Dyna-2 pretrained on 1M hours of video discovers new robot scaling laws
JasonMa2020 · x · 2026-08-11
Dyna Robotics introduced Dyna-2, a world-action model pre-trained on 1 million hours of human video. At this scale, they discovered several new scaling laws:
- Human data scaling: World-action models exhibit scaling laws on human data across four orders of magnitude (1k to 1M hours).
- Cross-embodiment transfer: The human data scaling law implies a scaling law on unseen robot data.
- Objective & data matter: World modeling and scaling on video data are essential for cross-embodiment transfer to emerge.
Related event: Dyna-2: Million-Hour Human Video Pretraining Unlocks Embodied Scaling Laws(13 posts)→
More from Embodied
- MIT and Tsinghua introduce GeoPT: physics as AI's third modality — MIT News AI · 2026-08-11
- Amazon Gutting Nova AI Models, Pivoting Hard to Robotics — max_paperclips · 2026-08-11
- Big Tech Robotics Engineers Make $400K+, Startups Face a Math Problem — ATTlKA · 2026-08-11
- Next-Gen BCI: Could Nanoparticles Sprayed Up Your Nose Replace Brain Surgery? — Chris_Armstrong · 2026-08-11
- Meta Ray-Bans Roasted with 'Not My Problem' Filter Meme — jonippolito · 2026-08-11
- MemoMind Smart Glasses Exceed Kickstarter Goal 100x, Raise ~$9M — chrisgrayson · 2026-08-11