Unstable Behavior with TGS Solver and Non-Identity ChildFrame in Articulations with Limited Spherical Joints about physx HOT 4 CLOSED

@preist-nvidia did you have time to look into this issue?

from physx.

Not yet, but it's on my list. And thank you for the excellent repro/bug report.

FYI, we are aware that the spherical articulation joints have several issues/inconsistencies that we are going to address after some higher-priority issues are done. A fix in 5.2.0 is unlikely at this point, however.

from physx.

@dtotsila - any chance you can try to work around this issue using an identity CmassLocalPose?

from physx.

Hi @dtotsila
thank you very much for reporting the issue and providing an excellent reproduction case.
I traced the instability down to a problem where the conversion from a quaternion to the angle-axis representation caused a flip of the joint angle by two pi. While visually not visible from the link state, this flip triggered the joint limits, which would then aggressively push the joint angle out of the limit and cause oscillatory motion.

The fix will appear in an upcoming release of PhysX. In the meantime, you can apply the following changes to your local version:

1. In the function PxQuat computeSphericalJointPositions(...) in file DyFeatherstoneForwardDynamic.cpp and
2. In the function PxTransform FeatherstoneArticulation::propagateTransform(...) in file DyFeatherstoneArticulation.cpp, add the lines

if(jointRotation.w < 0.0f)
  jointRotation = -jointRotation;

before that quaternion is converted into angle-axis format.

I hope this solves your problem. Please re-open the issue if it still persists after applying the changes.

from physx.