The process of training humanoid robots can take various forms. Unitree’s G1 robot for example, is trained partly through teleoperation whereby a human operator wears a motion capture suit or uses controllers to perform particular movements or entire tasks, with the robot mirroring the movements in real time. The process generates data that feeds into imitation learning algorithms, providing the robot with new autonomous skills. Additional reinforcement learning hones the model to make...
