Eelco Hoogendoorn

If I zero out the friction local var in OneWayCollider _position_contact, I seem to get unbreakable static friction. That strikes me as the opposite behavior from what it should be;...

In one of the recent PDB papers they argue that its best anyway to leave the iteration count of that loop at 1; and throw all compute at the outer...

Also one thing I dont really understand; whats up with the 'performs two physics substeps per collision update' in _pbd_step? Is that a manual unrolling of the solver loop, effectively?...

> Alright, elasticity should be behaving correctly now. There was a bug making it basically always be elasticity=0. I hadnt paid attention to the elasticity yet. Friction still seems funny...

I do not think it has to do with multiple contacts; it shows up for a single one. Ive been thinking about how to make this into a small reproducible...

  Here is how a spinning box thrown on the ground behaves for me; not sure that either pbd or legacy is without quircks; but they sure are different....

Hmm I guess my moments of inertia are artificially inflated in the above example, contributing to a sense of irrealism (and the y-z dimensions of the cube are swapped in...

To be more clear about my comment above; these two lines still appear inconsistent with one another, wrt to the appearance of the friction variable in only one of the...

Some additional findings; there are various ways to trigger instabilities in the PBD simulation. Lower inertia values tend to be less stable. Blocks of styrofoam with a metal ball inside...

Looking around for problems (and not finding root causes thus far) another nit to pick: https://github.com/google/brax/blob/730e05d4af58eada5b49a44e849107d76e386b9a/brax/physics/colliders.py#L330 This line strikes me as undesirable compared to the two-way equivalent, in the sense...