Re: [Algorithms] GJK

[Pierre T. <p.t...@wa...>]

Ok, I have my very first version running - and my head hurts.

Pretty basic for the moment, but it seems to work. The computed distance
does not look very accurate nonetheless. For example I saw it reported as
0.000261 altough the two objects obviously collided. Duno if it's normal. I
skipped the backup procedure (I actually *forgot* about it) but my basic
tests worked anyway. I think Gino was right when he discarded it. Running
time is correct, provided you use hill-climbing. If you don't.... *grin*, if
you don't, just forget it. (ok, unless your meshes have few triangles). I
used Rabbitz' subroutine for Johnson's algorithm, so that I could build the
Gilbert part on a working basis. Now I'm going to recode that one my way.

I think some day I'll also pack all my notes in a ppt file for interested
people.

Pierre