Hotel housekeeping
- Bed making & linen change
- Bathroom & surface cleaning
- Amenity restocking
- Room reset checks
01Intro
- Head-mounted POV
- Real shifts, real rooms
- Consent-first
Egocentric video of real work, for embodied AI
We record first-person footage of skilled hotel and hospitality work — housekeeping, kitchens, front desk, laundry — so robots and models can learn how hands actually get things done.
Worker & venue consent in writing
Faces de-identified before storage
Delivered to your spec
02Capture
What we record
Long-horizon, hands-on tasks in the places robots are expected to work next.
Kitchens
- Prep, chopping & plating
- Dishwashing & racking
- Station clean-down
- Storage & restocking
Reception
- Key & card handling
- Luggage handling
- Desk organisation
- Paper & device workflows
Laundry
- Sorting & loading
- Folding sheets & towels
- Ironing & pressing
- Stacking & trolley packing
Short-let & Airbnb turnovers
- Full apartment turnover between guests
- Kitchen & appliance reset
- Linen swap & staging
- Small-space, varied-layout navigation
03Process
How we work
Consent and privacy are steps in the pipeline, not paperwork at the end.
- .01
Written worker consent
Before capture >
Every person who wears a camera signs an informed consent form and can stop or withdraw.
- .02
Venue agreement
Before capture >
The hotel, kitchen or property owner agrees to recording, scope and areas that stay off-limits.
- .03
De-identify before storage
On ingest >
Faces and sensitive details (documents, screens, guest belongings) are blurred or removed before footage is stored.
- .04
Quality control
Review >
Each clip is checked against your spec — framing, task coverage, continuity — before it's offered for acceptance.
- .05
Pay for accepted footage
Delivery >
You review deliveries; billing is tied to the hours you accept, not hours recorded.
04Formats
Formats & metadata
We don't push a fixed product. Technical parameters are set per customer specification.
Raw videoFirst-person footage of complete tasks, delivered as captured after de-identification.Standard
Resolution · FPS · codecPer customer specification.Spec
Session metadataEnvironment type, task description and session context; fields agreed with you up front.Spec
Camera poseOptional, where the capture setup supports it.On request
AnnotationTask segments, actions or object labels — scoped to your taxonomy.On request
DeliveryFile structure and transfer method agreed per project.Spec
session_manifest.jsonExample · fictional
{
"session_id": "example-0001",
"environment": "short_let_apartment",
"tasks": ["linen_change", "kitchen_reset"],
"view": "egocentric_head_mounted",
"video": "per_customer_spec",
"camera_pose": "on_request",
"annotation": "on_request",
"deidentified": true,
"consent": { "worker": true, "venue": true }
}05Pilot
Start with a pilot
20–50hours
of accepted footage
of accepted footage
A pilot is a small, scoped batch that lets you test the data in your own pipeline before committing to volume. We agree the environments, tasks and technical spec together, capture, run QA, and you decide what to accept.
Week 0 >
Scope & spec
Environments, tasks, format, metadata fields.
Capture >
Record & de-identify
Consented sessions, processed before storage.
Review >
QA & acceptance
You review samples and accept against the spec.
Next >
Decide on scale
Adjust the spec or expand to more sites and tasks.
06Trust
Privacy & consent
This describes our process. It isn't legal advice, and each project's terms are agreed in writing.
- >
Opt-in, in writing
Workers sign informed consent before any recording, and can pause, stop or withdraw.
- >
Venue sign-off
Recording happens only with the site's agreement, within agreed areas and hours.
- >
Guests are not the subject
We avoid filming guests; any incidental people, faces or personal items are de-identified.
- >
De-identified before storage
Faces and sensitive data — documents, screens, room numbers — are obscured before footage is stored or shared.
- >
Purpose-limited use
Usage terms for each dataset are set out in the agreement with the buyer.
07Help
FAQ
What is egocentric video?
Footage recorded from the worker's own point of view, usually with a head-mounted camera. It shows hands, tools and objects the way the person doing the task sees them.
Who is this data for?
Teams training robot manipulation policies, world models, video-language models and other embodied AI systems that need examples of real, unscripted physical work.
Is the work staged?
No. We record people doing their actual jobs in working venues. Task lists can be agreed with you so coverage matches your needs.
What resolution and frame rate do you deliver?
Whatever your pipeline needs, within what the capture hardware supports. Video parameters, metadata fields and file structure are set per customer specification.
Do you provide annotations or camera pose?
Raw video is the core deliverable. Camera pose and annotation can be added on request, scoped to your taxonomy.
How do you handle consent and privacy?
Written consent from every worker, agreement from every venue, and de-identification of faces and sensitive details before footage is stored. See the Privacy & consent section above.
How does pricing work?
Pricing depends on environments, spec and volume, and is quoted per project. Billing is tied to accepted footage. Book a call and we'll scope it with you.
How big is a pilot?
Typically 20–50 hours of accepted footage — enough to test in your pipeline before scaling.
08Contact
Request a sample of real-world work
Tell us what your model needs to learn. We'll reply by email.