Dyna-2 trained on 1M+ hours of human video; data partner ships 450K hours monthly
Scobleizer · x · 2026-10-29
Scobleizer highlights Dyna's robotics data pipeline: Dyna-2 was trained on over a million hours of human video, and its largest data partner has since delivered 2 million more hours. With 30+ teams under contract and a monthly capacity of 450,000 hours, the next million hours takes roughly ten weeks — a glimpse of how data-hungry embodied AI models are.
More from Embodied
- SiC waveguide costs could fall to ~$100 per unit, but an optics skeptic pushes back — chrisgrayson · 2026-10-29
- Archon Robotics demos whole-body tactile intelligence: feeling inside a bag to grab items — micoolcho · 2026-10-29
- KUKA CEO on physical AI: how 600,000 deployed robots learn new tricks — lukas_m_ziegler · 2026-10-29
- Open-Source Stackchan Desk Robot Chats, Dances and Shows Memes — Full Build Guide Included — alexcovo_eth · 2026-10-29
- Robot steered by cloud AI: Dream Lab demos world-action model controlling MakerMods arms via public API — Scobleizer · 2026-10-29
- MASEM (NeurIPS '26): diverse state sampling for goal-conditioned robot RL — KyleMorgenstein · 2026-10-29