AI training data · Egocentric video

First-person video data for AI & robotics

We recruit participants, record real everyday tasks from a head-mounted camera, then clean, annotate and deliver the footage as training-ready datasets — scoped to the tasks your models need to learn.

Real-world data

Everyday tasks, first-person.

Participants wear a head-mounted camera and record their own real work — at home or on the job. No actors, no staged sets.

  • Real tasks, real environmentsGenuine homes, kitchens, warehouses and shops with natural lighting and clutter.
  • Hands and objects in frameCamera placement and scripts keep the manipulation your model needs clearly visible.
  • Consented and privacy-safeSigned releases from every participant and redaction before anything is delivered.
What we capture

Two environments. Hundreds of tasks.

Pick from our standard task library or send your own list — we script, record and quality-check each one.

At home — Residential

  • Cooking & kitchen prep
  • Cleaning & tidying rooms
  • Laundry & folding
  • Dishes & organising

At work — Commercial

  • Warehouse & logistics
  • Retail stocking & checkout
  • Workshop & repair
  • Food service & assembly
How it works

From task list to training-ready dataset

01

Scope

Tasks, environments, hours, regions and specs.

02

Recruit

Matched participants sign consent and releases.

03

Record

Pilot batch for your approval, then scale.

04

QA & label

Review, redaction and annotation.

05

Deliver

Video and labels to your bucket.

Data specs

Captured to your spec

Every project is set up around your model's input requirements. Typical options:

Camera
Head-mounted phone or action camera
Resolution
1080p up to 4K
Frame rate
30 or 60 fps
Extra streams
Audio, narration, IMU (optional)
Clip length
Single tasks or long continuous sessions
Formats
MP4 + JSON / CSV labels
Delivery
S3, GCS, Azure or secure link
Annotation

Raw footage or fully labelled

Step segmentation

Each task split into timestamped steps with start and end times.

Hands & objects

Bounding boxes for hands, tools and the objects being handled.

Narration & captions

Spoken or written descriptions of what the participant is doing.

Clip metadata

Environment, task, region, device and recording conditions.

Trust & compliance

Data you can safely train on

Consent & releases

Informed consent and a signed video release from every participant, stored with the dataset record.

Privacy redaction

Bystander faces, screens, documents and addresses are blurred or removed before delivery.

Clear licensing

Exclusive or non-exclusive licences, secure transfer and documented chain of custody.

FAQ

AI video data FAQ

What is egocentric (first-person) video data?

Video recorded from the point of view of the person doing a task, usually with a head-mounted camera. It shows hands, tools and objects exactly as the worker sees them, which is what AI and robotics models need to learn real-world manipulation.

Can you record the specific tasks we need?

Yes. We work from your task list and write a recording script per task — environments, steps, duration, camera placement and what must stay in frame — then run a pilot batch for your sign-off before scaling.

How do you handle consent and privacy?

Every participant signs an informed consent and video release before recording. Recordings are reviewed, and faces of bystanders, screens, documents and other personal details are blurred or cut before delivery.

Which annotations can you provide?

Task and step segmentation with timestamps, action labels, hand and object bounding boxes, narration or captions, and per-clip metadata. You can also take raw footage only.

How is the data delivered and licensed?

MP4 video with JSON or CSV labels, delivered to your cloud bucket or a secure download link. Datasets can be licensed exclusively to you or non-exclusively, as agreed in the contract.

How is pricing calculated?

Pricing is per hour of accepted footage, with annotation priced separately. It depends on task difficulty, environment, region and turnaround. Send us your requirements for a quote.

Get started

Request a video dataset

Tell us the tasks, environments, hours of footage and annotation you need — we'll reply with a pilot plan, timeline and quote.

We only use your details to reply to this enquiry.