Berkeley's Workhorse trains humanoid G1 for whole-body manipulation from human data only
pabbeel · x · 2026-10-07
UC Berkeley's Workhorse learns robust whole-body loco-manipulation directly from human demonstrations — no robot, teleop, or retargeting. A demonstrator wears 5 trackers and a chest camera (3.5h of data); a visual planner predicts five-link targets while an RL whole-body tracker executes on a real Unitree G1. The robot topples and climbs a suitcase, catches thrown boxes, and sorts boxes with hands and a kick, recovering from human pushes. Sim results: 77% task completion, 64% under 40 N·s pushes.
Related event: UC Berkeley's Workhorse Teaches Humanoid Robots From Human Demonstrations(2 posts)→
More from Embodied
- Kawasaki Builds a Four-Legged Welding Robot That Climbs Steel Ship Hulls, Trials in 2027 — chris_j_paxton · 2026-10-07
- Kolibri-1 plays Doom with 72 action combos at 39ms median latency — kmodi · 2026-10-07
- Tesla Powershare Home Backup goes live: 30-minute Supercharge adds ~2 days of whole-home power — kleffew94 · 2026-10-07
- Chinese team creates light-activated hydrogel that evolves with living tissue — yogthos · 2026-10-07
- Scoble visits a quick-turn Berkeley machine shop that robot prototype builders rely on — Scobleizer · 2026-10-07
- GroundedSLAM debuts, decisively beating all methods on Meta's egocentric SLAM benchmark — Scobleizer · 2026-10-07