hi, we found that when in slomo mode, such as slomo 0.01, the stacking results(3 cube stack) are unstable, will introduce a big accumulated impulse which cause the cubes separate each other. It happens for geometry collection also.
You can test by stacking 3 cube and set sleep family to very low(mulltiply 0.01).
before slomo, [Image Removed]the accumulated impulse:
[Image Removed]
the contact net push out and net impulse:
[Image Removed]
after slomo 0.01: [Image Removed]a big increase of accumulated impulse, with net impulse and netpushout/dt
[Image Removed]
[Image Removed]
is there anyway to solve this? it seems pbd based algorithm will causing this inevitably as well as precision issue.
[Attachment Removed]
Hi Junhe,
How are you implementing this slowmotion side? I think this is something we may have fixed in a more recent version - if you can send me a repro I can verify that and see if there is a specific change you can take
Best
Geoff
[Attachment Removed]
Hi Geoff,
Steps to repro:
1、drag 3 cube to stack each other, [Image Removed]enable simulating
[Image Removed]set the sleep threshold multiply to 0.01(to avoid sleeping) [Image Removed]2.play in editor,
at first, the result is stable [Image Removed]after a while, enter slomo 0.01 in the console command, you will see the three cubes begin seperate each other [Image Removed]
[Attachment Removed]
Thanks! I’ve repro’d it on my end so I will reach out to dev and see if this is on their radar.
[Attachment Removed]
This looks to be a result of passing in a tiny time delta to the simulation (which we’d probably want to keep for the final iteration step - otherwise we’d lose energy) - but when we end up with time slowing down loads like in the case here, it ends up constantly ticking with a very small delta (which could blow up for any 1/dt type operations).
I’m verifying this at the moment, and then I’ll have a chat with dev for next steps
[Attachment Removed]
Hi Junhe, and apologies for the delay on the response to this.
It looks like my advice above wasn’t actually correct. In this case, the jump from one timestep size to anything greater than around 50X difference will trip up the contact solver.
I’ve played around with it a bit, and if you set a larger step as a compound of say 10X jumps, the simulation remains stable.
Best
Geoff
[Attachment Removed]
No worries Junhe, apologies again for the delay!
Geoff
[Attachment Removed]
just using console command: slomo 0.01.
I have tried this in 5.6 release, the same problem
[Attachment Removed]