A framework for fast and accurate collision detection for haptic interaction

We present a framework for fast and accurate collision detection for haptic interaction with polygonal models. Given a model, we pre-compute a hybrid hierarchical representation, consisting of uniform grids and trees of tight-fitting oriented bounding box trees (OBB-Trees). At run time, we use hybrid hierarchical representations and exploit frame-to-frame coherence for fast proximity queries. We describe a new overlap test, which is specialized for intersection of a line segment with an oriented bounding box for haptic simulation and takes 6-36 operations excluding transformation costs. The algorithms have been implemented as part of H-COLLIDE and interfaced with a PHANToM arm and its haptic toolkit, GHOST, and applied to a number of models. As compared to the commercial implementation, we are able to achieve up to 20 times speedup in our experiments and sustain update rates over 1000 Hz on a 400 MHz Pentium II