How can robots learn object manipulation for everyday tasks just by watching videos of humans in different unstructured settings? In 'Zero-Shot Robot Manipulation from Passive Human Videos', a new paper by Meta AI researchers working with teams at CMU, we outline our Zero-Shot Human2Robot (H2R) framework. Trained with diverse human videos, our H2R framework learns to predict future action trajectories given a scene, and maps it to the robot’s frame for execution. We evaluate on several coarse ma

Few-shot and Zero-shot Learning - Part 02