A data company for robot learning.
We send small crews to places where skilled people work, record that work with their consent, and label it so robots can learn from it.
How we operate
Five rules that decide what we record, how we pay for it and what we tell buyers.
Go where the work is.
We record in working cells during real shifts. Staged demonstrations teach staged motion.
Pay for the hour.
Operators and sites are paid for every recorded hour. Nobody is asked to film their own work for free.
Publish the thresholds.
Sync, calibration, pose error and label agreement are measured on every delivery and reported against limits we publish.
Label what a policy uses.
Hands in 3D, contacts, actions and steps come before captions, and we label only the layers a buyer trains on.
Say where it fails.
When a task needs teleoperation or on-robot data instead, we say so before a project starts.
How it compounds
What each site and each delivery leaves behind for the next one.
Sites stay on.
A site paid for every recorded hour has a reason to host the next task, and every new site widens the work we can record.
Schemas carry over.
Label schemas, SOP alignment and QA tooling move from one task family to the next, so each new task starts from what the last one built.
Evals steer capture.
Buyers run every delivery through their own evals. What moves their policies decides which tasks, operators and layers we record next.
Provenance cannot be backfilled.
Consent, pay and de-identification are recorded at capture. Footage gathered without that record cannot be given one later.
Careers
We hire field crews, calibration and ML engineers, and annotation leads who know the tasks they label.
Tell us about something you have built, recorded or labeled, and the role you want to grow into.
Start with one episode.
We send a sample episode with every stream and label layer, in the format you train on. No NDA needed.