Meshpart Physics Glitch Out

Bug Reports Engine Bugs
,
#1

As seen on the video, when I press play, one of the parts falls to the ground (expected since it is unanchored and not welded), then randomly bounces and then disappears. No idea what causes this.
robloxapp-20240806-1425408.wmv (661,2 kB)

2 Likes
#2

Maybe it’s because of the aerodynamic feature. I assume that the mass of that MeshPart is very light thus it’s easier to get flown away by the air.

1 Like
#3

well aerodynamics are disabled completely so that couldn’t cause it

1 Like
#4

Or maybe the MeshPart is too light that the physics engine does not handle it very nicely

1 Like
#5

it is ~33 RMUs so mass shouldn’t be an issue either

1 Like
#6

Hi! Thanks for posting this issue. Could you please provide a repro file so we can test this locally?

Thanks!

#7

18833341626.rbxm (219,0 kB)

note that the wing is split into parts (originally intended for wing flex but hingeconstraints behave similarly to the described issue) and only the wing root is welded, the 2 outer segments arent, which are the ones behaving oddly

1 Like
#8

The issue here is that the moment of inertia of the MeshParts are “negative” (technically, the principal moments of inertia are negative). This produces incorrect physics, similar to if the part had negative mass.

The reason the inertia is negative is because the mesh is not closed and has large “holes” in it:

image
image719Ă—225 25.8 KB

The resulting convex decomposition of these parts ends up being degenerate, resulting in negative inertia.

There are a few ways to fix this:

  1. Change the CollisionFidelity of these parts to Hull. This produces a better decomposition geometry and the physics no longer glitch out. One caveat: we are updating the way that mass and inertia are computed, so you will need to make sure not to update the collision fidelity of these parts in the future.
  2. Fix the meshes: as I mentioned, the central issue is that the mesh is not closed. By making the mesh manifold, the inertia should no longer be negative. This will require more effort, but will produce correct results even if collision fidelity is changed.
2 Likes
#9

Thank you, setting it to hull was a fix.

2 Likes
#10

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.