First-person footage
The camera stays close to the hands, tools, surfaces, and decisions involved in physical work.
Industrial egocentric video
datasets for physical AI.
First-person video recorded during real work in textile manufacturing, metal production, and agriculture.
Request a dataset review ↗First-person video
from real industrial work.
We represent an existing archive that shows real environments, tools, movement, and work as it happened.
Recorded during real work
Continuous footage captures the clutter, variation, timing, and unexpected events that simulations often miss.
Rights reviewed before access
Ownership, participant consent, facility permission, and data history are part of the review from the start.
Why real-world video matters
Before a model can act in the real world, it needs to see how the real world works.
Physical AI needs real examples before it can understand actions. The archive shows settings, objects, motion, and outcomes from inside real work.
It gives robotics, embodied-AI, and computer-vision teams a focused source of industrial video for training and evaluation.
proprietary · recorded on-site · available for review under NDAWho this is for
Real-world data for teams building physical AI.
For technical teams that need real industrial video for model training, testing, or deployment.
Robotics labs
Train and test manipulation, navigation, and control models in realistic environments.
Physical-AI teams
Teach general-purpose models about the settings, objects, movement, and outcomes of real work.
Embodied-vision teams
Build first-person visual understanding for action, planning, and behavior.
Industrial automation companies
Find task-specific video for systems moving from demonstrations and simulation to deployment.
A first-person view of real industrial work.
The proprietary archive covers textile manufacturing, metal production, and agriculture. The hours below were reported by the rights holder. The amount accepted for use is confirmed during evaluation.
Archive by industry · represented hours
These figures are estimates until the footage and usage rights are reviewed under NDA.
Textile manufacturing20,000reported by the rights holder
Metal manufacturing3,000subject to evaluation
Agriculture1,000subject to evaluation
Represented hours
The amount of footage the rights holder reports as present in the archive. These totals remain estimates until the relevant footage is reviewed.
Accepted usable hours
The footage that meets the agreed ownership, quality, data history, privacy, and project requirements.
Rights and compliance review
Read about rights and licensing ↗- Ownership history
- Participant consent
- Facility permission
- Data provenance
- Privacy safeguards
- Responsible-use terms
Technical proof
What buyers can verify during evaluation.
The review separates what can be seen in a compressed website preview from what must be confirmed in the source footage and supporting records.
First-person perspective
Review how closely the footage shows hands, tools, surfaces, actions, and the surrounding work environment.
Work over time
Inspect whether sequences preserve the order, timing, variation, and context needed for the intended model or evaluation.
Three industrial domains
Evaluate relevant material from textile manufacturing, metal production, and agriculture against agreed requirements.
Data and rights together
Assess controlled samples alongside available manifest information, rights records, provenance, and acceptance criteria.
Website clips are 21-second, muted, compressed H.264/MP4 previews at 640 × 360. They are not a statement of source-file resolution, full sequence length, sensor coverage, annotation state, or accepted usable volume. Those details are confirmed during evaluation.
Evaluation path
A clear path from sample review to production use.
Each step defines what you can review, what the data must meet, and how it can be used.
See the full evaluation process ↗
- 01
NDA
Set the terms for a private review.
- 02
Sample
Review a small set of footage.
- 03
Acceptance criteria
Define what the data must meet.
- 04
Paid evaluation
Test fit in your workflow.
- 05
Production license
Set long-term access and usage terms.
How we work
A practical review process.
A simple process for technical buyers to decide whether the data is useful, properly documented, and ready for their workflow.
- 01
Tell us what you need
Share the capabilities, environments, tasks, or model gaps that matter to your work.
- 02
Review rights and provenance
Review the available ownership, consent, facility, data history, and privacy records under NDA.
- 03
Choose access and delivery
Agree on the sample, review limits, delivery format, and next step for your data pipeline.