A fast collision detection method based on point clouds and stretched primitives for manipulator obstacle-avoidance motion planning

Abstract

It is essential to efficiently perform collision detection for robotic manipulators obstacle-avoidance planning. Existing methods are excellent when manipulator links are simple and obstacles are convex. But they cannot keep the accuracy and the efficiency at the same time when manipulator links or obstacles are nonconvex. To decrease the computing time and keep a high accuracy, this article presents a collision detection method based on point clouds and stretched primitives (PCSP). In traditional methods, obstacles are often represented either by a convex body or enormous amounts of points. But this needs a trade-off between the accuracy and the computing time when obstacles are concave. In the proposed method, we represent obstacles and complex manipulator links as stretched geometric bodies while simple manipulator links are enclosed by capsules with different sizes. The stretched body is constructed by the original point cloud from sensors but it only requires a small number of points to approximate the original object. We conducted the simulation experiment in our specific scenarios, and the results indicated that PCSP required less computing time while maintaining a high level of accuracy compared to existing methods. We also conducted standard benchmark tests in general scenarios, which showed that PCSP had advantages over libraries based on bounding volume hierarchies when concave objects are close together. Finally, we implemented PCSP for a manipulator obstacle-avoidance motion planning in a real-world environment, which demonstrated that PCSP was effective.