How we researched and ranked robotics data companies
Last updated: 2026-11-09
What we did
- Built a candidate list. We searched current public sources for companies that collect, annotate or license robotics and physical-AI training data: teleoperation, egocentric video, UMI demonstrations, and LiDAR or sensor-fusion annotation.
- Checked each candidate on its own website. We read each company's official product pages on 2026-11-09 and recorded only what those pages state.
- Scored each company on five criteria:
- Relevance: how directly the offering targets robotics and physical AI.
- Coverage: collection modes and modalities (teleop, egocentric, UMI, depth, IMU, LiDAR, force/torque) and annotation depth.
- Completeness: whether the company covers collection, QA, annotation, delivery and evals.
- Transparency: public specs, formats, samples, provenance and pricing.
- Evidence of operating at scale: named customers or partners and stated volumes, as claimed by the company.
- Ranked by overall fit for a typical robotics team that needs training data. Specialists for narrower needs, such as perception labeling or footage-to-episode conversion, are ranked lower but called out for the cases where they fit.
What we did not do
- We did not test any product, run pilots, or train models on vendors' data.
- We did not independently check volumes, sync accuracy, workforce sizes or customer results. Where we quote numbers, they are the company's own claims and are credited to the company.
- We did not accept payment for placement. The list contains no affiliate links.
Limits
Most of our evidence comes from pages the vendors control. Several companies are young, and a live website shows that an offering is marketed, not that the company can deliver it at your volume. Ask for samples and run a pilot with written acceptance criteria.
Updates and corrections
We review the list when vendors change their offerings. If a company believes something here is out of date, the fix should cite a public source on its own site.