Real-Time Physics Simulation Forum

Re: bullet/dynamica and deterministic results

2012-02-18T21:11:59+00:00

Thanks for the update! Very cool. I look forward to trying it out next time I've got a gig requiring Bullet.

Statistics: Posted by meeotch — Sat Feb 18, 2012 9:11 pm

Re: bullet/dynamica and deterministic results

2012-02-18T08:53:28+00:00

It took some years before I sat down to fix the code, but Dynamica is now deterministic across different runs (just make sure to reset to the first frame every time before simulating)

It should be fixed from revision 139 onwards, http://code.google.com/p/dynamica/source/detail?r=139

Note that under Linux you can use the Extras/MayaBullet/Makefile, and for Windows you can use cmake
Thanks,
Erwin

Statistics: Posted by Erwin Coumans — Sat Feb 18, 2012 8:53 am

Re: bullet/dynamica and deterministic results

2010-03-29T15:54:59+00:00

Thanks!

Obviously, this issue is a stumbling block to using Maya Dynamica in production situations, where one needs to lock in animation for approval. (You can still bake the animation, obviously - but you can't use the usual tweak-playblast-tweak cycle without also caching each iteration, then quitting & restarting when you want to tweak again... which is a bit unwieldy.)

Statistics: Posted by meeotch — Mon Mar 29, 2010 3:54 pm

Re: bullet/dynamica and deterministic results

2010-03-25T20:08:00+00:00

Bullet is deterministic, as long as you reset all values properly when you restart.

We don't do this at the moment in Maya Dynamica, so the simulation is different each time. I filed an issue here issue right here, so you can keep track of the progress.
Thanks a lot,
Erwin

Statistics: Posted by Erwin Coumans — Thu Mar 25, 2010 8:08 pm

bullet/dynamica and deterministic results

2010-03-12T22:16:16+00:00

I've got a fairly simple file that shows a high degree of non-determinism when run. (Maya 2009-x64 on CentOS 5.4, Bullet 2.76.) It's a shattered object, with the individual pieces having initial pos/rot/vel/spin set, plus a ground plane. I didn't see an option in the solver for randomization - perhaps it's not exposed?

Is there something I'm missing? I was under the impression that Bullet was deterministic. Is there something else I need to do to reset the solver between successive runs? File is attached - run it multiple times, and watch where the pieces end up... not even remotely in the same position.

testBullet2.tgz

Statistics: Posted by meeotch — Fri Mar 12, 2010 10:16 pm