Vertex Block Descent

[RT2Code]

Hello,

A new promising physics solver was introduced recently in the latest siggraph conference: https://ankachan.github.io/Projects/VertexBlockDescent/index.html

What do you think about it? I suppose that replacing XPBD with it in Jolt would require a tremendous amount of work, but do you think it is something that could be done in the future?

[jrouwe]

Hello,

I am aware of this paper. I need to do some investigation before I can tell if this would be an improvement for Jolt or not. I think it does indeed mean a substantial refactor.

As far as I can see now, the way the VBD solver works is really beneficial for GPU, but since Jolt currently runs on CPU it has much less problems with the 'vertex coloring' issue as mentioned in the paper. Also, I think the solver loop requires calculating some quantities per vertex instead of per constraint which XPBD does (edge constraint = 2 vertices, volume constraint = 4 vertices), which again is fine on GPU but less so on CPU.

[AnuraDev]

bump. right now i am searching for a solution to simulate simple cloth which should be purely visual and scale decently well for number of instances, some compute shader solution seems like best for this, cpu solutions are imo too expensive just for visual stuff like this. (ex. rts game with units who wear capes, collision should be probably just with the primitive shape of unit) right now i am not sure what something lightweight/performant like this exists (at least open source).
so the "question" is, are there any plans for gpu soft body simulation in jolt? (the gapi probably will be "problematic" but hlsl should be ok cause you can port it relative easily)

[jrouwe]

Yes, eventually I would like this (and other parts of the simulation) to be able to run on GPU. But this is a lot of work so will not land any time soon.

[AnuraDev]

yeah thats obvious, but good what it's on the roadmap. jolt ftw