In this research project we intend to investigate the problem of planning feasible paths for car-like robots in environments cluttered by obstacles. Planning algorithms often rely on the notion of clearance from obstacles which is closely related to the length of the shortest feasible path to an obstacle (the two best known examples are skeletonization and potential field methods). To make the use of the shortest path metric appealing in real-time applications it is necessary to develop i algorithms that compu efficiently the length of these paths. Studies on this subject have been conducted in a joined work by Jean-Paul Laumond and the applicant. in the context of the EC project ESPRIT 3 Basic Research :6546 (PROMotion). l providing a new efficient geometric algorithm for distance computation (see the publications of the applicant). These promising results must be validated by integrating the proposed algorithm in a planner for a real-time moving platform. It is particularly advisable. to carry out this experimentation at LAAS-CNRS since in the laboratory three mobile robots (one carrying a trailer) are available. Another interesting question, arising from the previously mentioned work, which has not been vet investigated. deals with the characterization of the visibility sets for a car-like robot. This is closely related to the visibility problem classically defined in the framework of Euclidean metric. In the previous collaboration with Jean-Paul Laumond the applicant has characterized the visibility sets in position (i.e. the sets of positions in the plane reachable by a collision free shortest path unaffected by the presence of the obstacles). The extension to the 3-dimensional configuration space l (position in the plane and orientation of the robot) will be accomplished in the context of the proposed project to study the l complexity of the problem of planning optimal paths for a car-like robot in presence of obstacles. Training content (objective, benefit and expected impact) The results obtained in nonholonomic motion planning and control in the last six years are supported by sophisticated tools coming from Computer Science, Control Theory and Mathematics and call for further development. Moreover. the limitation of existing techniques stems from their inability to integrate uncertainties, perception functionalities and sensor based motion primitives at the kernel of the motion planning and control algorithms. The requested training is particularly opportune since the know-how of the applicant on sensor-based motion planning in unknown environments will join the experience characterizing the host institution in the theoretic study of nonholonomic motion planning problems. A joined work will allow overcoming the limitation of the existing techniques previously mentioned besides of providing a new efficient planner based on the results already obtained in the joined work by the applicant and Jean-Paul Laumond.
Links with industry / industrial relevance (22) The group Robotics and Al at LAAS has close relationship with industry involved in the market of Mobile Robotics
Thomson-MidiRobots, Alcatel. Framatome...). The research subject of the applicant lies at the kernel of the mobility function in Mobile Robotics. The objectives of the work is to make the expected results on obstacle avoidance transferable in the framework of these industrial relationship.