Stacking cylindrical models keep shaking and jittering indefinitely - Chaos Physics

[Armitage1982]

Hi,

I am currently working on a project that uses Havok for physics.
I would be interested in switching to UE to open myself to more opportunities.

However, my few attempts with the Chaos Solver are not particularly convincing (close, but not enough).

My problem is :
The stacking of cylindrical models continues to jump around indefinitely. It doesn’t matter what combination of parameters you try (and I’ve tried many).

Physics on Objects :

Project Physics Settings :

As you can see, I even try ridiculous Iterations values or cheat with Mass & Inertia Tensor Scale but without success.
This does not seem to happen with simpler models like spheres or boxes.

Is my object not thick enough or is it an energy conservation problem?

My simulation must be able to handle 2000 cylinders, thick enough to provide good depenetration, but slices at least 3~4 times smaller than the radius.

As the size has an impact on many physical engines it is the environment, mostly composed of static elements, that will be adapted.

The most important feature is STACKING. The resting phase must be stable and one must be able to stack dozens of Rigid Bodies without shaking.

The performance must be the best possible for real time. To give you an idea I’m under 2 ms per frame on most situations with Havok (which is incredible).

Would someone be able to confirm that these needs can be met with UE5 Chaos Physics?

Any feedback, video material or documentation about Chaos Physics is also welcome.

Thanks

[Armitage1982]

I tried with the recent updates and even disable the Sleep with :

p.Chaos.Solver.SleepEnabled 0

But nothing helps.

I wonder if, as with Jolt Physics, the cylinder would simply be a shape that offers the least stability in the absence of a cache system for the simulation.

If someone has an idea, I’m still interested…

[CedUE]

Hi
Have you tried a non uniform inertia scale using a larger value for the thinner axis ?
I would also suggest playing with the convex margin parameters ( assuming you are using convex shapes ) to see if that helps

[Armitage1982]

Hi,
Thank you very much for the suggestions.

The shapes are indeed convex so we can play on these parameters.

Reducing the margin parameters seems to restore a better fidelity to the collisions but the problem still persists.
I have tried what you suggest for the non-uniform scale value for the thinner axis, but the phenomenon does not seem to be related to this parameter and does not correct the problem.
Also, whether the shape is uniform in scale or not, it’s all the same.

I also tried different simulation scales, but in order to get the desired simulation I have to vary the gravity force which gives the same result. The accumulation of successive forces prevents any stability.

[Hawk_UK]

As far as I understand it, isn’t this simply an issue related to trying to do something the default physics sub stepping settings don’t produce an accurate enough solution to resolve this particular need?
There is a setting in the details panel to allow and increase the number of sub steps to get a higher resolution to solve these more detailed physics issues more accurately, but of course by increasing the substepping it also uses more resources, so you have to test and balance the use of the substepping for your projects needs, as far as I understand it?

Just a thought… I hope this helps you, Armitage1982.