We are building robot manipulation models that make robots work in unconstrained and diverse environments.
Current robot solutions fail as soon as there is diversity in tasks and objects. For example, if Instacart wants to build automated grocery stores, current models wouldn't work. We are solving this.

We are training models on human demonstration data collected from our data factories. We hire people just to collect data, performing a wide variety of tasks—like picking an apple, cutting a garlic, or folding a shirt.
Inside the Factory
As hardware converges to five-finger robot hands, we are building the largest corpus of dexterous hand data that no other competitor currently has.

We designed and built custom gloves from scratch. They capture ground-truth joint angles and contact forces directly, enabling collection of vision, action, and force as one multi-modal dataset.
Operators wear head-mounted cameras synced with the sensor gloves. This creates a perfect loop of what the operator sees and the exact forces they exert, directly training the model's spatial reasoning.

We are able to collect this data at 1/10 of the cost in the US. By scaling training data and compute globally, we position ourselves as one of the only companies in the world able to solve generalized robotics.
We can also retarget this rich, dexterous data to lower DOF hardware like grippers for current applications.
Join us in building the largest corpus of dexterous hand data.