Claru profile: egocentric video and robot data at volume

Contents

1. What Claru does for robotics teams
2. Published evidence
3. Strengths and limits
4. Pricing
5. Questions to ask Claru
6. Sources

Type	Human data network
Pricing	Not published
Website	claru.ai

Claru builds training datasets for frontier robotics, embodied AI and world models. It reports a network of more than 34,000 contributors in over 70 countries recording real-world video⁠1Training data for physical AIClaru.

Best for: Labs training world models or vision-language-action models that need large volumes of egocentric human video across many real environments.

Not the right fit for: Teams whose model needs robot state and action data from their own robot, shift after shift.

What Claru does for robotics teams

Claru runs three capture pipelines: contributors with wearable cameras in kitchens, workshops, warehouses and outdoor spaces; managed teleoperation, in which trained operators follow structured task protocols on client robot hardware; and game environments that log synchronized video and control inputs. Raw video is enriched with depth maps, segmentation, human pose and captions, then labeled by people for action boundaries and object affordances⁠2Training data for roboticsClaru.

Its own pages draw the line that matters for buyers. Teleoperation produces observation-action data on the exact robot the policy will run on, while egocentric video captures how people do the task across many environments⁠2Training data for roboticsClaru. Claru argues that lab teleoperation is limited in volume and in environmental diversity, and positions its network as the answer to the diversity problem⁠3Teleoperation dataset collection for robot learningClaru.

Published evidence

No manufacturing case studies are published.

Claru reports volumes such as 1.5 million egocentric videos captured for one world-modeling lab, without naming the client⁠1Training data for physical AIClaru.

Strengths and limits

Strengths

• Environmental diversity at volume: kitchens, factory floors and streets across six continents.
• Every clip is enriched with depth maps, pose estimation and segmentation, then labeled by people.
• Also manages teleoperation campaigns with trained operators on structured task protocols.

Limits

• Mostly human video, not robot state-action pairs; its own page notes that teleoperation is what captures the exact robot embodiment.
• No prices are published.
• Scale figures are its own and are not tied to named, linked engagements.

Choose another option if: Your policy needs demonstrations on your own robot in a training format such as LeRobot or RLDS.

Pricing

Not published.

Questions to ask Claru

• What share of our dataset would be robot teleoperation, and what share human video?
• On whose robots do your teleoperation campaigns run, and where?
• In what format are trajectories delivered?
• How are contributor consent and privacy handled in homes?

See how Claru compares with the other providers in the main guide.

Sources

1. Training data for physical AI. Claru. Accessed 8 Nov 2026.
2. Training data for robotics. Claru. Accessed 8 Nov 2026.
3. Teleoperation dataset collection for robot learning. Claru. Accessed 8 Nov 2026.

Further reading

• Buyer's guide · Updated quarterlyThe best teleoperation companies in 2026An analysis of nine companies that operate robots remotely or collect robot training data, compared on seven criteria, with the evidence behind each one and the established path to a first dataset.
• Guide · The marketWhat teleoperation companies do, and which kind you needOne word covers four different businesses. This guide separates them, explains the two jobs they do and shows what each kind hands back at the end of a shift.
• Guide · Choosing a partnerHow to choose a teleoperation partner for your robotsThe differences between providers show up after the contract: who trains the operators, who owns the data and what the files look like. Here is how to surface them first.