MPC failure

[Saeed-Mansouri]

Hi Dr. Caron,

You have considered a condition for MPC failure for the last DSP of the motion (mentioned below). As far as I understood, if the MPC couldn't be solved during the last DSP, it means that there is no capture trajectory and the walking will resume. To do this, you change stopDuringThisDSP_ variable from true to false in DoubleSupport.cpp. I have seen the MPC may not have a solution when the walking resumed after this moment and the robot would not be stable. I have attached an image of this problem that can occur in the experiments. Did you have seen such a problem in your experiments?

  if(ctl.updatePreview())
  {
    timeSinceLastPreviewUpdate_ = 0.;
  }
  else
  {
    LOG_WARNING("No capture trajectory, resuming walking");
    stopDuringThisDSP_ = false;
  }

[stephane-caron]

We didn't run into that corner case, but it is definitely a bad corner to be in. Looking at your plot, though, the top problem seems to be the steady increase in number of MPC failures rather than the logic of resuming walking when double-support MPC fails.

That logic is, by design, very wrong 😅 I don't remember why I implemented it this way, but the corner case it handles should never happen if the footstep plan is correct (which can be verified in sim since MPC integration in this controller is open loop).