is this still the case? sphere collider are more effecient than cube collider?

http://forum.unity3d.com/threads/34254-Capsule-vs-Box-Colliders

 

Its very likely to be still true (by that I mean it is but I'm too lazy to write a test to prove it). But the chances that you are doing something where the difference would matter are slim.

 

Yes because a sphere has 1 side whereas a cube has 6 sides.

 

AFAIK none of them have any faces but instead are just maths, with the sphere it probably just does some fanciness with the radius and whatnot when the cube is more complex (And the capsule as well).

 

Spheres probably check for the distance towards the center, if it's less than radius then it's colliding.
And cubes check if the point is between min and max bounds, for x, y, and z axis.

Are you going to do many collisions per frame?

 

What? I had no Idea. I'm going to test this tonight, cause the thing I'm working actually needs a lot of colliders.

 

Spheres/circles have pretty much always been faster than rectangles because, as said, you can just check the distance from the center of the sphere vs its radius... rectangles can take up to 4 tests to see if there is an overlap, or even more complicated when it rotates.

 

Right...unless math itself changes fundamentally somehow, sphere/circle colliders will always be faster than cube/box colliders, forever.

--Eric

 

*ponders if a signed distance representation would help the poor cubes and other non-spherical objects*

 

There was a technique used on GPU driven particles to speed up collisions between cubes, which worked essentially by treating each edge of the cube as a capsule, so you basically ended up with a rounded-cube collision "mesh". Capsules are much easier to do collision checking than cubes also (though not as much as spheres), but it was pretty efficient compared to actual box collision.

The obvious drawback is that not all cubes have rounded edges, and maybe rounded edges won't work for your needs. Might be worth a test with compounded capsules against the cube collider to see what wins out though. The GPU method was pretty optimized at a shader level though, so maybe the overhead of having 12 capsules will be worse than having the one cube if you just do it the 'easy' way.

Cheers

 

very interesting. i had never thought about it that way (or at all really). seemed like since cubes are flatter they would be easier, but this makes perfect sense now that i think about it.

 

Pardon the necro, but are there any calculations or estimations regarding the actual overhead difference between sphere and cube colliders?

 

Think of it as this:

A sphere only (in some cases) needs to check if the distance to a point is larger than radius. One check and a distance calculation.

A box in 3D needs to check if min and max bounds are bigger or smaller than each other in all axis. Up to six checks.

However it is not always that easy. It depends on what type of form you are check against each other. Checking a sphere vs a sphere is not the same as checking a sphere vs a box.

With more complex forms and rotation comes more checks. Like separating axis collision detection.