Egocentric video data, licensed or recorded to your spec.
Egocentric video is first-person footage filmed from the wearer's point of view on smart glasses or head-mounted cameras. Robotics and embodied-AI teams train on it to learn how people use their hands to handle objects and finish tasks. Indri Labs licenses first-person footage from the companies that already hold it, and films new footage to your spec when none exists.
Updated · Indri Labs
Why do robotics teams train on first-person video?
A robot that works with its hands needs examples of hands at work, seen from roughly where its own cameras sit. First-person video shows that, and it is easier to collect at scale than robot teleoperation data. In 2026 NVIDIA researchers trained a manipulation model on 20,854 hours of egocentric human video and reported that its performance kept improving as they added hours.
Demand now reaches past the open datasets. Companies pay people to film everyday and workplace tasks on head cameras, because the footage a given robot needs (a specific warehouse, kitchen or production line) rarely exists in public.
Should you license existing footage or record new footage?
License what exists when a company already films the work you need. Record new footage when nobody does. Most projects mix both, and the brief prices each source on its own line.
- Licensed: existing first-person and fixed-camera footage from companies such as logistics operators
- Recorded: new footage filmed to your spec on smart glasses and head cameras
- Both: one contract, one invoice and one merged dataset
What do you specify?
A call or a page of notes is enough. We turn it into a spec, then into a priced brief with a sample from every source.
- Tasks and environments: which work, where, and how many sites
- Volume: hours of footage and how many people filming
- Coverage: regions, languages and edge cases
- Camera: smart glasses, head cameras, or both
- Format, rights and deadline
What arrives?
One dataset in your format, checked against your spec, with duplicates and personal data removed and the source and license attached to every record.
Questions, answered.
What is egocentric video data?
Egocentric video data is first-person footage recorded from a camera worn by a person, usually on smart glasses or a head mount, so the frame shows what the wearer sees and what their hands do. It is used to train robotics, embodied-AI and video models on real tasks.
Where can I buy egocentric video for robotics training?
Indri Labs sources it two ways: we license existing first-person footage from companies that hold it, and we record new footage to your spec with people wearing smart glasses and head cameras. Every source is listed with a sample and a price in a free brief before you commit.
Can you record first-person video to a custom spec?
Yes. You set the tasks, environments, sites, camera type and hours, and we film it with people wearing cameras and smart glasses, then deliver it with personal data removed and a license for every clip.
How much does egocentric video data cost?
It depends on hours, environments, camera setup and exclusivity, so we price each project in a free brief with every source and our fee on separate lines. For reference, Defined.ai's CEO told Reuters in 2024 that short videos sold for $2 to $4 each and long-form film for $100 to $300 per hour.
Can I use open egocentric datasets instead?
Sometimes. Check each dataset's license first: some allow commercial training, others are limited to research or non-commercial use. Public sets also cover the scenes their authors chose, so a specific task, workplace or region usually has to be licensed or recorded.