[ODE] TriMesh/Box Problem

Pierre Terdiman pierre.terdiman at novodex.com
Mon Feb 9 09:54:15 MST 2004


> Trimesh/box collision are indeed much more accurate now than they
previously
> were, but there are still problems that need to be resolved. I've made a
lot
> of tests using custom 3dsmax scenes and I can still find some cases where
a
> box will start sliding where it shouldn't, get "hooked" into geometry or
> will simply explode.

Trimesh/box is an ill-defined problem. As soon as you let them penetrate,
i.e. as soon as your engine works with penetration depths, you just *can't*
solve it in all cases. A lot of those cases are ambiguous, and there's
simply no way to select the correct contact normal by just looking at a
single instantaneous frame (what I call the "static problem"). To resolve
all the ambiguities you have to use more information, such as the collision
history (where is the box coming from ?), or at least the velocity vectors.

All in all it's a very hard problem, a lot more difficult than what people
usually expect. I spent a huge amount of time implementing robust box-mesh
contact generation for Novodex... Last time I checked ODE's box-mesh code,
it was very very incomplete - to tell the truth I simply didn't expect it to
work at all. The problem you mention is only a particular case among a lot
of other painful cases :)

Pierre Terdiman

- Novodex AG (www.novodex.com)
- Personal : www.codercorner.com





More information about the ODE mailing list