Meshed FLIP sim as deforming "solid" RBD collider--not working as expected

[bentway23]

I have a meshed FLIP sim (cached out, .bgeo files read back in) that I want to use as a "solid" object and have a rigid body rest on top of it (the idea is that something solid "ripples" psychedelically and the objects resting on top of it move as the ripples go underneath--but the original object is still a solid, and the objects don't sink into it). I just can't get it to hold up, however--when the RBD object is initially "dropped" onto the fluid mesh, it bounces and settles as expected, and does okay with simple early movement, but eventually the collisions become less and less precise and the RBD object just sort of sinks in. How can I get the mesh to read as just a run-of-the-mill deforming object?

I've tried RBD solver, bullet solver (obviously doesn't work with the shape of the mesh), volume and surface collisions, jacking up collider volume resolution to make sure all the little waves are getting captured, and boosting the substeps, but the result has been pretty consistent (except more substeps = super jittery RBD as it's reading all of the little pre-big-movement mesh ripples). On the attached setup (a simpler version of the project giving me trouble) I even ran the meshed surface through a "clean" just in case any of the lingering velocity attributes were throwing it off, but that didn't help either. I'm wondering if the fact that the topology itself changes so much might be the problem, but if so I'm not sure how to fix that.

Thanks for any suggestions!

flipmeshtest.mp4

meshtest.hiplc

[jamesr]

Hi, this effect sounds neat. I don't know if you'll have much luck with the rigid body solver in this case the way you've got it (I could be very wrong though so anybody please chime in!). Shuffling your DOP setup around and changing the density of the RBD object seems to do sort of what I think you were saying? The torus is affected by the FLIP sim, but in a rigid sort of way. Otherwise, I'd be looking for ways to transfer the motion of the flip sim to the torus in SOPs post-sim.

 

EDIT: Also if you're getting any jittery stuff on your object you wanna smooth out, try just using a little bit of a lag or filter CHOP on P and blend it back in. It can help with some small stuff.

 meshtest_jamesr.hipnc

Edited September 14, 2019 by jamesr

[bentway23]

Thanks for your reply! Unfortunately, the FLIP is, I'm guessing, partially incidental--what I'm looking for is a rigid torus dropped on top of a deforming mesh--I don't want any fluid animation. I included the FLIP elements as I'm assuming that somewhere in there is the source of why it's not working.

Also, in the "real" project, the fluid-like solid mesh is done, big and dense and pretty, after a four-day sim. I am out of FLIP-land and into crass ol' geometry-land now.

I did mess with the density of the RBD torus, but that didn't really do anything as the collider itself is a non-density object.

 

The quest continues. (Your example was great guidance on combining FLIP and RBD, though!)

[thebrettwalter]

Done something kind of similar sounding in the past but I didn't go with flip. I don't have houdini open right now, but off the top of my head, you could try ripple solver on a grid and then extrude volume the cache. If extruding your rippled grid is messy you could ray a highish poly grid onto the cache for tidier collision surface geo, then volume extrude the rayed grid and cache (You may not need to extrude if your polys allow for surface collision instead of volume on your static object dop node). Plug it into your sim as a static deforming object. You could then popsolve points down from above which land onto the ripple/undulating surface, which instance packed rigid body objects. If you definitely need flip though, this approach may not be so useful.

[bentway23]

Using the rayed and then volume-extruded grid in addition to bumping the substeps way up (5 min/10 max), using volume collisions and the RBD solver are doing the trick on my quick test mesh. Haven't tried it on the final big guy yet--I'm guessing more substeps will be in order.