Efficient contact determination between solids with boundary representations (B-Rep)

Overview & Synopsis

With the development of advanced robotic systems and complex teleoperation tasks, the need to perform simulations before operating on physical systems becomes of increasing interest for feasibility tests, training of the human operators, motion planning, etc. Such simulations usually need to be performed with great accuracy of physical phenomena if, e.g., the operator is expected to face the same ones in the real world and in the virtual scene. Collision detection, i.e., the computation of contact points and contact normals between interacting rigid bodies, occupies a time-consuming part of such a physical simulation. The accuracy and smoothness of such contact information is of primary importance to produce a realistic behavior of the simulated objects. However, the quality of the computed contact information strongly depends on the geometric representation of the parts of the virtual scene directly involved in the mechanical simulation. On the one hand, discrete volumes-based and tessellation-based geometric representations allow very fast contacts generation at the cost of the potential introduction of numerical artifacts due to the approximation of the interacting geometrical shapes. On the other hand, the use of boundary representations (issued by CAD modelers) composed of smooth curve and surfaces removes this approximation problem but is currently considered being too slow in practice for real-time applications.This Ph.D focuses on developing a first complete collision detection framework on solids with smooth boundary representations that achieves real-time performances. Our goal is to allow the real-time simulation of industrial scenarios that require a high level of accuracy. Typical applications are insertion tasks with small mechanical clearances. The proposed approach is based on a bounding-volume hierarchy and takes advantage of key features of industrial mechanical components which are often modeled with surfaces describing functional contacts with canonical surfaces (cylinder, sphere, cone, plane, torus) while contacts over free-form surfaces like B-Splines are mostly accidental and encountered during operations of maintenance and manufacturing. We augment this hierarchy with the identification of supermaximal features in order to avoid redundant exact localization of contact points on canonical surfaces that may be represented as distinct features of the CAD model. In addition, we define polyhedral normal cones that offer tighter bounds of normals than existing normal cones of revolution. Moreover, we extend our method to handle configurations that involve beams modeled as deformable dilated Bézier curves. Finally, parallelization of the full approach allows industrial scenarios to be executed in real-time.