Probabilistic Roadmaps for Path Planning in High-Dimensional Con(cid:12)guration Spaces

Real-time robot motion planning using rasterizing computer graphics hardware. In Proc. OY82] C. O'D unlaing and C.K. Yap. A retraction method for planning the motion of a disc. A local approach for path planning of manipulators with a high number of degrees of freedom. a path generation algorithm into oo-line programming of airbus panels. Motion planning for many degrees of freedom-random reeections at c-space obstacles. REFERENCES 31 iterative collision checker. We observed no dramatic slowdown of the planner. A challenging research goal would now be to extend the method to dynamic scenes. One rst question is: How should a roadmap computed for a given workspace be updated if a few obstacles are removed or added? Answering this question would be useful to apply our method to scenes subject to small incremental changes. Such changes occur in many manufacturing (e.g., assembly) cells; while most of the geometry of such a cell is permanent and stationary, a few objects (e.g., xtures) are added or removed between any two consecutive manufacturing operations. Similar incremental changes also occur in automatic graphic animation. A second question is: How should the learning and query phase be modiied if some obstacles are moving along known trajectories? An answer to this question might consist of applying our roadmap method in the conngurationtime space of the robot Lat91]. The roadmap would then have to be built as a directed graph, since local paths between any two nodes must monotonically progress along the time axis, with possibly additional constraints on their slope and curvature to reeect bounds on the robot's velocity and acceleration. A motion planner with performance proportional to task diiculty. 30 In KL94a, KL94b, O S94] prior versions of the method have been applied to a great variety of holonomic robots including planar and spatial articulated robots with revolute, prismatic, and/or spherical joints, xed or free base, and single or multiple kinematic chains. In Sve93, SO94] a variation of the method (essentially one with a diierent general local planner) was also run successfully on examples involving nonholonomic car-like robots. Experimental results show that our method can eeciently solve problems which are beyond the capabilities of other existing methods. For example, for planar articulated robots with many dofs, the customized implementation of Section 5 is much more consistent than the Randomized Path Planner (RPP) of BL91]. Indeed, the latter can be very fast on some diicult problems, but it …