Volume Based Collision Fails On Human Figure?

[Atom]

Hi All,

I am working on a piece where I want the feet of my character to interact with water. This will be a focus of the piece so I would like fairly accurate collision. I have exported my character as a .bgeo.sc sequence and I use that as collision geometry for FLIP. When I try to increase the detail on my collision mesh I get terrible spikes. At the setting of 192 I have a clean midres collision mesh but when I push it higher to 256 spikes start to appear. I have tried Raytrace, Volume and Meta Balls mode types.

Is there another way to construct hi-res collision geometry?

[stickman]

Might it be that the spikes are from insufficient resolution in the source mesh? Maybe try a subdiv node before the convert to volume?

[Atom]

Thanks, that does help. I can go a little higher on divisions.

Edited December 31, 2016 by Atom

[Skybar]

Making a VDB and using that as a volume sample is usually both more stable and faster. Also if you only need the feet to collide the whole upper body is essentially unnecessary to include. 

[chrisdunham95]

I would second what skybar/david has said above - volume based collisions/vdb cached to disk much faster espicially with flip! 

Edited December 31, 2016 by chrisdunham95

[Atom]

But isn't that essentially what the static object is doing already? It is creating a compatible volume based upon the supplied geometry, which in this case is a deforming .bgeo.sc sequence.

So you are suggesting I create a VDB sequence out of the .bgeo.sc mesh sequence? Then supply that to the AutoDopNetwork. What node do I use to supply the VDB sequence as a collision object?

Edited December 31, 2016 by Atom

[Skybar]

30 minutes ago, Atom said:

But isn't that essentially what the static object is doing already? It is creating a compatible volume based upon the supplied geometry, which in this case is a deforming .bgeo.sc sequence.

So you are suggesting I create a VDB sequence out of the .bgeo.sc mesh sequence? Then supply that to the AutoDopNetwork. What node do I use to supply the VDB sequence as a collision object?

Well yes, its doing that. But it is creating a standard houdini volume, and recreating that every time you resim. So instead you create a vdb which is faster than houdini volumes, and you cache that to disk so you dont have to recreate it all the time. In the end you'll save time and computing power. Its simply copying your vdb into the sim. 

All you have to do is set the mode to volume sample on the static object, scroll down and reference the volume sample path to your vdb. 

[ryew]

Hey Atom,

Attached is a quick example of how I typically cache and re-use VDBs for collisions in the AutoDopNetwork, this one using one of the built-in mocap models and a simple flip tank.  

As others mention, along with culling your geometry to only the necessary collision areas, referencing cached VDBs into DOPs tends to be easier and more efficient than wrestling with the automatic collision functions.

vdbcollisions.hipnc

[Atom]

Thanks for all the suggestions and the example file. I am going to try out these approaches when I get back to it. I guess I don't even need to go to .bgeo.sc. I'll just write that out as VDB instead.