Simulation in reduced/generalized coordinate introduces instabilities?

[yzd0014]

I was reading the documentation of PhysX, it mentions the cons and pros for doing simulations in maximal coordinate and generalized/reduced coordinate. It says that, one problem for reduced coordinate is that it may not be stable. I wonder if somebody knows there are any papers or publications that talk about this in detail since I didn't quite get their explanation. Also I wonder if mujoco did something to tackle the instability issue they mentioned.

[yuvaltassa]

Their explanation is not very informative. What I'm guessing they are trying to say is that in reduced coordinates bias forces (Coriolis, centripetal) can be quite complicated and fast-changing, so are hard to integrate accurately. MuJoCo has quite a few solutions to this, most notably the advanced integrators we provide.

PS Velocity clamping is a horrible idea. Never ever do that. Somewhat counter intuitively, it can inject energy into the system and make it go unstable.