Re: [Algorithms] Using RAPID within a rigid body simulator

[<Chr...@Pl...>]

Pierre Terdiman wrote:
>In rigid body simulations, one needs a correct estimation of the distance
>and contact points between two bodies, in order to:
>- backup the simulator in search of the exact collision time
>- compute an accurate collision response
>
>If you followed the recent GJK thread, you know this algorithm works well
>for such tasks. But what about other collision detection methods? For
>example, say I'm just using OBBs, as with RAPID. RAPID will detect
>collisions, but won't give me a distance or a pair of closest points. Does
>it mean it can't be used efficiently within a rigid body simulator ?


A potential problem with RAPID is that it works from a 'polygon soup', and
as such has no concept of solidity to distinguish between insideness and
outsideness.

So, depending on how small or fast your objects are, RAPID might not be
able to detect all your collisions (i.e. object A fully inside object B)
unless it's enhanced to do so. I guess this is probably not a problem
in general, as you're probably running your physical simulation at a fairly
high frame rate, but it could be an issue in some cases. Of course, the
triangle-triangle test could of course be extended to handle velocity,
which would fix this problem.


Also, I'm no expert on physical simulation, but one important aspect is
computing the contact manifold (the set of points of contact) and again
this is something RAPID would have to be enhanced with, to be able to
provide.


Christer Ericson
SCEA, Santa Monica