Global Shutter Egocentric Headset with Real-Time, On-Device Hand Detection
- A global-shutter stereo headset built for all-day egocentric capture
- Two 157° global-shutter cameras, a 400 Hz 9-DoF IMU, and stereo audio — hardware-synced at the source
- On-device real-time hand detection ensures data quality at capture time
- Under 200 g, powered from any USB power bank
The whole frame. One instant.
Most cameras scan a scene line by line. EGO1GS captures the entire frame at once — so fast head turns and quick hand motion stay sharp, straight, and true to geometry. No motion skew means better data: cleaner SLAM trajectories, more reliable depth reconstruction, tighter hand pose. And with a 157° field of view, both hands stay in frame through the full reach of your arms.
The headset knows where your hands are.
Meet EgoHand — our proprietary detection model, purpose-built for finding hands in egocentric video. It runs in real time on the headset itself, checking every frame as it's captured. If the wearer's hands drift out of frame, they're warned on the spot — so quality issues get fixed at capture time, not discovered in training. And every detection is written into the recording on the same clock as video, IMU, and audio.
Four streams. One clock. One file.
Two global-shutter cameras, a 400 Hz 9-DoF IMU, and stereo audio — synced in hardware at the source, captured into a single time-aligned .mcap file. No stitching. No clapper board. No drift to correct. Every session comes off the headset with vision, motion, and sound on one clock — ready to become the data that physical AI trains on.
See what the headset sees.
Pair over Wi-Fi or USB for a live view of both camera streams, with EgoHand detections drawn in. Frame the work volume, check the pitch, and start the session with confidence. Metadata lives where it should: operator, environment, and location are set in the app before capture begins, so every recording arrives labeled. And it's all built on a documented device API — ready for you to build on.
Pull the card after a field day — a 128 GB V30 card ships with the unit, pre-formatted FAT32.
The device uploads to a configured endpoint on its own; set the endpoint once from the companion app.
Real-time streaming transfer over USB while the session keeps recording.
| GI EGO1GS | GI EGO1 | |
|---|---|---|
| Cameras | Dual 1080p at 30 fps · Global shutter | Dual 1080p at 30 fps · Rolling shutter |
| Field of view | 157° × 84° per camera | 120° × 72° per camera |
| Stereo baseline | 63 mm — matched to human IPD | 63 mm — matched to human IPD |
| IMU | 9-DoF at 400 Hz | 9-DoF at 200 Hz |
| Audio | Stereo, dual microphones | — |
| Hand detection | EgoHand — real-time, on-device, every frame* | — |
| Synchronization | All four streams on one hardware clock | Cameras and IMU on one hardware clock |
| Encoding | H.265 / HEVC, on-device | H.265 / HEVC, on-device |
| Output | One time-aligned .mcap file | One time-aligned .mcap file |
| Storage | MicroSD — 128 GB card included | MicroSD — user-supplied |
| Connectivity | Wi-Fi + USB — live preview, upload, direct transfer | Wi-Fi — live preview, offload |
| Voice prompts | Built-in speaker — session and status announcements | — |
| Mount | Adjustable pitch bracket | GoPro-compatible, adjustable pitch |
| Power | Any USB power bank — runtime scales with capacity | Any USB power bank — runtime scales with capacity |
| Weight | ≈ 200 g, headset only | ≈ 200 g, headset only |
* Detector accuracy, throughput, and the exact on-device inference budget are still being characterized; figures will be published with the production firmware.
Bring planet-scale human dexterity to your robots.
Sample recordings are available on request, and we're partnering with labs and teams collecting at scale. Let's talk.