Collisions become unpredictable when changing localScale of a sphereCollider's GObject per-frame.

I am designing a little VR catching game in which the thrown ball changes physical size during its approach to the user. Because this is VR, the user is catching using a disc/flat cylinder that is a child of the motion-tracked VR wand.

Currently, rescaling of the ball is implemented by manipulating the game object's localScale within the Update() loop. The resize is a gain term which adds/subtracts a fraction of the ball's physical size on each frame. The change in size is uniform along the 3 spatial dimensions, and is not allowed to drop below 0.05 m, and does not grow beyond 1.5m.

The ball is supposed to stick to the user's hand-held paddle upon collision, and this is implemented as an OnCollisionEnter() reparenting of the ball gameobject to the paddle. When the resize manipulation is off, everything works beautifully. When the resizing manipulation is on, the collisions are unpredictable, and the ball often bounces away from the paddle before coming into visual contact with it.

The temporal resolution of the physics engine has been greatly increased to improve reduce issues related tunneling. I have tried all forms of collision detection mode, and this does not seem to help things.

Any input would be greatly appreciated!

Thanks.

 

Thank you for your diagnosis, Melv. I'm afraid I understand it only superficially, and not enough to speculate on possible remedies. Do you have an accompanying remedy to suggest? I have a few questions below that, if you have the time to answer, might help lead this horse to water.

"Physics doesn't run per-frame, it runs per fixed-update (by default)."

I understand that, but am not sure how it would impact the current problem. In any case, the issue persists whether the resizing method is called in Update() or in fixedUpdate().​

"if you size-scale-up colliders then you're likely causing overlaps which then the solver has to resolve, some of which won't necessarily happen instantly during a single simulation step but might take several steps, none of which are per-frame."

There are only two colliders in the scene: one on the ball gameobject, and one on the paddle gameobject. I'm a bit unclear on what an overlap is. I assume that the only overlap would be when the ball tunnels into the paddle, so correct me if I'm wrong.

Apologies, the language "per-frame" was clearly just not specific enough. I'll use more specific language (ie update, fixedUpdate) in the future. Are you suggesting that it will take several rounds of fixedUpdate for the solver to deal with overlaps?

The resizing of the ball should stop when OnCollisionEnter() is called, and the ball should then stick to the paddle, but does not. Instead, it bounces off, and often before the collision occurs, visually, as if the ball is bouncing off of a force-field. Could this issue you're raising also be the cause of the failure to stick to the paddle?


​

Thanks again!