EasyMimic: A Low-Cost Framework for Robot Imitation Learning from Human Videos

Researchers have unveiled EasyMimic, a low-cost framework that enables robots to learn manipulation tasks by watching standard RGB video of humans performing them. The system extracts 3D hand trajectories and maps them to a robot's gripper control space, bridging the human-to-robot domain gap with visual augmentation. Co-trained with a small amount of robot data, it significantly reduces the need for expensive, large-scale robot data collection, paving the way for affordable home robots.