Robot Training Data Companies
Robot learning is bottlenecked by demonstration data: trajectories of real robots doing real tasks, usually recorded through teleoperation. A small industry has formed around collecting, curating and selling that data, next to open efforts that publish it freely.
This list mixes commercial data providers with the organizations behind the large open datasets, because in practice teams source from both.
Last updated 2026-08-09 · 9 companies listed
AY-Robots
Germany
On-demand robot data collection through a global teleoperator network, plus a marketplace of ready-to-train datasets in LeRobot format.
Scale AI
United States
Data engine for frontier AI labs; offers robotics and embodied-AI data collection and annotation programs.
Hugging Face
United States / France
Home of LeRobot, the open-source robot learning stack, and host of most public robot demonstration datasets.
Physical Intelligence
United States
Builds robot foundation models on top of one of the largest in-house teleoperation data collection operations.
Google DeepMind
United Kingdom / United States
Coordinates Open X-Embodiment, the cross-institution robot dataset effort, and trains its own robotics models on it.
NVIDIA
United States
Combines real teleoperation data with large-scale synthetic generation (Isaac Lab, Cosmos) for its GR00T robot foundation models.
Humanoid maker that published AgiBot World, one of the largest open humanoid manipulation datasets.
Unitree Robotics
China
Robot manufacturer that releases open datasets and models alongside its humanoids and quadrupeds.
Encord
United Kingdom
Data annotation and curation platform used for video and sensor data, including robotics workloads.
Need robot training data instead of a list?
AY-Robots collects demonstration data on real robot arms through a global teleoperator network, and sells ready-made datasets on the marketplace.