[ODE] does ODE need convex polyhedra collision?

[nlin at nlin.net]

On Mon, Oct 13, 2003 at 05:37:15PM +0200, Sergio Valverde wrote:
> Anyway, the problem is not in the penetration depth computation.
> 
> If you plan to integrate those closest-features collision
> detectors you should be able to devise a way to compute not only the
> SINGLE pair of closest points but the full set of contact points 
> required by ODE in order to avoid interpenetration 
> (think of face-face, edge-face, and the like). 
> In general, every contact consists of SEVERAL contact points. 
> The algorithms you commented they only return ONE contact point, which
> is not enough.

This isn't directly related to the concerns you raise, but it's worth noting
that you may be able to use a "hack" to compensate for inaccurate collision
detection.

I assume that the full set of contact points required to avoid
interpenetration is needed mostly for stable stacking (in other cases,
objects fall down anyway, so accuracy is, in games at least, not
important).

Stacking may still be possible even with single-point penetration depth
information by putting the rotational component of objects "to sleep" when
they are moving slowly, by setting the inertia tensor to infinity. This
effectively simplifies the rigid body motion to particle motion. For more
information see Gino van dan Bergen's USENET article:

http://groups.google.com/groups?hl=en&lr=&ie=UTF-8&oe=UTF-8&selm=a91htu%244jp%241%40news.tue.nl&rnum=8

The last paragraph is the relevant bit.

I'm not sure how well this hack works, and it's rather unphysical, but it may
be of use in some situations.

-Norman