ZMP during the DSP

[Saeed-Mansouri]

Hi Dr. Caron,

As far as I know, the plane of the support foot is considered to calculate the zmp of the pendulum (pendulum.zpm()) during the DSP. However, for estimating the zmp of the stabilizer (stabilizer.zpm()) during the DSP, you considered the average of the left and right foot planes (zmpframe_). I think that beacuse of this inconsistency that occurs during walking on the stair, the below term of Stabilizer::computeDesiredWrench() should be modified.

zmpError_ = pendulum_.zmp() - measuredZMP_;

[stephane-caron]

That's right, that's why there is a comment in the code next to the line you mention 😉

zmpError_ = pendulum_.zmp() - measuredZMP_; // XXX: both in same plane?

The two are in the same plane during single support, but during double support, the pendulum ZMP is computed in the previous contact frame (see states/DoubleSupport.cpp) while the measured ZMP is computed in a "ZMP frame" midway between the two contact frames (see Stabilizer.cpp).

The discrepancy affects the DCM derivative term (which was disabled on HRP-4 in 2019, but active on HRP-2Kai) and only applies during stair climbing. The DCM derivative term is not so impactful but this could yield some undesired reactions during stair climbing.

@arntanguy @gergondet FYI. This is still around jrl-umi3218/lipm_walking_controller.

There are two straightforward ways things to try from there:

1. Compute the measured ZMP in the previous contact frame (straightforward to implement)
2. Compute the pendulum ZMP in the midway ZMP frame (requires a bit of frame passing, nothing too hard)

@Saeed-Mansouri thanks for pointing this out 👍