A general-purpose sewing robot, trained on data from real production.
As cheap as the cheapest garment worker on Earth. No hiring. No training. No downtime.
Curious how we’ll get there?That is what it takes to sew the world’s clothes. Fabric is soft and unpredictable, so even the most automated lines still need a person at every station.
Most urgent: US defense uniforms. By law, they’re made in the US, where labor is costly and skilled sewers are retiring. The Department of Defense spends $1.7 billion a year on them: the need, the funding, and the scale, in one customer.
Built with robot learning. The data comes as a byproduct of revenue.
Sasha built DROID the same way. A global coalition of 15 institutions collected it and became its first users. It worked best in their labs, it’s still their default dataset, and their use grows every year.
98% sellable, at 70% of a skilled sewer’s speed. A person loads each piece.
Sellable T-shirts, sewn start to finish by our robot under teleoperation.
Hanging shirts at 99% success. A person resets the shirt and hanger each time.
Trained with our handheld data collection prototype. Performance held in four manipulation experiments.
One neural network controls the robot and the sewing machine, reacts within 20 ms, and uses episodic memory to stay consistent over long runs.
Flexible, but slow to scale: hard to hire, train, and keep, especially where wages are high.
Fast, but rigid. Each new garment needs new engineering, and a person still loads the fabric.
Could learn to sew, but need factories that want to keep them. We’re earning that place first.
Close to a decade of robot learning research at UC Berkeley, Stanford, and CollectedAI. Led DROID, a robot manipulation dataset 100× more diverse than any before it. To build it, he raised $2M and ran a data collection network of 100 people across 15 institutions, including a distributed 40-person operations team and 20 frontier robotics researchers.
30+ years building anything and everything, including 10 automated factories across Asia and the US, garment automation at Shima Seiki and CreateMe, and robots at Boston Dynamics, Google, and J&J. Has won and delivered grants and contracts for the Departments of Defense (including DARPA), Energy, and the Treasury.
A factory station that runs itself.
A person drops off piles of pieces and walks away. The robot loads them, attaches the sleeves, and passes the work down the line. We’ll measure speed, quality, and success rate against a skilled sewer, and report them as they are.
The US government, as a customer.
We’ll partner with a US uniform manufacturer that holds government contracts. Together, we’ll apply for government grants and more contracts, and grow that relationship while we build the demo.
We are lean by design. If the demo lands on time, our goal becomes sewing the whole shirt autonomously.