Can a company license its egocentric video for AI training?
Yes, provided the company owns the footage and the people in it can either be covered by an appropriate legal basis or removed and de-identified. Egocentric video is footage recorded from the wearer's point of view: a helmet camera on a building site, smart glasses on a maintenance round, a chest camera on a delivery route, a head-mounted rig used to film a procedure for new hires. Often it was captured for safety, quality or training, then archived.
Data-collection startups are paying gig workers to film household chores for robot training. The Globe and Mail reported that one contributor recorded 17 hours of household tasks and was paid US$50, about US$2.94 an hour (The Globe and Mail, Oct 9 2026). Footage of trained people doing skilled work on real equipment, in working facilities, is a different thing and is much harder to stage in a kitchen.
What do AI teams use egocentric video for?
They use it to learn how hands, tools and objects interact during a task, seen from roughly the angle a robot's own head or wrist camera would have. The main uses:
- Robot manipulation. Video of a person doing a task teaches a model what the steps are and what success looks like before any robot attempts it. One collection vendor matches rig to training goal: plain action cameras for pretraining visual encoders, instrumented glasses for retargeting human actions to robots, stereo-depth helmets for metric 3D (Dexset, Jul 14 2026).
- World models. Long first-person sequences show how a scene changes as someone acts on it, which is the core of predicting what happens next.
- Simulation and digital twins. Footage of real workspaces helps rebuild them for training in simulation. The Logic reported a buyer asking for phone or action-camera footage of homes and offices together with depth information and 3D meshes of the objects in view (The Logic, May 25 2026).
- Benchmarks. The best-known research set, Ego4D, holds 3,670 hours from 931 camera wearers, collected with consenting participants and de-identification where relevant (Ego4D, arXiv).
What makes egocentric video more valuable?
Footage is worth more when a buyer can see what task is being done, with which tools, from start to finish, and can line the video up with other signals. In rough order of impact:
- Hands and work in frame. Clips where the wearer's hands, the tool and the workpiece stay visible are the core of manipulation data.
- Whole tasks. Complete jobs, from setup to clean-up, beat short highlights. Footage of mistakes and how they were fixed is useful too.
- Labels you already have. Work orders, checklists, SOP step names or inspection results tied to a clip turn raw video into supervised data.
- Synced sensors. Depth or stereo, an IMU, gaze and audio, with calibration files. One marketplace listing describes head-mounted stereo with a 6-DOF IMU at 300 Hz aligned to each exposure and calibration per session (Opendatabay listing).
- Metadata. Date, site type, equipment model, task name and the wearer's role (not their name).
- Span and variety. Many sites, crews, seasons and equipment generations, recorded over years, at original resolution and frame rate rather than compressed exports.
What rights and consent does egocentric video need?
Your company needs to own the footage, and the new use has to be lawful for the people who appear in it, including the person wearing the camera. In practice that means reading what workers, clients and site visitors were told when the footage was recorded. The gates that come up most:
- Identifiable people. Under EU and UK data protection law, footage of identifiable people is personal data. The EDPB's video guidance adds that footage only becomes biometric (special category) data when it is put through specific technical processing to identify someone (EDPB Guidelines 3/2019). The UK regulator covers body-worn video and action cameras in its own guidance (ICO).
- Biometrics. Illinois' Biometric Information Privacy Act counts a scan of face geometry as a biometric identifier (photographs are excluded) and bars private companies from selling or otherwise profiting from biometric identifiers or information (740 ILCS 14). Keep face templates or face analytics derived from footage out of any licence.
- Employees. Whether employees' consent extends to a sale of their data is, in an Emory law professor's words quoted by Business Insider, "an ethically gray area" (Business Insider, Oct 13 2026).
- Client sites. Footage shot on a client's premises may fall under the confidentiality terms of that job. Screens, documents and products in view can belong to someone else.
This page is general information, not legal advice. Have counsel review the notices and contracts that cover your footage.
What is egocentric video worth?
There is no public price list for licensed footage of professional work, so the figures available mostly describe what it costs to record new footage or what raw hours are listed for. They are reference points, each dated and labelled by market side, not offers.
| Reference point | Market side | Figure and source |
|---|---|---|
| Gig contributor filming household tasks | Pay to a collector, Oct 2026 | US$50 for 17 hours, about US$2.94 an hour (The Globe and Mail) |
| Head-mounted capture, by rig | Vendor collection cost, Jul 2026 | $15 to $20 (action camera) up to $25 to $40 (instrumented glasses) per finished hour (Dexset) |
| First-person data in China | Reported market quote, Sep 2026 | about CNY 100 per hour (Xinhua Finance) |
| 49.3 hours of raw industrial stereo, non-exclusive | Marketplace listing, checked Oct 2026 | 8 USD per hour, listed at $395 (Opendatabay) |
| 25,000 hours of factory assembly-line footage | Marketplace listing, checked Oct 2026 | $111,800 (Opendatabay) |
| Video licensed for AI training, general | Reported licence rate, May 2026 | about $1 a minute, specialised scenes up to $30 per 30-second clip (The Logic) |
Even the labelled 25,000-hour factory listing above works out to under $5 an hour (Opendatabay). Hours alone are not what is scarce; depth is: skilled work with synced sensors, work orders and outcomes attached, and the rights cleared. fearofgod.ai does not publish per-hour prices. Partnerships typically range from $50K to $3M+.
How does licensing egocentric video work with fearofgod.ai?
You license footage your company already holds, securely, and earn additional, long-term revenue from it. If it already sits on a camera vendor's platform or a cloud drive, licensing it means re-licensing what you already hold. No new data collection is needed.
The steps are the partnership form, a conversation, scope and terms, data preparation, then delivery and payment. You approve the scope before anything is shared, and anything you want left out, such as a site, a client or a period, stays out. fearofgod.ai de-identifies the data before it goes to buyers. Raw data is deleted according to a schedule agreed with the partner.
A rough first description is enough: hours, cameras, kinds of work, and the notices given to staff. Talk to our team when you are ready. Companies with robot logs as well as human video can read licensing robot and teleoperation data.
Selling egocentric video to AI companies
A company that owns first-person footage of its work can sell it to AI companies by licensing it with fearofgod.ai, securely and for long-term revenue. Partnerships typically range from $50K to $3M+. Partners can be paid more than once, through renewals and repeat licences.