Dyna-2 Pre-Trained on 1M Hours of Human Video Achieves Cross-Embodiment Transfer
iamfakhrealam · x · 2026-08-13
Dyna Robotics introduced Dyna-2, a world-action model pre-trained on 1 million hours of human video data. At this scale, the team discovered several new scaling laws:
- Human Data Scaling: World-action models exhibit a scaling law on human data across four orders of magnitude (1,000 to 1,000,000 hours).
- Robot Data Transfer: This human data scaling law implies a corresponding scaling law on never-before-seen robot data.
- Objective Matters: Both data and objective are crucial; world modeling and scaling on video data are essential for cross-embodiment transfer to emerge.
This points to a promising path for utilizing low-cost human experience to enhance robots' real-world manipulation skills.
Related event: Dyna-2 Model Pretrained on 1M Hours Video(2 posts)→
More from Embodied
- Tesla Unveils 'Flying Roadster' Reportedly Equipped with Cold Gas Thrusters — nptacek · 2026-10-16
- AutoAgent0: VLM-powered agentic runtime lifts long-tail driving success by 28.3pp — yining_hong · 2026-10-16
- CarryOn: dual-loop harness adds personalization and safe recovery to frozen VLA robots — Zixuan Wang · 2026-10-16
- FoveaVLA replaces wrist cameras with learned foveation, hitting 70.8% success — Haoyu_Xiong_ · 2026-10-16
- Musk teases two-way Grok x Optimus integration: Grok in the bot, bot in Grok — beffjezos · 2026-10-16
- SimVTA uses simulator-rendered video to post-train driving world action models with RL — OpenDriveLab · 2026-10-16