[ODE] Stable trimeshes ?

[Jeffrey Smith]

Marco wrote:
> Are there any projects currently going on for providing stable trimeshes ?

I am working on improving the trimesh-trimesh collisons.  There are
basically two large peices of functionality that ODE needs to do
trimesh-trimesh collisions "right."  

1) Nonconvex decomposition - Take a nonconvex object and turn it into
   a set of watertight convex objects.

2) Penetration depth calculation for convex objects

I have basically finished part 1, implementing a modified version of
"Convex Decomposition of Polyhedra and Robustness.", Bajaj, C., and
Dey, K., after the cut-and-split technique of Chazelle.  My employer
is supportive of contributing this code to the ODE community, but it
probably won't happen for a few months.

The second part is much harder. According to what I've read, the
expanding polytope algorithm (EPA) should give good results for
penetration depth (and witness point) calculations.  However, I've
been testing an implementation of this and the results aren't that
great.  It's possible I'm using this code incorrectly, but it's
equally probably that ODE's current technique of handling collisons
is unsuitable for accurate convex collisions.

Jeff