I’m using big meshes as walls for a map. Mesh parts are not physically accurate to how they appear, though; physical quality is greatly reduced for the sake of performance. The result is that huge, open spaces in the map are actually solid and inaccessible. The solution here would be a fourth CollisionFidelity which ensures that the physical model is identical to the visual model. Currently the only way around this is to either not use MeshParts or divide them into a bunch of smaller meshes until they’re simple enough that the visual representation is no longer a lie.
These open spaces that aren’t filled in aren’t supposed to be solid. I shouldn’t need to hack around this behavior just because there’s no option to disable it.
Please! I’d love to be able to create fantastic terrain using meshes, though right now the only decent alternative is using a .obj importer to have a bunch of invisible parts, which is a bit clunky of a solution and takes some meddling to make sure everything’s aligned properly.
We would like to improve this in the future, but we don’t have anything on the roadmap for this right now.
For now our convex decomposition (CollisionFidelity.Default) obviously isn’t that great. It’s a lose fit and the decomp is a bit sub-optimal with an excessive amount of convexes (you can see them by enabling ShowDecompositionGeometry in Studio settings, under Physics). Your best bet is doing that yourself. Make your own convex meshes for collision and set CollisionFidelity.Hull on them. If your mesh is already convex then CollisionFidelity.Hull is pretty much exact and results in a single convex collision shape.
