DYNA-1

Dyna Robotics' first-generation robot foundation model, released in 2025, best known for folding napkins autonomously for hours at a stretch.

DYNA-1 is the first embodied foundation model from the US startup Dyna Robotics, released in June 2025, with the official full name Dynamism v1. It is built for continuous operation in commercial settings: on a restaurant napkin-folding task, the company reports 24 hours of unattended autonomous operation with a 99.4% success rate, folding more than 850 napkins at about 60% of human speed. Its core approach is to put a self-developed reward model — a model that scores the robot's own performance — inside the training loop, letting the robot explore and self-correct, and continuing to generate new data during deployment. In November 2025, the company also said its pretrained base model, without any further fine-tuning, could fold clothes and sort packages in environments it had never seen, and could learn a new task from about an hour of demonstration. Architecture details have not been disclosed.

ExampleAccording to the company's blog, in week 1 the robot could only fold napkins continuously for 5 minutes; by week 6 it could run non-stop for a full 24 hours. DYNA-1 was then deployed folding napkins at a paying customer's site.

Also called

Dyna-1, Dynamism v1, DYNA-1 Foundation Model

Related

Dyna Robotics · DYNA-2 · Reward Model · Real-World Reinforcement Learning · Data Flywheel · Embodied Foundation Model

Sources

Dynamism v1 (DYNA-1) Model: A Breakthrough in Performance and Production-Ready Embodied AI (Dyna Robotics, 2025-06)
DYNA-1 Pre-Training: Zero-Shot Dexterity Is Here (Dyna Robotics, 2025-11)

As of

2025-11