Refactor contact models and add tests

[diegoferigo]

I wanted to polish and uniform the great work done in the last few months about contact models. Likely, we need to keep working on them to optimize their performance, I took advantage of going through them more carefully than my previous reviews.

I have some doubts on few details of the RigidContacts and RelaxedRigidContacts. I will comment my edits in this PR to discuss my doubts and proposed changes.

---

📚 Documentation preview 📚: https://jaxsim--260.org.readthedocs.build//260/

[diegoferigo]

@flferretti I suspect that the usage of the spatial inertia of the link $\mathbb{M}_L$ is wrong here. Accordingly to the following resources, it seems to me that Mujoco initializes it to $\text{diag}(J^\top_{W,C} M^{-1} J_{W,C})[0:3, 0:3]$ computed at $\mathbf{q}_0$ (default configuration from the URDF). It's not a big deal now, let's keep in mind to validate it as soon as possible.

• https://github.com/google/brax/blob/865d974c0285d345425d76df30c38614c828c8c8/brax/generalized/constraint.py#L169
• https://github.com/google/brax/blob/865d974c0285d345425d76df30c38614c828c8c8/brax/base.py#L316C5-L316C14
• https://github.com/google-deepmind/mujoco/blob/5e989d18c349e914fb82ae937006ff4832b011ce/src/engine/engine_setconst.c#L87-L100

Note that I didn't fix it here in this PR. We should replace M_L with something else after some investigation.

[diegoferigo]

Delayed to a next PR.

[diegoferigo]

In this PR, I also added some simple test of all contact models so that we at least know that we won't break anything related to them in the future. Regardless, since I touched some internal details of the RigidContacts and RelaxedRigidContacts, please @flferretti @xela-95 during your review make sure that your scripts still work as expected.

[flferretti]

Just an early comment, would it make sense to include in this PR avoid repeating the contact_params attributes in JaxSimModelData and in ContactModel?

[diegoferigo]

@xela-95 w.r.t. to the original PR, I updated the direction of action of the baumgarte stabilization. Now it produces a force towards the terrain's normal, and this behavior is aligned with RelaxedRigidContacts. While reviewing, do you mind running the walking script with Baumgarte stabilization enabled? I'm curious to see if it works well now.

Just an early comment, would it make sense to include in this PR avoid repeating the contact_params attributes in JaxSimModelData and in ContactModel?

Let me think about it. Having nominal parameters already defined in the contact model itself is quite useful, but the dynamic override from data is not super intuitive. I'm in favor of this change, but I need to play a bit with it before proceeding. I'll try to do it in a new PR.

[flferretti]

Thanks for working on this! I left a couple comments

[diegoferigo]

@flferretti @xela-95 I removed from this PR the new logic of the RelaxedRigid contact model, only leaving some minor refactoring. I need some more time to investigate how to properly update it, as there are behaviors that do not sum up. The comodo notebook we were discussing f2f should work with these changes.

There are few PRs that depend on this one. I want to merge these changes and propose again the new logic of the RelaxedRigid contact model in a new one.