Published September 2026
The next wave of AI is physical. Robot foundation models, vision-language-action (VLA) models and world models all need the same raw material: thousands of hours of video showing how people actually do real work, seen through their own eyes. That footage is called egocentric video, and it cannot be generated in a studio. It has to come from people who know their trade.
Dovra is the project that builds that supply, and Simera backs it. This post explains what Dovra does, why it fits Simera, and where to go if you are buying data or want to record.
What Dovra does
Dovra produces the human layer of the physical AI stack: first-person capture of real work. Contributors mount their own phone on their forehead and record their normal shift — a mechanic changing oil, a cook plating, a manicurist filing, a warehouse worker palletizing — inside a capture app that ties every clip to a task, a timestamp and a device.
- Scale: standing capacity of about 5,000 hours per 4-week cycle.
- Coverage: 32 countries, 15 work domains and 84 catalogued task types.
- Speed: first clips within days in a covered domain; new domains in two to three weeks.
- Metadata: domain and task from a fixed taxonomy, scene type, country, device model, duration, timestamps and a human quality rating.
Consent and provenance come first
Egocentric footage is filmed inside workplaces. Done badly, it becomes the buyer's legal problem long after delivery. Dovra built the paperwork before the library: every contributor signs an agreement and is paid per approved hour; every business gives written permission; minors, documents, screens and identifiable customers are never filmed; and every clip can be traced to who recorded it, where, when, on which device and under which agreement.
If you are evaluating vendors, Dovra published a practical checklist: Buying egocentric video data: 10 questions to ask a vendor.
Why Simera backs it
Simera has spent years building remote talent networks across Latin America: recruiting, vetting and paying people in dozens of countries, compliantly. Dovra applies the same muscle to a new kind of work. Recording real tasks at scale is, above all, a people operation — finding contributors inside each trade, training them, getting business permissions, and paying them reliably. That is what Simera knows how to do.
It also matters where the data comes from. Most large egocentric datasets are concentrated in two or three countries and in clean environments. Latin America brings different tools, layouts, products and working practices — precisely the distribution shift that breaks models in deployment.
A new income for small businesses and independent workers
On the other side of every clip is a person earning money. In Dovra's network, whoever records is paid USD 4 per approved hour, the same in every country, and local leaders who build a city's network earn between USD 1 and USD 2 per approved hour on top — up to USD 1,000 a month when their network delivers 500 approved hours. For a nail technician, a car-wash worker or a small bakery, it is extra income from the work they already do. Details in Spanish are at dovra.ai/grabar and in Dovra's guides.
🚀 Training a robot or world model? Request a free Dovra sample pack
FAQ
What is egocentric video data?
First-person video recorded from a head-mounted camera at eye level, showing what a person sees and does with their hands during a real task.
Who uses it?
Teams training robot foundation models, VLA models, humanoids, mobile manipulation, industrial automation, world models and video generation.
How is Dovra's data priced?
Per approved hour, depending on the domain, recruiting difficulty and whether annotation is needed. Samples are free.
How is Simera involved?
Simera backs Dovra and brings its experience building and paying remote talent networks across Latin America.
