How we researched and ranked robotics data providers

Last updated: 2026-11-09

What we did

Found candidates. We searched the public web for companies offering robotics and physical AI data: teleoperation, egocentric capture, manipulation demonstrations and multi-sensor annotation.
Checked official sources. We read each company's own current web pages in November 2026 and recorded the URLs. A company was listed only if its official site clearly described a robotics or physical AI data offering.
Scored capabilities. We ranked providers on: (a) breadth of robotics-specific collection, including teleoperation, human demonstrations and egocentric capture; (b) annotation depth for robot learning, such as action labels, pose, VLA captions and 3D/LiDAR; (c) delivery infrastructure, including formats, calibration, security certifications and deployment options; (d) how specific their public evidence was, such as named facilities, published processes and stated volumes; and (e) transparency on pricing and pilots.
Wrote drawbacks honestly. Drawbacks reflect gaps in what each vendor documents publicly, or limits of its model, such as labeling-only or video-only offerings.

What we did not do

We did not test any service, buy any data or audit any delivered dataset.
We did not independently verify vendor-reported figures such as contributor counts or hours collected. These are attributed to the vendor.
No vendor paid to be included or to change its position.

Inclusion notes

A capability marked "not stated" means we found no public claim. It is not proof the vendor cannot provide it. We left out some companies because their official robotics evidence was too thin to verify, or because they said they were not yet selling commercially. The market changes quickly, and we update the list as vendors publish new information.

Back to the ranking · About this site