Re: [Algorithms] N-body processing

[David B. <db...@bt...>]

> Now, my concern is this: how do I track the possible colliding pairs? How
do
> I store them? How do I handle them? For the moment Adam seems to be the
only
> one who actually understood the problem: that's why he introduced the
sparse
> matrices.


I just use a hash table, and a simple modulus hash key(with the polyhedra
indexes), no performance problems with this and very easy to
implement(especially when you use STL:-). Objects are stored in an unsorted
linked list off each hash entry.
I don't use the pair table for anything other than caching closest feature
pairs to improve performance.

When you actually detect that there has been a collision, i.e. the closest
distance algo returns something smaller than your epsilon. You need to look
at the features(verts, edges or faces) surrounding the closest points and
determine if they are as close to each other as the closest points. This is
determining the contact manifold, which can be two points,  point and edge,
edge and edge, point and face or face and face. The contact manifold is then
used to determine the collision response.

David

http://www.dblack.btinternet.co.uk
ICQ #: 24402391
Mobile: (UK) 0778 7836188