Egocentric data for physical AI

Real-world human demonstrations from warehouses, cafés and auto shops — captured to your spec.

Global-shutter stereo, IMU, hand pose and action labels, recorded by real workers in real workplaces. Delivered in LeRobot, MCAP or HDF5.

For imitation learning, VLA pre-training and co-training, world models and hand–object interaction research.

Synchronized left and right first-person views of a hand reaching for a cup on a desk Early test capture
Left camera Raw synchronized stereo, 2 × 1920 × 1200 Right camera

Where we capture

Real workplaces, real workers — not a lab set.

We have access to working sites and record people doing their actual jobs, with written consent. Tell us the task; we find or arrange the environment.

01

Warehouse & logistics

  • Pick, scan, pack
  • Box handling and palletizing
  • Sorting and shelving
  • Barcode and label work
02

Café & barista

  • Espresso and pour-over
  • Cup, lid and tray handling
  • Serving and clearing
  • Counter cleaning
03

Car wash & auto service

  • Interior and exterior cleaning
  • Hand-tool work
  • Parts handling
  • Inspection routines

Need another environment — retail, home cleaning, light assembly? Ask us.

What's in every episode

One episode, every layer in sync.

Included

Synchronized stereo video

Two global-shutter cameras, 1920 × 1200 each, 30 or 60 fps, wide field of view.

Included

Hand pose

21 keypoints per hand, per frame, with confidence.

Included

Action segments + language

Task, sub-task and action boundaries with short natural-language descriptions.

Included

Episode metadata

Task spec, outcome (success / failure), environment, operator ID, device and capture settings.

Included

Quality report

Per-episode check of frame count, dropped frames, timestamp continuity and file integrity.

In validation

IMU + exposure timestamps

On-board accelerometer and gyroscope, hardware-timestamped against each frame's exposure.

On request

Depth & camera trajectory

Stereo depth and 6-DoF head trajectory for calibrated captures.

On request

Objects & custom labels

Object tracks, segmentation, contact events or your own label schema.

See the data

What a delivered episode looks like.

Raw left-camera frame: a hand grasping a paper cup on a desk Raw left camera · early test capture

Panels show the delivery format. The full warehouse sample episode is in production. *IMU layer in validation.

Action timeline
reach grasp cup lift move to target place
Episode record
task
Pick up the cup and place it on the target spot
outcome
success | failure
video
stereo 2 × 1920 × 1200, 30 fps
layers
hands · actions · QA · IMU*
Sample pack
  • 2–3 full episodes from one environment
  • Stereo video, hand pose, action labels, QA report
  • Data card with capture protocol and calibration
  • Your choice of LeRobot, MCAP or HDF5
Request the sample

Capture rig

Head-mounted stereo with on-board IMU.

A lightweight head-mounted camera tethered to a belt recorder. Workers keep their hands free and their normal pace.

Sensors2 × AR0234, global shutter
Resolution2 × 1920 × 1200, side-by-side
Frame rate30 or 60 fps
Field of view≈ 144° horizontal, ≈ 92° vertical
Inertial6-axis IMU (accelerometer + gyroscope)
SyncHardware-synchronized stereo
RecordingLossless raw or high-bitrate H.265
Raw sensor frame: an embedded data strip on the left, then the left and right camera views
Raw sensor frame as recorded: embedded sensor-data strip, then left and right views. We keep it lossless until it is decoded.

How we work

From spec to dataset in weeks.

  1. 01

    Spec

    You define the task, environment, layers, volume and format. We propose a capture protocol.

  2. 02

    Pilot

    A 10-hour pilot delivered in 2–3 weeks, so you can test the data in your pipeline first.

  3. 03

    Scale

    More operators, sites and hours on the same protocol, with QA on every episode.

Formats & delivery

Ready for your training stack.

  • LeRobot v3
  • MCAP
  • HDF5
  • MP4 + Parquet

Delivered via S3, Hugging Face (private) or physical drive. Calibration files and a data card are included with every delivery.

Consent & licensing

Clean rights, documented.

  • Written consent from every person recorded
  • Faces of bystanders blurred on request
  • Commercial training license included
  • Exclusive datasets available

FAQ

Questions robotics teams ask us.

Is this robot data or human data?

Human demonstrations, recorded from the worker's point of view. Teams use it for pre-training and co-training alongside their own robot or teleoperation data. We do not collect robot joint states or teleoperation.

Can you record our specific task and objects?

Yes. Send us the task definition, objects and success criteria. We write a capture protocol, agree it with you, and record against it — including your own items shipped to the site if needed.

How is the data priced?

Per accepted hour, depending on the environment, the layers you need and exclusivity. We quote after the spec and always start with a pilot.

Who owns the data, and can it be exclusive?

Every delivery comes with a commercial training licence. Exclusive collections — recorded only for you and never resold — are available.

How do you handle privacy and consent?

Everyone we record signs a written consent for AI-training use. Bystanders' faces and screens can be blurred, and we can avoid sensitive areas on site. Consent records are kept and available for audit.

How fast can you deliver?

A pilot typically takes 2–3 weeks from an agreed spec, including on-site capture, processing and QA. Larger volumes are scheduled per project.

Tell us what your robot needs to learn.

Send the task, environment, layers, volume and format you need. We reply within one business day with a capture plan and a sample.

emil@vayta.tech