High‑DOF glove for robot hands
- Origami Robotics, a Y Combinator Winter 2026 startup in Millbrae, says it has built a high-degree-of-freedom robotic hand and a matching data glove to capture human motions for robot training. - The company says the glove and hand “match each other exactly,” aiming to cut the “embodiment gap” between human demonstrations and robot hardware; it also says Amazon physical AI labs bought its hands. - The pitch lands as robotics firms chase better manipulation data, not just better models, for factory and logistics tasks. (ycombinator.com)
Teaching a robot hand usually starts with a mismatch: a human hand moves one way, the robot hand moves another. Origami Robotics says it built a glove-and-hand pair to reduce that mismatch. (ycombinator.com) The company, a Y Combinator Winter 2026 startup based in Millbrae, says its product combines a high-degree-of-freedom robotic hand with a co-designed data-collection glove. The idea is to record a person’s finger motions and map them directly onto the robot hand. (ycombinator.com) (origami-robotics.com) In robotics, “degrees of freedom” means the number of independent ways a jointed system can move. More degrees of freedom usually means finer control, but also a harder control problem and more ways for training data to go wrong. (robohorizon.com) That is where the glove comes in. Origami says the glove and hand were designed to “match each other exactly,” so a human demonstration can become training data for the same motions the robot will later execute. (ycombinator.com) The problem it is trying to solve has a name: the embodiment gap. That is the difference in kinematics, contact, and sensing between a human demonstrator and a robot, which can make a policy trained on one body fail on the other. (ycombinator.com) Origami says existing robot hands often rely on large transmissions, which make the hand harder to model and reduce force transparency, or the operator’s ability to feel and command subtle contact forces cleanly. Its answer is a direct-drive hand with in-joint motors. (ycombinator.com) (origami-robotics.com) The company says it wants to deploy the hardware in factories and logistics centers to collect real-world manipulation data at scale. It compares that ambition to “Tesla like data,” meaning large volumes of field data gathered from real use rather than lab demos alone. (ycombinator.com) Origami also says it has already sold hands to “Physical AI Labs like Amazon.” That detail matters because manipulation hardware has often stayed in research labs instead of reaching buyers who can test it on production tasks. (ycombinator.com) (robohorizon.com) The broader bet is that better robot dexterity may depend as much on better data collection as on bigger models. A glove that mirrors a robot hand is one way to turn human skill into cleaner training examples for grasping, turning, pinching, and in-hand manipulation. (ycombinator.com) (robohorizon.com) Origami’s pitch is straightforward: if the robot hand learns from demonstrations made on a matching glove, the jump from human motion to machine action gets smaller. For robot hands, that is the difference between collecting motion data and collecting usable skill. (ycombinator.com)
