Why Robots Can't Learn from Video Alone: The Need for Action Data

chris_j_paxton · x · 2026-08-13

Commenting on a video showing human manipulation skills, Baoyu Li highlighted that robots cannot learn skills solely from video data.

Humans can quickly learn from videos because they already possess relevant "motion/skill primitives." Acquiring these underlying primitives requires actual action data, making visual input alone far from sufficient.

Original post →

More from Embodied

Embodied channel →