I'm seeing the same problems in a project. Managed to characterize the issue narrowly enough to report it via Feedback Assistant, since I'm not sure it's been reported yet (FB21197208). Seems to be two main issues to me: Physics simulation stops moving objects during collisions when their velocity is still high enough that it looks quite unnatural (relative velocity around 2.0 or less). Even with everything set such that the collision should produce 100% elasticity, it doesn't. Effect is worse with a box-shaped collider, but still happens with sphere-shaped colliders. Combination of the two means repeated collisions will eventually lead to objects stopping, regardless of the 100% elasticity. Note: I haven't tried the scaling workaround yet, but it sounds like that might just be putting off the issue (more collisions needed to lose enough energy to stop), and I'm starting to think implementing my own movement system would be better (I don't actually need mass/inertia and other true-to-physics behavior, just perfect elastic bouncing).