Hi, I don’t know if this is a common occurrence here, but it turns out that we have to simulate physics on character animations that last longer than several minutes (3 to 7 minutes).
From frame 10000 onwards (at 60 fps), the simulation starts to make a mess of things, and for the time being we’re obliged to restart the simulation in increments of -10000 frames.
Has anyone here ever encountered this problem and knows its origin?
Much thanks.
Hi @Antonin_Delboy, welcome to the forums 
This is certainly uncharted territory on our side, we test our software with 100-200 frames; the typical length of film and cinematic performances. I can’t think of a reason why 10k+ would be a problem, but it’s likely there is something we’ve overlooked.
What does memory of the Maya session look like at these numbers? Could it be running out of it? We store the simulation at every frame in intricate detail, such that it can be rewound later, and it’s possible it’s simply eating up your RAM.
Can you confirm whether the simulation runs fine as you record? Namely, is it the conversion to keyframes that break, or the physics? During record, you’ll find it runs through the Maya timeline twice, the first time is simulation and the second time is conversion to keyframes.
I’ve run a quick test now with 20,000 frames using 2024.07.01 on Windows and was unable to reproduce the issue.
Can you try this same scenario on your machine?
In the above example, I also added a turbulence field and animated it along Translate X to give the entire duration a randomised look.
Finally, could you describe what it looks like once it breaks? Do positions end up in the wrong place, do keyframes turn into stepped keys? Is scale affected?
Thank you for taking the time to investigate. We conducted the suggested test, but with over 10,000 frames at 60 fps, rather than 24 fps, which seems to be the frame rate of your video.
We encountered a corrupt result during the first simulation pass, not just during the key recording process. I can provide you with a file where we performed three tests across three layers. As you’ll see, with the dead ragdoll test on the animation layer named “rSolverLayer1Test3,” the issues start appearing after frame 9500 (60 fps).
Much thanks
zip file available here : https://we.tl/t-dNUZM1Vg25
Thanks, I’ll look into this next.
Playback and recording are independent of each other, so I don’t expect the playback rate to matter - 10,000 frames is 10,000 frames, no matter how quickly you play through them.
To confirm, I did a quick recording with playback set to 60 fps, and still was unable to reproduce any issue. Can you try recording the Manikin like in my example? Do you still experience the issue?
Regarding your file, we are unable to open .mb files, would it be possible to upload the file as an .ma?
Hi, here is the test in ascii format if you want to have a look.
Thanks
https://we.tl/t-Np5v5gHGrA
Thanks, here’s what we’re looking at.
Base Layer
There is some animation around the 1,000 mark, the simulation would attempt to follow this.
Layer 1
Something is missing in this scene, I’m not seeing the Pin Constraint used to pull him around. But presumably, you’ve animated a pin constraint and added a field to your scene to make him move around, all appears well so far.
Layer 2
Here I need clarification; did you leave Layer 1 enabled when recording? If so, what result are you expecting? Whatever the animation looks like at the time of simulation is what the Markers will attempt to follow, given their Rotate Stiffnesses.
In this case, it looks like the simulation is run again, using the same Pin Constraint, but this time using the recorded simulation from last time as animation input.
Layer 3
Where the problem begins at frame 9,500.
However, if the previous layer was enabled at the time of recording, the Markers would again attempt to follow this animation, which explains why he suddenly starts moving at frame 9,500. He’ll be moving the whole time, except the Rotate Stiffness isn’t strong enough for it to have an effect.
Can you tell me more about these 3 layers, and whether you expect simulation from a previous recording to influence the next recording? I don’t see anything technically wrong here, the simulation is following your animation in each layer, for the full 20,000 frames.