I accidentally found something strange about RBDs and wonder if somebody here can take a look and try to make the simulation in the attached scene to somehow work. I am running out of ideas how to stabilize it.

The point is not to use constraints, but to rely on rbd collisions only.
Feel free to apply any changes to the shape and topology of the objects.
Just keep the “swing” effect.

The sim is pretty light.
… and breaks after the first 10-20 frames.

I did and few more tests ( different scenarios than this one ) with concave objects and got some mixed results.

Is there any reason why you're using a staticsolver for boxA/B instead of making them passive objects? I get stable results if i toggle off “Create Active Objects” for boxA and boxB, and increase “Resolve Penetration” on the rbdsolver (I set it to 3) and delete the staticsolver node.

Or wire boxA and boxB in to a merge and then a static solver.
Merge the static solver and the rbd solver in to another merge then apply the gravity after all that.

Resolve penetrations in this case is probably fine at 2 but works also works at 1.

Thanks.

I left the static solver in the scene, because it is from a more complex setup where it's needed.
I am wondering, is that some sort of bug, or the order of the nodes in this case matters ( opposite to what the help claims ) ?
I clearly see now why some of my previous tests didn't work. They were little chaotic and i didn't notice the pattern that if i have a static solver somewhere at the end of the networks the things get funky.

Once again - thanks for the feedback.