Slow sway on robots with larger ankle flexibility

[stephane-caron]

On HRP-2Kai, which has larger ankle flexibility than HRP-4, we observe some slow oscillations after walking:

The robot successfully stops, then seems to sway back and forth on its own "volition".

[stephane-caron]

The problem is solved by a (major) one-line change in Stabilizer::computeDesiredWrench():

• Before (zero-moment axis goes through reference CoM):

    return {pendulum_.com().cross(desiredForce), desiredForce};

• After (zero-moment axis goes through measured CoM):

    return {measuredCoM_.cross(desiredForce), desiredForce};

Oscillations previously observed vanish:

However, this also makes balance control more "nervous" and DCM gains should be tuned down accordingly.

This is a major change in the controller. Once tested and documented it will get its own release.

[stephane-caron]

Closing this issue as this has been tested on HRP-2Kai. The main consequence of this change is that we can go to lower DCM gains (e.g. P=3, which was previously marginally stable), and should actually do so because the controller is now more "nervous", i.e. the stabilizer reacts faster with larger ZMP deviations compared to the previous behavior.