Hi, I am loading an usd asset in IsaacSim, and it does not fall on the ground as expected. The rendering mesh and the collision mesh looks perfectly fine in the viewport. I test the same usd in Isaac Sim 2022.11 and everything works fine.
It seems the component transform is not treated properly, even though the rendering looks just fine. Is it a bug in IsaacSim 2023.1.1?
Link to usd to reproduce the issue - CollisionTransformBug_240130.zip - Google Drive
When simulation happens the output is either saved in USD or to Fabric. In IsaacSim the omni.physx.fabric extension is usually enabled by default, because it provides significant speedup. Hence output of simulation ends in Fabric. You see then USD properties that are not updated. If you would disable omni.physx.fabric extension you should see them updated properly.
Do you mean PhysX Fabric? I think I have already disabled (by default).
I think you are descriping the final transform after simulation is not updated in Property panel, and that is not exactly my issue.
Here is a video of issue. The issue is that the second actor from left does not fall onto the ground as expected as the first one or the cube do.
I see, then I would suggest to display whats inside PhysX SDK. Go to Window->Simulation->Debug, there is PhysX Simulation debug view, you can enable it and display the collision shapes for example to see whats there.
Here is a screenshot with Collsion AABBs enabled. Clearly the collision and the visual mesh do not match.
Did a few experiment and the error ocurs whenever there is a negative sign in the component prim scale, and the collision mesh does not match visual mesh. Tried the same thing in IsaacSim 2022.2.1 and it does not occur. Is it a bug of Isaac Sim 2023.2.1 and would this be fixed in later release?
Here is what happens in Isaac Sim 2022.2.1, where the collision matches visual mesh perfectly.
That sounds like a regression, I am not sure on top of my head where. Please would it be possible to share the stage with me? If so I will make sure the fix is in next IsaacSim update included. Sorry about the issues.
Looks like this is a cooking regression on our side, the cooked meshes are wrong :-( sorry, we will fix it for next release.
Thanks! Currently I fixed my use case by apply the transform to all mesh vertices to avoid the issue, yet it would be great if it can be supported officially.
Yes, I confirm too that this is a regression.
It should only affect convex decompositions with negative scaling.
Using Convex hulls should be working fine even in this release, so please do try them if they can solve your problem.
Next release will have a fix as stated in previous message, sorry for the inconvenience.