Robots should learn from human video turned into 3D/4D trajectories plus reinforcement learning in simulation, not from teleoperation. Contact and forces are the unsolved core, and manipulation claims should face real be
The Physical AI Voices by Herbert Scale Experts.