Question on automatic motion annotation: about PhysCap

[siyuzou]

Thanks for the publishing such a large scale dataset!

I'm interesting in the automatic motion annotation details, and have some questions about the local pose optimization procedure. Equation 3 in the paper is as follow:

I have 2 questions here:

1. PhysCap framely solves torques and accelerations, then use finite difference and PD controller to manipulate an proxy robot to imitate the input (noisy) motion sequence. I can't see there is any "physical loss" that can be directly integrated in the pose optimization procedure. I wonder how you construct this physical loss term?
2. Could you kindly provide the weighting factors?