RE: [Algorithms] Explicit minkowski sums (long'ish)

[Adam P. C. <ac...@st...>]

On Fri, 25 Jul 2003 chr...@pl... wrote:

> Chris Butcher wrote:
> > Halo 1 used exactly this for sphere/capsule vs polygon mesh. The nice
> > thing about doing your character movement checks this way is that by
> > raycasting against the minkowski sum of your environment, you are
> > guaranteed that the capsule can never penetrate or fall through the
> > environment.
> This needs to be qualified to hold true, specifically how the ray
> test is done.

> Of course, adding epsilon terms into your ray tests effectively
> adds thickness to your polygons, which allows the day to be saved.

Given - and I think this is probably the usual fix;  does anyone actually
try to do penetration depth computation instead?  This might be a bit
nasty on a mesh or BSP - but for OBBs or something, might not be so bad.
Another thought, does the 'nearest plane' on the CSO to an internal point
(which would be computed to give penetration depth) somehow identify the
penetration points/features? Hmmm... will have to draw a picture later :)
Does anyone use a backup solution to deal with penetration?  Or is this
sufficiently rare (with a large enough epsilon) that you don't worry?  It
seems like you're bound to end up with trucks crushing players into the
ground, and having the fall-through problem because the collisions can't
be resolved.

Adam C.