RE: [Algorithms] Dumb OBB-tree optimization ?

[Justin J. <ju...@mo...>]

> You know, power plants and such..... Here's what got me thinking about all
> this:
> http://www.cs.unc.edu/~larsen/290/project.html
>

Ah. I'm enlightened. How naive of me not to consider stuff like this.

> The boolean answer is enough to find Tc, the collision time. It has been
> discussed in many places, for example in Brian Mirtich's thesis.
> Baraff then suggested a distance estimate would be better to do the job,
but both
> methods work.

Its been a while since I've read Baraff's papers... I'll have to dig them
out again.
But you are quite right of course.

> Now, this has actually little to do with penetration depth, which is only
> required with penalty methods to compute collision response. In order to
> find the collision time, a distance estimate is enough. There's a subtle
> difference here: computing separation distances, even between non-convex
> bodies, is relatively easy. For non-convex penetration depths, I
> don't even know where to start....

Yep. Non-convex penetration depths is a nightmare. I wouldnt even bother
thinking about it.

Seperation distance is all you need to know.

> Finally, as far as the "time and normal" information is
> concerned, there are
> two things:
> > Triangles and point of intersections would be really useful though.
>
> As said above, you almost got them :)

True enough. Hmmm. I need to figure out how to get a meaningful contact set
using just a true/false collision check and bisection.

Justin Johnson - Technical Director
Mobius Entertainment
ju...@mo...