[ODE] collision detection optimalizations for humanoid

Jon Watte hplus-ode at mindcontrol.org
Thu Jan 6 17:08:59 MST 2005


> well the problems starts when 2 human models are going to collide
> mutually. I need to know about every collided box, so the pesimistic
> complexity is O(n^2) (which gives me 1600 collision checkings when
> models have 40 box colliders) - i've tested that doing collision
> detection so many times is VERY expensive.


Make sure that each of the humanoids are in their own dHashSpace 
space. HashSpace<->HashSpace collision is reasonably efficient, if 
you tweak the parameters right, because you won't ever get the O(n2) 
tests like you would if they're all in a single space, or each in 
a dSimpleSpace.

Cheers,

			/ h+




More information about the ODE mailing list