Physical AI needs data the internet doesn't have.
OmniRL works with construction crews and site operators to record how skilled people do their jobs. We also design and run custom data collection for robotics and video AI teams, built around what their models need.
Our mission
Teach machines real work, learned from the people who do it best.
The skill of a framer, an electrician or a welder was never written down. It lives in their hands. We're capturing it on the job, so the robots that come next learn from experts.
What we do
- Custom data acquisition
- Data that doesn't exist yet. We scope the tasks, sites, people and sensors with you, then collect until you have enough.
- Site access
- Your robots and capture teams on active construction and industrial sites.
- Egocentric and exocentric video
- Head-worn first-person capture of workers mid-task, plus multi-view footage from fixed cameras.
- 3D hand pose and kinematics
- Per-frame hand keypoints and joint angles, time-synced to the video.
- Labeled and annotated
- Raw or annotated to your schema: action segments, bounding boxes, keypoints, captions. Faces blurred, licensing documented.
- Something else?
- Tell us what your model needs. Request a briefing