[ODE] Octree space vs hash space vs higher level solution

[Gabriel Gambetta]

I have a car driving over a terrain (one Landscape geom) which also
contains track pieces ("straight", "curve"), where the collision geoms
of each piece is a set of boxes. There are also other cars driving
around.

I don't want to perform too much collisions - for example a curve road
segment has 33 collision boxes.

I think I can put each car's geoms into its own Simple space, put the
geom boxes of *each tile* in a separate Simple space, and on each
timestep, calculate the tile where the car is, and collide the car space
with the terrain geom, the simple space with the "closest" track geoms,
and with the other car(s) simple spaces.

Does the Quadtree space, which is "exceptionally quick for large amounts
of objects in landscape-shaped worlds", have any advantage over this
method? Since I have higher-level knowledge of the scene, I believe I
can cull much faster than the Quadtree (finding the space which has the
relevant boxes involves two divisions). Besides, I think I tried using a
quadtree or hash space before and some collision functions weren't
implemented for some combinations.

Thanks in advance,
	--Gabriel