Motion Planning for Carlike Robots Using a Probabilistic Learning Approach

Motion planning for many degrees of freedom-random reeections at c-space obstacles. In Proc.bilistic roadmaps for path planning in high dimensional conngura-tion spaces. LS89] J.-P. Laumond and T. Sim eon. Motion planning for a two degrees of freedom mobile robot with towing. A motion planner for car-like robots based on a global/local approach. 37 learning method remains a topic of further research. 9 Discussion and conclusions We have described and extended a probabilistic technique for solving the learning motion planning problem in static environments, and we have applied it to two types of car-like robots. In the learning phase a probabilistic roadmap is incrementally constructed, which can subsequently be used for solving individual motion planning problems in the given scene. The method, which has previously proven to be very fast for free-ying robots and (high dof) articulated robots, achieves very good results for both general car-like robots as well as for forward car-like robots, in diierent types of scenes. Conceptually it is a very simple method, and the graphs that it builds are small representations of the free connguration space. The method is highly exible. In order to apply it to some particular robot type, all that is needed is a local method which computes paths feasible for this robot type, and some (induced) metric. Experimental results ((Mas92], Sve93]) indicate that very simple local methods achieve the best results. Hence, it is usually easy to nd a suitable local method for some given robot type. For robots with the reversibility property, we have also shown that a proper choice of the local method guarantees probabilistic completeness of the learning method. Another property which one may wish is that (concatenations of) local paths describe \smooth" motions of the robot. The paths generated by the planners described in this paper are solely composed of straight line-and circle segments. In practice, due to various mechanical constraints, such paths will cause the robot to jerk while traveling along its path (in points where the straight line-and circle segments meet). There exist however local methods which take into account the smoothness constraints imposed by the control engineering level ((Nau94]). We believe we can incorporate such local methods into our global approach. This will be investigated in the future. Currently we are working on the application of the method to a number of diierent robot types, such as solid robots in 3 dimensional workspaces, and robots with …