Egocentric data from the people who do the work
First-person capture of skilled workers performing real tasks, structured for manipulation learning and VLA training
What we capture
Every session is multi-view, time-aligned and labeled at the action level
Head-mounted POV video
Third-person views
Hand & tool tracking
Depth & IMU
Narrated steps
Task metadata
Trades we cover
Skilled work across industries, captured where it actually happens
How a collection works
From scoping to delivery, with QC at every step
We agree with you on the tasks, environments, hardware and volume your models need.
We verify experience and certifications, then train contributors on the capture rig.
Sessions happen where the work happens: job sites, workshops, factory floors and warehouses.
Each session is segmented, labeled and checked by reviewers before it is accepted.
Data ships in the format your stack uses, with documentation and consent records.
Frequently asked questions
Working tradespeople and operators with verified experience in the task being captured. Every contributor is paid and has consented to how their data is used.
Yes. Every collection is scoped with you: tasks, settings, hardware, viewpoint and volume are all agreed up front.
Common robotics formats such as MP4 with JSON labels, MCAP, HDF5, RLDS and LeRobot. We can match an internal schema if you share it.
Faces, screens and other identifying details are blurred or removed, and bystanders are excluded from capture. Contributors can withdraw future sessions at any time.
It can be. Exclusivity is agreed per collection.