Reconstruction tridimensionnelle et exploration autonome d'environnements non structures (French text)

Unstructured indoor or outdoor environments present difficult challenges caused by uneven terrain, the presence of obstacles and the absence of easily distinguishable landmarks for localization.; Our approach is based on a range finder mounted on a mobile robot aiming at the ground and also returning environment profiles. The terrain model is then computed by merging these slices into a global mesh representing the terrain. The advantages of such a system are low computation cost and high precision. Navigation algorithms grant safe navigation of the platform in the explored environment. Odometer and inertial sensors are used to compute the platform 3D localization. This pose is then used to place the profiles and calculate the points position in a global frame. Those points are placed into a mesh based on a Delaunay triangulation in order to model the terrain surface. Several filter are applied to remove redundant points, to allow the operator to set the map resolution, and to ensure the model consistency. A real-time visualization module allows the operator to supervise the exploration process and see the map during its construction on a remote computer. The reactive navigation is based on this 3D map to detect obstacles to build tire traversability map, to plan a safe trajectory and to detect unexplored areas. (Abstract shortened by UMI.)...