Berkeley's Do as I Do Trains Dexterous Hands from Human Video

A UC Berkeley team including advisors Pieter Abbeel and Jitendra Malik released Do as I Do, which retargets ordinary monocular RGB human videos into training data for multi-fingered dexterous hands. The work tackles the scalability bottleneck of dexterous manipulation data, and was featured on RoboPapers podcast episode 107.

2026-09-24 ~ 2026-09-25 · 3 related posts

• Berkeley's Do as I Do turns everyday human videos into dexterous robot hand training data — micoolcho · 2026-09-24
• Do As I Do retargets internet and generated videos into executable robot hand data — chris_j_paxton · 2026-09-25

1 near-duplicate retellings: chris_j_paxton