FLIP substeps for fast moving boat

[dleonhardt]

Hello dear odforce users.

I have a question regarding substepping with fast moving colliders.

I have a catamaran speed boat which is driving in a straight line at roughly 60km/h. I added some procedural animation to the boat with the help of the ocean spectrum, that drives the wavetank, and CHOPs. Currently the simulation uses 6/8 Substeps on the FLIP Solver and the Solver has access to the sub-frame collision geometry.

The problem I'm running into is that the trail left by the boats keels has visible stepping in position and velocity, which causes tendrils to shoot up in the boats wake. This doesn't look good together with motion blur.

Should I increase substeps further to get rid of the stepping by brute force?
Work out a whitewater setup to mask this behavior?
Or is there anything else you can recommend in this case?

I'm thankful for any suggestions.

P.S. Unfortunatly I can't provide any scene files

[jimeng20]

Hi there,

Just curious what do you mean the Solver has access to the sub-frame collision geometry? Have you tried to use a trail sop to trail your animation yet?

If not, I would try to create all the sub-frame animation as much as possible before completely relying on solver's substep. Based on the speed, put down a timeblend and then set something like Trail Length 10, Trail Increment 0.1, then convert it to sdf.

Hope this would help.

[dleonhardt]

1 hour ago, jimeng20 said:

Just curious what do you mean the Solver has access to the sub-frame collision geometry?

 

What I mean by that is if I go through the timeline in 1/x increments, my collision SDF is moving accordingly inside the DOPnet.

I set up my collision object similarly to this quicktip: 

But I will still try out your suggestion with the timeblend. Thank you.

Edited December 13, 2019 by dleonhardt

[jimeng20]

I would trail the collision geo and convert into sdf in the sop, cache out and you can either use it as a proxy volume in the static object or, you can create colliison and collisionvel and use in the dop as collision fields.

If you are going to try the collision/collisionvel, make sure you wire the volume source to your THIRD INPUT of the flipsolver instead of the forth.

Basically, only rely on the substep in the solver is very expensive, the dop is trying to figuring out the sub-frame information during the sim, and apparently, you can prepare the sub-frame info in the sop, cache out, and dop just need to read it in. 

It will be much faster if you have trailed your collision animation 1/5 increment and put substep 2 in solver than try to calculate sub-frame while siming with 10 substeps.

Hope this would help.

 

[dleonhardt]

Hey Jie, 

unfortunatly your two suggestions didn't solve my issue. But I'll keep the collision/collisionvel method in mind, as it seems to be very fast and works very well with slow moving colliders.

Still thank you for your time.