Algorithmique de la planification de mouvement probabiliste pour un robot mobile

Abstract

In this thesis, we are studying incremental probabilistic motion planning. Our studies present a new fast algorithm to expand Rapidly exploring Random Tree (RRT) and a new irregular cell partition based on visibility. Our algorithm improves the existing successful probabilistic path planner called RRT by restricting each expansion step to the first collision free configuration. The analysis of the principal sampling's properties used in probabilistic motion planning leads us to propose a new irregular cell partition based on visibility. This new decomposition is tested in narrow environments and in cluthered ones. Results show that this new algorithm and this new decomposition are two significant componants of RRT methods. The motion planner we developped is implemented for mobile robot, evolving in a static well-known environment.