MultiSAP aabb overlap test

[Ronin]

Hi, sorry if this has been asked before, but I didn't find any on this forum, if so please redirect me to the appropiate topic...

I am using the multi sap, but experienced this problem also with "normal" sweep and prune. In this test case I use

Code: Select all

btVector3 aabbMin(-1000,-1000,-1000);
btVector3 aabbMax(1000,1000,1000);

as extents of the world aabb. I divide into 10 sap areas on each axis. I have 2 objects, both the same trimesh, one located in the origin and one 500 units away.

When I look at the aabb overlap test in btMultiSapBroadphase::testAabbOverlap() it happens that both aabb's contain the exact same coordinates. It seems like the aabb in object space to me. Hence the test always returns true and the objects are not excluded from narrowphase no matter how far they are seperated.

Has anyone any idea what could cause this?

And just out of curiosity: What is this btDbvtBroadphase, which gets introduced in 2.69?

Thanks,
Ronin

[Ronin]

Nevermind, I resolved this issue. It was totally unrelated to bullet, sorry...

[Erwin Coumans]

And just out of curiosity: What is this btDbvtBroadphase, which gets introduced in 2.69?

btDbvtBroadphase is a broad phase based on dynamic AABB trees, one static and one dynamic. It was originally developed for Bullet soft bodies.

The update and change of topology (adding, removing, changing AABBs) performs very well, and it has the extra features: frustum culling capability for graphics rendering (so you don't need to keep two separate acceleration structures around).

Thanks,
Erwin

[Ronin]

That sounds really interesting, are there any benefits when using extremely large worlds, too? This post kinda suggerate it:

http://www.bulletphysics.com/Bullet/php ... f=9&t=2142

For example in comparison to the multiSAP broadphase, is btDbvtBroadphase able to handle worlds in similar sizes?