Objects go through convex hull

[ThibaultLemaire]

First off, bevy rapier is great. It was easy enough to setup in Bevy, well documented, and provides pretty satisfying physics for simple collider shapes. So thanks a lot!

Now, I am building a convex hull directly from a single mesh exported from Blender.

I have attempted Collider::convex_hull() and Collider::convex_decomposition() and they both produce the same result:
A Collider shape that looks perfectly fine, but has one half where all objects fall through it.

The behaviour is a bit strange though, as objects seem to somehow react to the collider, as they start spinning pretty fast as they fall through.

The following clip will be much more telling than my clumsy words: (tiles have their RigidBody switch from Fixed, to KinematicPositionBased, to Dynamic as they are selected and released when clicked)

simplescreenrecorder-2025-07-27_18.59.35.webm

I have tried inputting the same mesh to Collider::convex_mesh() but that gives me None, without further explanation.

My shape is simple enough that I'm convinced it should be convex (I can provide a blend or gltf file), so what am I doing wrong?

(I use the same method for the tiles and they work perfectly fine with both convex_hull and convex_decomposition while put on top of a cuboid collider, or other tiles)

[ThierryBerger]

I'd bet on an edge case triggered by points being too close to one another, or triangles too "flat":

• WIP: Convex hull invalid topology parry#353 or QueryOptions parameters (GjkOptions) parry#298 may give you some insights.

In the meantime, if you want a quick fix, I'd try to make a simpler mesh: there seems to be a lot of long triangles which may be simplified ahead of time?

[ThibaultLemaire]

Thanks a lot, I will try to approximate by building increasingly complex meshes, starting with a box.

Newb question though: How is a long triangle a problem? And what do you mean by simplifying it? Break it down into shorter triangles? (We're talking about the sides right? Not the large triangles on the top and bottom?)

[ThibaultLemaire]

So, a simple box works, but as soon as I cut the corners it starts displaying the weird behaviour documented above. So it seems trivially reproducible. (One thing I have not tried though is to only cut one of the corners.)

Here is my workaround: I have sliced my target volume into extruded triangles (following those created by Blender's "triangulate faces" function) treated as different entities with each its own convex hull collider.

This is working perfectly for my purposes. I'm leaving this issue open for tracking.

[ThierryBerger]

Thanks again for reporting! I'm glad you found a way.

I believe the issue you faced is better tracked through the links I shared above so I'll go ahead and close this.