Robotics labs use egocentric data as imitation-learning training signal for humanoid robots: a human wears head-, chest-, or wrist-mounted cameras (sometimes stereo pairs) while doing everyday work, and the footage is annotated and mapped onto a robot's joints so a model learns human-like motion without a robot present during collection. It captures visual and positional information but not force or tactile feedback, which is why some collection hardware, such as the UMI Gripper, pairs cameras with grip-sensing devices.