Is the time step used for calling saveKinematicState() correct ?

[gjaegy]

Hi,

I'm investigating an issue where some cable/chain made out of rigid-body capsules, attached to a kinematic rigid body, kind of oscillate in a weird way.

Using the debug view, I've noticed the kinematic object the cable is attached to seems to somehow decorrelate from the corresponding rendering object, in a non-constant manner (i.e. offset is irregular).

That irregular movement of the main kinematic object seems to be correlated with the oscillation of the rigid-bodies chain.

After some digging, I've seen that saveKinematicsState() is called with "fixedTimeStep * clampedSimulationSteps". However, we use motion states, and the getWorldTransform() method of these are basically returning position/orientation that are computed according the real time step (i.e. "timeStep", time elapsed between two render frames).

This means, velocities get computed by dividing a distance/movement that matches the rendering frame elapsed time, with a different time, leading to velocities being discontinuous between physics update steps.

Not sure my explanation is clear, if not please let me know...

Is that an expected behaviour ? Why not using the time step passed down to stepSimulation() to call saveKinematicState() ? Using current implementation, I understand the motion state should provide a transform that has been computed using the same "fixedTimeStep * clampedSimulationSteps", which is not the case in our implementation (and I guess, in others either) ?

Thanks for any feedback.