks amidst possibly occluding obstacles. We have first proposed techniques able to fulfill simultaneously the two previously mentioned objectives. However, avoiding both collisions and occlusions often over-strained the robotic navigation task, reducing the range of realizable missions. This is the reason why we have developed a second approach which lets the visual features loss occurs if it is necessary for the task realization. Using the link between vision and motion, we have proposed different methods (analytical and numerical) to compute the visual signal as soon it becomes unavailable. We have then applied them to perform vision-based tasks in cluttered environments, before highlighting their interest to deal with a camera failure during the mission.