REC 00:00:00:00 · CAM-01 · EU
Groundry records professional workers — cleaners, cooks, technicians — performing real tasks in real European environments, and delivers annotated, training-ready first-person video for robot foundation models.
One frame, as we deliver it: hands, objects and actions — labelled.
The pipeline
Head-mounted, hands-centred first-person video, recorded in lived-in European homes and workplaces.
Framing, blur, hand visibility, task completeness, privacy. Sub-spec footage never reaches you and is never billed.
Temporal action segments, object and interaction labels and task metadata, in your taxonomy.
JSON, COCO-style or your own schema, with documentation. Your pipeline ingests it on day one.
Gloved hands mid-task, annotation boxes on.
Why it matters
The best-performing robot foundation models are now pre-trained on thousands of hours of egocentric human video — but most available footage is bulk-collected, redundant, and blind to the environments where home and service robots will actually be deployed.
We exist to fix that: deliberate demonstrations, by professionals, in the European homes and workplaces your robots are heading for.
What you get
Our workers aren’t gig recruits with a phone strapped to their head. They are trained professionals who understand the capture protocol: hands in frame, natural pace, full task coverage, deliberate handling of edge cases.
How we captureEvery delivery ships with temporal action segmentation, object labels and task metadata to your taxonomy — so your team spends GPU hours training, not cleaning data.
See the specEU households and workplaces, EU data residency, GDPR-grade consent chains. The deployment environments your US and Asian datasets don’t cover.
Rights & complianceCleaning today; kitchens, hospitality and light logistics next. If a professional can demonstrate it on camera, we can spec it, capture it and annotate it.
Explore use casesAbout Groundry
We come from e-commerce content operations and quality assurance — the craft of making messy real-world information machine-readable. Now we point that craft at physical AI.
The team at work — annotators in front of their screens. Natural, candid, not staged.
A 30-minute scoping call is all it takes to define a free, spec-matched sample — delivered within 48 hours.
