Collisions detected too late

[Lollo12589]

Hello,

I´m still very new to BepuPhysics and I´m experiencing an issue with collisions detected too late. The scenario is very simple. I only have a static environment - meaning there is no movement in my bodies - and I set positions of the objects externally via the RigiPose. The simulation only includes kinematic bodies since no physical forces should act on them.

The shapes are either convexHull or BigCompound. Collision detection between two ConvexHull bodies work flawlessly. Collision detection between two BigCompoundBodies also works perfectly after I increased the speculative margin a bit. However, I'm having problems with collision detection between a BigCompound and a ConvexHull.
In that case, collisions are detected too late. The BroadPhase does indicate potential contact, but the NarrowPhase still rejects it even though a contact already exists. What could be the reason for this, and are there any other parameters or thresholds I can adjust?

In my case, very sensitive collision etection is crucial. I'd rather have a contact detected that isn't actually there than miss one that is.

I am grateful for any help :)

[RossNordby]

(sorry for slow response, seen this and will respond later when I am not being eaten by workmonsters)

[RossNordby]

I exist!

The BroadPhase does indicate potential contact, but the NarrowPhase still rejects it even though a contact already exists.

Not sure I understand; the narrow phase has generated a contact point for a child pair, but then filters it out? The only filtering that should happen is speculative margins and then contact reduction. Contact reduction only kills of contacts beyond the number that can fit in the manifold, so it should never result in a total lack of contacts. Speculative margin filtering operates at the level of convex pairs; convex pairs don't report contacts with negative depths with magnitude larger than the pair's speculative margin.

I'll probably need more information or a repro to narrow it down; there's no particular reason I'd expect that to happen under normal conditions. Random guess: Did the compound have its children modified without having its Tree acceleration structure updated to match their new state?