Abstract : A robot system basically acts by moving in a physical world. Motion planning is therefore a key component of its autonomy. This is the main topic adressed by our work. The presentation is organized along four axis. The first axis concerns algorithmic motion planning. Our objective has been to favour effective planning methods. Hence, most of the real problems resist to exact algorithms; we present alternative methods, satisfying a weaker form of completeness (eg. probabilistic), but more adapted to the needs of the applications in terms of algorithmic efficiency and robustness.Then we consider extended problems more directly related to the notion of robot task planning. The second part addresses manipulation planning in presence of movable objects. We propose a geometric formulation allowing to extend the configuration space concept to the automatic planning of such tasks.Also, robot motions are executed in real world settings. The robustness of the algorithms has to be extended to the whole motion planning and control loop, in order to face uncertainties of the models and the inaccuracy of the execution. The approach proposed in the third part combines planning with uncertainty and reactive modes: the planning algorithms use the a priori knowledge about the task to produce sensor based motion primitives adapted to mobile robots.Finally, the last work concerns motion planning for outdoor wheeled vehicles moving onto natural and unstructured terrains. The proposed techniques allow to deal with complex locomotion systems and highly irregular terrains.