Dyna Robotics releases Dyna-2.1, a physical agent with hour-long autonomy using a world-action model trained on one million hours of video

Dyna Robotics released Dyna-2.1, a physical agent that combines a new semi-humanoid robot with a world-action model trained on over one million hours of human video. The company describes Dyna-2.1 as the first physical agent capable of reliable super long-horizon whole-body autonomy, enabling ultra-long real-world workflows.

In demonstrations, Dyna-2.1 achieved a 99.4% success rate folding restaurant napkins for 24 hours continuously, including recovery after disruptions such as pulling over the entire stack. The agent is also deployed at commercial sites like Din Tai Fung, where it folds thousands of napkins daily.

Dyna-2, the underlying world-action model, predicts future video states of the world and uses those predictions to generate robot actions. This architecture is designed to mimic how humans implicitly plan by anticipating the consequences of their actions, improving dexterity, robustness, and generalization to unseen environments and tasks.

The company reports that robots can learn new tasks from as little as 45 minutes of demonstrations, attributing this efficiency to the world-action model’s ability to transfer knowledge across embodiments and scenarios. Dyna argues that synthetic environments and self-generated training data could further unlock progress in robotics by addressing the data bottleneck.

Dyna-2.1 integrates a custom semi-humanoid hardware stack with an agentic system built around Dyna-2. The company emphasizes the need for on-device inference, model distillation, and custom hardware to scale physical AI, citing hardware limitations such as overheating dexterous hands as major bottlenecks.

Dyna’s approach contrasts with traditional vision-language-action (VLA) models, which rely primarily on vision inputs. The company contends that predicting the future of the physical world is key to achieving general-purpose robotics, and that scaling data—rather than engineering specialized solutions—is the path forward.