Monte-Carlo collision detection
This paper presents a method for detecting collisions between objects
under the hard real-time constraints of a virtual reality
simulation. A list of potential collision regions is computed and
updated over time, using temporal coherence to reduce the cost of this
update. New samples are constantly randomly generated on every object
in order to discover new interesting regions. The objects are then
efficiently tested for collision using a multiresolution layered shell
representation, which is locally fitted according to an evaluation of
the objects' distance. Amortized algorithms allow the user to trade
accuracy for speed, in order to reach real-time
performances. Deformable objects and auto-collisions are handled by
our algorithm without any change, with a validity that decreases with
the amplitude of the deformation. We show how a multiresolution
deformable object simulation can be linked with the collision
detection process in order to optimize the simulation. We demonstrate
our method in a context of virtual reality by simulating realistic
dynamic collisions between several and possibly deformable objects,
with a guaranteed frame rate. Benchmarks indicate that the method
favorably compares to alternative methods, including those which are
restricted to (and optimized for) rigid objects collision detection.
Images and movies
BibTex references
@TechReport\{GD04, author = "Guy, St\'ephane and Debunne, Gilles", title = "Monte-Carlo collision detection", institution = "INRIA", number = "RR-5136", month = "March", year = "2004", keywords = "Collision Detection, Stochatic", url = "http://artis.inrialpes.fr/Publications/2004/GD04" }