How to use OdeQuadTreeSpace

Secator · May 30, 2009, 10:21am

How to use OdeQuadTreeSpace ?? What should be passed to the constructor, and is there any example of using that?

rdb · May 30, 2009, 10:24am

Notnasiul · June 3, 2009, 10:42pm

Hi!

I’m playing with ODE too. I managed to use a trimesh as a colliding surface, where many (ode) spheres interact. However, I felt a little bit dissapointed when framerate dropped to less than 25 with only 30 objects. I know that trimeshes are tricky but… only 30 objects?

Then I realized I was using the SimpleSpace… so I tried to use QuadTreeSpace instead, and I found this thread.

Well, according to ODE manual:

ode.org/ode-latest-userguide.html

dSpaceID dQuadTreeSpaceCreate (dSpaceID space, dVector3 Center, dVector3 Extents, int Depth);

Creates a quadtree space: center and extents define the size of the root block. depth sets the depth of the tree - the number of blocks that are created is 4^depth.

So I tried in Panda with

self.space = OdeQuadTreeSpace((0, 0, 0), (2, 2, 2), 3)

and it worked… but the framerate didn’t rise. I though the extents vector was clearly too small for any world, so I changed it to (20,20,20), not really knowing if the scale is correct.

After a few frames, ODE crashes The same if I use 6 depth levels. No way.

Are ODE spaces really working properly on Panda? :m Any known working example? Thanks!

rdb · June 4, 2009, 6:53am

Yes, they are working properly.
Performance problems can be solved by using multiple spaces and reparenting spaces to others.

lethe · June 4, 2009, 7:16am

Also, try the hash space - its easier to setup than the quadtree space (Default parameters are usually fine.), and copes with height also - if your level is mostly vertical the quad tree space will do little.

Notnasiul · June 4, 2009, 8:18am

Thanks! The HashSpace works much better. Now I get around 80 spheres before framerate drops below 30.

I’ll play a little bit with HashSpace levels. This information from the ODE Wiki may be useful for anyone using Ode:

How do I come up with the optimum values for the parameters to dHashSpaceSetLevels?

They just really depend on the typical sizes of collidable geom (groups) in your space. Set minlevel and maxlevel such that your full range of geom sizes are encompassed. Say that your largest geoms are of size 41 units (in X or Y or Z dimension) and your smallest are 0.25 units in size. Then your dHashSpaceSetLevels? call should be: dHashSpaceSetLevels?(space, -2, 6) to set the min and max cell sizes to 2^-2 = 0.25 and 2^6=64 (2^6 being the smallest power of 2 that ‘41’ will fit in). It’s more important to get the max right than to get the min right, assuming that your tiniest objects don’t all clump together in the bottom of a hole. You can be over-generous with these values and just suffer a smidgen of time+space overhead. But if you’re too stingy, your collision times start to look decreasingly like O(n) and more like O(n^2).