ClaruCLARU

// training data

Training data for physical AI.

Egocentric, manipulation and navigation datasets captured in the real world — annotated, verified and delivered in days, not months.

Sign UpLogin
500k+
egocentric videos
4M+
annotations
10,000+
global collectors
100+
cities

The only training-data company 100% focused on Physical AI — real kitchens, factories and city streets.

// library

Millions of videos.

Across every environment your model needs.

Fine manipulation at home
Folding clothes — egocentric capture, hands free, 4K/60fps.
Household tasks
Kitchen and cleaning — full trajectories with context.
Depth map
Per-frame depth for 3D reconstruction and sim-to-real.
Annotated segmentation
Instance masks overlaid, human verified.

// capabilities

More than video.

Captured, enriched, annotated and delivered to your workflow.

01

First-person egocentric video

For embodied AI agents, captured in residential and commercial environments worldwide.

02

Teleoperation and fine manipulation

Robot training for pick-and-place, logistics and structured navigation, with synchronized trajectories.

03

Urban navigation at scale

Scene understanding with pose, action and behavior in real streets and suburban environments.

04

Sim-to-real environments

Reinforcement learning and world models, with 3D worlds captured on a proprietary platform.

05

Cinematic video annotation

High-quality licensed footage, annotated for RLHF in video generation models.

// scale

Built for frontier labs.

Proven at scale.

4M+
annotations delivered
100+
cities covered
10,000+
active collectors
5+
data modalities

// catalog

Explore our datasets.

Off-the-shelf or collected to your exact specification.

Egocentric

First-person household tasks

Cooking, tidying, folding and cleaning — full trajectories with action labels.

Manipulation

Industrial pick-and-place

Real factory floors, with hand pose, object and task outcome.

Teleoperation

Teleoperated robot arms

Synchronized demonstrations for behavior cloning and VLA.

Navigation

Urban and suburban routes

Scene understanding at scale, with pose and pedestrian behavior.

Sim-to-real

Captured 3D worlds

Reconstructions for reinforcement learning and world models.

RLHF

Licensed cinematic video

Quality annotations for video generation models.

Login

// how it works

Tell us what you're training.

We scope the dataset and deliver in days.

  1. 01

    Briefing

    Describe the task, environment and output format your model needs.

  2. 02

    Scope

    We define modalities, volume, geographic diversity and acceptance criteria.

  3. 03

    Collection

    Our global collector network captures the footage with the right gear.

  4. 04

    Annotation

    Expert labeling with double human verification and quality control.

  5. 05

    Delivery

    Structured data in your workflow, in days — not months.

// who it's for

Built for teams shipping intelligent physical systems.

  • Physical AI research labs that need large volumes of egocentric and manipulation data.
  • Industrial robotics companies that need real factory-floor data.
  • Generative video AI startups that require high-quality RLHF annotations.
  • World model and reinforcement learning teams that need sim-to-real.
  • Autonomous vehicle companies that need urban scene understanding at scale.

// faq

Questions before you start

Days, not months. The exact timeline depends on volume, modalities and geographic diversity defined in the brief.

// next step

Tell us what you're training.

We scope the dataset and put the right data inside your timeline.

Sign Up