[ODE] collision detection optimalizations for humanoid

Piotr Obrzut piotr_obrzut at o2.pl
Sat Jan 8 10:59:29 MST 2005


8 stycznia 2005 (08:16:39) Erin wrote:

EC> I see your point; I checked and ODE does compute bounding boxes for each
EC> HashSpace. This would quickly cull the two humanoids if they are not
EC> touching. Unfortunately, once the two HashSpace boxes overlap, the code
EC> descends into an O(n^2) test. I think that overall this would be a loss
EC> compared to using a single HashSpace, but I could be wrong because I have
EC> not profiled the HashSpace (i.e. it might be slow).

i think that the best way will be own grater colliders for model and
one HashSpace for all lower colliders.

But i was thinking on sth like collider tree for humanoid model:
every collider connected to parent bone will "contain" all child
bones colliders; then it will be possible to estimate all boxes from
all possible realtive moves (from joints or/and from recorded
animations); but the win may not be worthy b/c it could take a lot of
time to do sth like this.

 Piotr Obrzut

More information about the ODE mailing list