Egocentric capture, at planet scale.
An all-in-one egocentric headset designed for collecting the data that trains the next generation of robotics foundation models — no phone, no laptop tether, all day.
Designed to disappear into the day.
No phone, no laptop, no rig — just the headset. Strap it on at morning coffee and forget it until evening cleanup. The data you collect is what people actually do, not what they do for a camera.
Three streams. One clock.
Synchronized exposure between the two cameras. Synchronized clock between cameras and IMU. The streams arrive on disk already time-aligned — nothing to resample, interpolate, or stitch together before training.
Built to see what you see.
Two cameras, mounted 63 mm apart at the average human inter-pupillary distance. 180° diagonal FOV, 1080P, 30 FPS per camera. Field of view, geometry, frame rate: your robot trains on the world the way you see it. The SDK includes depth reconstruction, so you don't have to roll your own.
Standard MP4. Compressed on-device.
H.265 encoding runs on the device. You copy off standard MP4 files that ffmpeg and every training pipeline can read directly. Typically 30–50% smaller than H.264 at the same quality, so the device records longer and the data moves off it faster.
- Cameras
- 2x 1080P at 30 FPS
- Field of view
- 150° horizontal × 80° vertical (per camera)
- IPD (center-to-center)
- 63 mm — average human inter-pupillary distance
- IMU
- 9-DoF, 200 Hz
- Synchronization
- Hardware-synced exposure and clocks across cameras and IMU
- Encoding
- H.265 (MP4 container), on-device
- Storage
- MicroSD card (not included)
- Connectivity
- Wi-Fi
- Battery
- Typical 5000 mAh · 6–9 hours battery life*
- Weight
- TBD
- Form factor
- Headband with fabric strap
- Colors
- Light · Midnight
* Battery life varies with working environment, ambient temperature, recording settings, and connectivity usage.
Bring planet-scale human dexterity to your robots.
We're partnering with labs and teams collecting at scale. If that's you, let's talk.