Dyna-2 World Action Model: Pretraining and Long-Horizon Autonomy Progress
Understand the story
On October 14, 2026, Latent Space published an interview with Jason Ma, co-founder and chief scientist of Dyna Robotics. In the interview, Jason Ma explained the technical approach of the Dyna-2 world action model: the model is pretrained on over 1 million hours of human video, combining a diffusion transformer with flow matching in its architecture, generating robot actions by predicting future video. The report focuses on Dyna-2's technical principles and its application to long-horizon autonomous robots. The information currently comes only from this interview, with no other follow-up developments disclosed. Note: the event's current title reads "Dyna-2.1 embodied agent achieves hour-level autonomy," but this new report covers only a technical explanation of the Dyna-2 world action model and does not mention Dyna-2.1 or hour-level autonomy.
AI overview of reporting · Updated 4 days ago
Reporting timeline
Explore different aspects of the story through its reporting.
- Latent SpaceDyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space hosted Dyna Robotics co-founder and chief scientist Jason Ma for a deep dive into the Dyna-2 world action model, pretrained on over 1 million hours of human video, combining a diffusion transformer with flow matching to predict future video and generate robot actions.
Story heat trend
Current heat3·Comparable peak10（10/14 08:00）·Comparable change over 24 hours -51%
The trend compares the same continuously observed sources, which may cover fewer sources than the current heat score. Hover or select the chart to inspect each hour; use the left and right arrow keys to navigate.