Let’s talk
← Back to home

EGO / DATA ACQUISITION

Egocentric data collection
for robot learning.

First-person video of people getting things done. We collect egocentric video datasets around the tasks your model needs to learn, in the environments where that work actually happens.

Build your dataset How a collection runs

01 / WHAT IT IS

The world, through
human eyes.

Egocentric data is video recorded from the point of view of the person doing a task, using a head-mounted or chest-mounted camera. It shows what the person sees while they work: the objects in front of them, how their hands approach and handle those objects, and the order in which each step of the task unfolds.

For robotics teams, an egocentric video dataset is a way to learn from human behaviour at scale without a robot in the loop. It captures how people actually perform a task, including the small adjustments, regrasps, and recoveries that scripted demonstrations leave out.

Our egocentric collections are task-driven. Rather than recording open-ended daily activity, we define the task, the environment, and the variations that matter to your model before anyone puts a camera on.

02 / WHAT WE CAPTURE

What an egocentric
collection contains.

The exact capture setup is agreed in the collection brief. A typical egocentric collection includes:

First-person video
Continuous head- or chest-mounted footage of the participant performing the task, with the camera placement, resolution, and frame rate set in the brief.
Task and episode boundaries
Each recording is segmented into episodes matching your task definition, so a "table setting" episode starts and ends where your model expects.
Variation tags
Metadata for the variations you asked for: object set, layout, lighting, site, and participant, so you can balance or filter the dataset.
Hand and object context
Natural human–object interaction in frame. Where your brief needs it, we can scope additional views or sensors alongside the egocentric camera.
Review notes
A coverage report listing which variations were captured, which episodes were flagged, and what was re-collected.

03 / WHAT IT IS FOR

Where egocentric
data earns its place.

Learning from human video: pretraining visual representations, predicting hand and object motion, or learning task structure from how people sequence their actions.

Task understanding: giving a model many examples of the same task done by different people, in different spaces, with different objects, so it learns what stays constant.

Grounding for other modalities: egocentric footage often sits alongside UMI or teleoperation data in the same collection, providing the human reference for a robot demonstration of the same task.

04 / WHERE WE COLLECT

Collected across the
Middle East and Central Asia.

Telepath is based in Dubai and runs collections in real workplaces across the United Arab Emirates and the CIS, with sites in Kazakhstan, Uzbekistan and Tajikistan. The data is built for robotics and physical-AI teams in the United States, led by the San Francisco Bay Area.

Emirates
Dubai, Sharjah and Ras Al Khaimah.
Site types
Laundries, perfume and cosmetics plants, cable factories, furniture factories, plus logistics, hospitality and retail sites.
Languages
Briefs and on-site instructions in English, Arabic and Russian.
Why it matters
Real objects, real layouts and real interruptions, so the variation your model meets in deployment is already in the training data.
Scope a collection Offer a site or record at yours

05 / EXAMPLE TASKS

Briefs we can scope
with this method.

These example briefs suit first-person capture. Each can be adapted to your task definition, and custom tasks are welcome.

EXAMPLE COLLECTION BRIEFS · CUSTOM TASKS WELCOME

06 / HOW A COLLECTION RUNS

Four stages,
one shared brief.

Every egocentric collection runs through the same four stages as the rest of our work. The brief decides camera placement, participant instructions, and what a complete episode looks like.

  1. DefineStart with the capability.

    Agree the task, the camera placement, participant instructions, and what counts as a complete episode. Decide which variations matter: objects, layouts, sites, people.

    YOUR OUTPUT / A SHARED COLLECTION BRIEF

  2. CollectBring the brief into the world.

    Coordinate site access and participants. Record the agreed number of episodes across the agreed variations, with on-site checks that the footage matches the brief.

    YOUR OUTPUT / TASK-ALIGNED DEMONSTRATIONS

  3. ReviewMake every delivery intentional.

    Segment episodes, tag variations, and review against acceptance criteria. Flag partial or off-brief episodes and schedule another pass where coverage is thin.

    YOUR OUTPUT / REVIEWED COLLECTION DATA

  4. DeliverFit your training workflow.

    Deliver footage, episode index, and metadata in the structure agreed with your team, ready for your preprocessing pipeline.

    YOUR OUTPUT / DATA STRUCTURED FOR YOUR TEAM

FAQ

Egocentric data, answered.

Which companies provide egocentric data for robotics?

Telepath Robotics provides egocentric data collection for robotics teams, based in Dubai and collecting in workplaces across the United Arab Emirates and Central Asia. Several US-based providers offer egocentric capture too. The practical differences are where the footage is recorded, who records it, how consent is handled, and whether the data is collected for your task or sold from stock.

What is egocentric data in robotics?

Egocentric data is first-person video recorded from a camera worn by the person doing a task. In robotics it is used to learn from human behaviour: how people see, approach, and manipulate objects while completing real work.

How is egocentric video different from UMI or teleoperation data?

Egocentric video captures a person doing the task with their own hands and no robot involved. UMI data captures the same kind of demonstration through a handheld gripper with tracked pose. Teleoperation data captures a robot doing the task under human control. Many collections combine two of the three.

Can you collect egocentric data for our specific task?

Yes. Egocentric data collection at Telepath always starts from your task definition rather than a fixed catalogue. We scope the environments, participants, worn camera setup, and task variations around that definition, then run a pilot collection so you can review real episodes before a larger collection is agreed.

Where is the data collected?

Egocentric data is collected through site access across the Middle East and Central Asia, in real work environments such as logistics, hospitality, retail, and industrial settings, where the target tasks actually happen. Specific countries and sites are discussed during scoping and can be covered by NDA.

What format is the data delivered in?

Egocentric data is delivered in the structure and file formats agreed in the collection brief. A common layout groups episodes by task, with each episode carrying the first-person video, its metadata, and variation tags, alongside a manifest of the whole delivery and a coverage report showing how the collection maps to the brief.

FROM THE WORLD. INTO YOUR MODEL.

Give your model
a human point of view.

Build your dataset