DOP RBD NaN problem

   4608   3   0
User Avatar
Member
1721 posts
Joined: March 2020
Offline
Hi,

I'm running a simulation of a few thousand RBDs (small shards – exploding windshield), and I realized that at the end of the sim almost half of my RBDs are “gone”. Upon further examination it turned out that the geometry point coordinates of the disappearing pieces became NaNs – this coming from the DOP network where these pieces' sim attributes became NaNs as well.

(The event that triggers this seems to be a collision, as on the same frame it becomes NaN-ned, an Impacts data also appears on the object)

I'm currently experimenting with various solver settings and whatnot that perhaps would get rid of this problem.

Anyone seen such a thing before?

cheers,
imre

Attachments:
NaN_WTF.png (123.2 KB)

Imre Tuske
FX Supervisor | Senior FXTD @ Weta FX

qLib -- Houdini asset library
http://qlab.github.io/qLib/ [qlab.github.io]
https://www.facebook.com/qLibHoudini [www.facebook.com]
User Avatar
Member
5 posts
Joined: Jan. 2016
Offline
Bumping this thread from spacetime but no one has really answered this on the forums.

Is there any ways to debug NaNs?

I'm in a somewhat similar situation, i'm pretty sure my collision objects which are growing and changing shapes are affecting the sim and blowing up forces (division by 0?), but I can't find a way to debug the scene(s).
User Avatar
Member
9384 posts
Joined: July 2007
Offline
If you are using POP Drag, POP Wind POP, Drag Spin or simply @targetv + @airresist your bullet sim is likely to be unstable and result in NaN's
So until that's fixed I'd suggest replacing them with wrangles that dampen v@v and v@w directly
Tomas Slancik
CG Supervisor
Framestore, NY
User Avatar
Member
1 posts
Joined: Feb. 2020
Offline
Hey

I just had the same issue and wanted to add, that in my case it was caused by particles getting a pscale of 0.
Changing this to a very small number instead of 0 fixed the issue.
  • Quick Links