[Feature Request] Soft-soft collisions and self-collisions

[josephrocca]

Hello, I'm very excited by Jolt's soft body support! Here it is mentioned:

Soft bodies can only collide with rigid bodies, collisions between soft bodies are not implemented yet.

I'm wondering if soft-soft collisions (including self-collisions) are on the near-term road map?

---

Please ignore this if there's not really anything interesting to say/compare here, but: Tangentially, in terms of trade-offs relative to Jolt's approach, I'm wondering if anyone knows how MuJoCo's new "flex" based soft bodies work? I don't have much experience here, but it doesn't look like XPBD?

• https://github.com/google-deepmind/mujoco/tree/main/plugin/elasticity
• https://mujoco.readthedocs.io/en/stable/modeling.html#deformable-objects

MuJoCo has also had a "rigid body composite" thing for a while - basically lots of rigid tetrahedrons connected by springs to emulate a soft body IIUC. Whereas with this new "flex" stuff the tetrahedrons (or lines or triangles - "simplicial complexes") themselves are squishy, so you can use fewer elements to achieve a similar result (though overall performance for same "effective fidelity" - in terms of FPS - currently seems similar-ish, but I haven't done extensive tests).

[josephrocca]

Also I haven't tried creating my own models yet, but think MuJoCo uses (or requires use of) gmsh and I also saw this comment:

To convert from tri-mesh to tet-mesh we used fTetWild.

and they end up with a tet-mesh that has more fidelity closer to the surface of the model, like this:

I'm wondering if an example/sample could be made showing a comparable process in Jolt? I.e. starting with a surface mesh (or an SDF, perhaps) and ending with a simple demo that drops a couple of them on top of one another (once soft-soft collision is supported).

[jrouwe]

Soft body physics is still in development but I haven't had much time to work on it lately.

Soft body vs soft body is something that I eventually want to support, but there are a lot of other things that have higher priority such as:

• Further optimizing and parallelizing the simulation
• Optimizing collision tests against soft body (= currently O(number of surface triangles))
• Collision listener callbacks
• Implementing missing interfaces like AddLinearImpulse etc.
• Supporting buoyancy

I'm wondering if an example/sample could be made showing a comparable process in Jolt?

I have put this on my list.

Tangentially, in terms of trade-offs relative to Jolt's approach, I'm wondering if anyone knows how MuJoCo's new "flex" based soft bodies work?

I haven't really looked into this, so I can't say much about this.

[ThomasCartier]

I literally gave it a try and was really happy with the result so far

It's complicated to see through, but yeah, past self collisions which are lacking, the animation looks beautiful.
Well done; really.