Sticking contacts with btConvexHullShape

[peterk]

Hi,

I'm using a voronoi cell based shatter tool to create a large amount (1000) of stacked convex shapes with btConvexHullShape as the collision geometry. I'm finding that while fairly stable for stacking, the resulting bodies have a tendency to stick together. Only a large direct collision impulse will separate them, creating unnatural looking results.

In contrast, I've found that ODE augmented with the Bullet GJK collider behaves quite differently, with the bodies separating easily after an impulse unless otherwise constrained.

The effect is reduced if friction is set to a very low value. Does anyone know would be the likely cause of the very different results?

Cheers.

[Erwin Coumans]

I've not seen such sticking behaviour.

Bullet constraint solver and ODE dWorldQuickstep are pretty much identical (aside from improvements such as warmstarting and SIMD in Bullet), or are you using dWorldStep?

Can you reproduce it in one of the demos and share it? One option is to export the scene to a binary .bullet file and attach it to this topic.
Thanks,
Erwin

[peterk]

No, I'm using QuickStep in ODE.

Unfortunately my scene doesn't seem to work with dynamica (or the Serialize Demo), looks like the inertia tensor is being calculated wrong. I'll try debugging this when I get some spare time and get back to you.