A global motion planner for a mobile robot on a terrain

Moving on a terrain (mountain landscape), a mobile robot must avoid collision with the ground, it must be stable without risk of tumbling and its wheels must touch the ground. The methods for solving the motion planning problem for such an all-terrain mobile robot can be divided into two classes. A local method plans the motion having only local information about the configuration space [4] and therefore has the drawback that the robot might get stuck in a deadlock. In contrast, a global method performs a preprocessing step to compute a representation of the complete space of the configurations of the robot that are collision free and satisfy the stability constraints. Both methods can be used in a complementary way: the global method that uses a simplified model plans the coarse motion and computes intermediate goal configurations for the local method that does the fine planning and thus tends to avoid deadlocks. There are a few papers presenting local optimization methods for the above motion planning problem. The —to our knowledge— only global method computes an octree representation of the space of collision free and stable configurations [2].