I’m using the exact same collision properties as a normal static mesh on a ProceduralMeshComponent but the ragdolls are overlapping via the ProceduralMeshComponent as if it’s collision was disabled.
I’v noticed the collision with ragdolls will work only when the mesh’s polygons doesn’t have any complex shapes.
As if the collision for ragdolls is different than the mesh’s normal collision, like as if it was generated as convex instead of per polygon as it should.
Edit2:
It seems like the collision is always good for particle systems. What is so unique about the ragdolls collision?
Edit3:
Here is a video that shows the problem. Once I add some more complex geometry the collision get messed up.
I’m simply refreshing the sections by using the CreateSection function after changing their vertices positions and other properties. and It works great while keeping the mesh simple.
It also works good for particles even after the mesh getting complex.
For some reason, It won’t do the same with ragdolls…
I’ve been working with the internals of PMC collision a bit recently, in theory it should work exactly the same as a static mesh except for the cost of cooking it on updates. The only thing that I know of that might be partially causing a problem is that mesh udpates (whether through Create or Update) kind of bypass the standard physx insert which can cause some weird issues for anything that was already in contact with it. One way around that is to move the PMC a tiny amount with swept collision on then move it back. That might work, but I have my doubts since they appear to walk around correctly after you edit it before you kill them.
I’m already moving it back and forth before every update to force the navigation mesh to regenerate.
I’v set the bSweep variable to true now and it doesn’t fix the problem.
Thanks for providing the link to the forum thread. This led me to discover that we have an issue in our database, UE-31286, that is related to enabling simulation on kinematic bodies causing them to fall through a trimesh. This issue is currently being investigated by our developers, and is targeted for a 4.13 release. However, there is currently no timeline on when exactly a fix will be released.
