Datoric

Datoric is a robotics training data company that supplies data for physical AI. It sells licensed, ethically sourced datasets and runs custom collections intended to move robot learning systems from research into real world deployment, with documented provenance and quality review on the data it delivers. Its customers are teams training robot manipulation and embodied AI models. Datoric works across three main modalities: egocentric video of household and workplace tasks with task, action and object labels; bimanual UMI demonstrations with left and right gripper views, end effector poses and gripper width; and robot teleoperation data with synchronized video, states, actions and sensor configuration captured on a specified platform. Tactile data is also offered. Catalogued datasets include a UMI dual camera annotated manipulation dataset of 500 hours of bimanual demonstrations, a custom robot teleoperation dataset, and a tactile glove and stereo manipulation dataset. What distinguishes the company is its research method. Datoric starts from a model failure or capability target, forms hypotheses about the data behind it, then runs side by side collection experiments that change one variable at a time, such as cameras, sensors, demonstrators, robots, objects, rooms or labels, while holding tasks and people constant. The method that transfers in downstream evaluation is then used for full scale collection, with ongoing drift checks on capture, labels and model results. Pricing is not published. Buyers request data or plan a collection with the team, so costs are quote based. The company states that it is backed by Y Combinator.