An open workshop on humanoid robot learning, end to end. A phone video of someone dancing becomes 3D human motion, the motion is retargeted to the Booster K1, a policy learns to follow it in simulation, and then it runs on the real robot. Participants went through the whole pipeline themselves and could launch their policy on the K1.
Tools: GVHMR (video to 3D motion), GMR (retargeting), MuJoCo and mjlab (simulation and training).
Photos and slides will be added later.