Egocentric capture for VLA & world models
Trinet
Third eye for your AI.
Tri-netra means “third eye”. With Trinet, your team gets a third eye, capturing rich egocentric data for training Vision-Language-Action models.
Built for all-day capture.
Purpose-built for teams collecting egocentric data. A wide-angle camera ensures you don't miss out on precious data. Lightweight, designed for long sessions, with thermal protection that saves the take before it steps in.
Ultra-wide
Fisheye lens
FHD 30fps
Recording
Hours
Continuous takes to microSD
USB-C
Runs from a power bank
400 Hz
9 DoF IMU rate
µs-level
Hardware IMU-video sync
Research-grade precision
Deterministic by design.
Hardware-synchronized capture - not software alignment. Every frame's timestamp is latched in hardware and corrected to mid-exposure; the IMU clock is disciplined to the same timeline. Any constant camera-IMU offset is pinned down by factory calibration - what remains is microsecond-level jitter that no post-processing can remove. Timing quality that holds up in VLA training, world models, and visual-inertial odometry.
~10µs
Frame timestamp jitter
<10µs
IMU timestamp jitter
<1ms
Drift across a full recording
1 clock
Video, IMU & audio, one timeline
Sync validation report: ~50 µs between cameras ↗How timing and sync work ↗
Zero-friction integration
UVC video with IMU stream.
Trinet streams synchronized IMU data alongside video over a standard UVC interface. The video opens in OpenCV, GStreamer, ROS, or v4l2 like any webcam - no custom drivers. IMU samples and hardware timestamps ride in-band in the same stream, and our SDK extracts them in one call. Or flip to onboard storage and record straight to the device from a power bank - no host computer required.
Standard UVC
Works with any UVC-compatible pipeline. No driver installs.
Inline IMU
9 DoF IMU samples delivered on the same interface as video - extracted in one call with our SDK.
Hardware-timestamped
Every frame and sample carries a sync-accurate timestamp.
Onboard storage mode
Flip a mode and record to microSD - crash-safe recordings that survive even a mid-take power pull.
New · Wrist Kit
From third eye to three eyes.
Add the Wrist Kit and Trinet captures from three viewpoints at once - your head and both hands - with every frame on one shared timeline, aligned to under a millisecond. Exactly the multi-view, bimanual data that VLA, manipulation, and world-model training are hungry for. And it's not just Mono units - Trinet Stereo pairs into the same rig, adding depth to your viewpoints.
Stereo GS head camera (top) and both wrist cameras (bottom) on one shared timeline. Depth and point cloud computed afterwards on a computer.
3
Cameras, one timeline
<1ms
Cross-camera sync
~30×
Tighter than the frame gap
Wireless
Genlock over a dedicated link
Genlocked capture
Frame capture is locked to a common rhythm, so corresponding frames across all three cameras are exposed within a couple of milliseconds - not the random offset independent cameras drift to.
Shared wireless clock
One camera anchors the timeline; the others continuously track their offset and natural drift, so alignment holds through long takes. Every frame and IMU sample is stamped on this shared clock.
Matched one-to-one
Aligned to under a millisecond - more than 30x tighter than the gap between frames - so frames pair up across cameras and motion interpolates to sub-ms agreement.
Built to keep recording
If a camera briefly loses contact it keeps recording locally and re-syncs automatically - a dropped link doesn't end the take.
Effortless to run
Pair once. Record forever.
One button per camera, clear LED feedback. Pair the group a single time - it's remembered across power cycles, and the cameras find each other automatically every time they power on.
Not limited to three
Head and both wrists is just the most common setup. Pair any number of cameras in whatever configuration your capture calls for - they all join the same way and stay on one synchronized timeline.
Audio & privacy
Hears in stereo. Mutes in hardware.
Every Trinet records stereo audio alongside video - the narration and ambient context that egocentric datasets increasingly rely on. And when audio isn't welcome, you don't have to trust a software toggle.
Stereo audio
Two microphones on every unit, recorded as AAC at up to 48kHz - on the device and over the stream, on the same timeline as video and IMU.
Hardware mute switch
A physical switch on the bottom of the camera cuts the microphones for video-only capture - for kitchens, labs, and factory floors where audio isn't welcome. Which way mutes ↗
Nothing phones home
No Wi-Fi, no cloud - recordings stay on your card or your machine. The radios carry only the sync link between cameras and a recording-status broadcast the Trinet app can read.
Start with one. Scale to a rig.
Begin with a single Trinet camera, step up to the Wrist Kit - a three-camera synced rig for both wrists and the head - or add Trinet Stereo for depth and 3D reconstruction. Bulk and custom configurations available.
Trinet
With motion sensing
Full sensor suite with accelerometer, gyroscope, and magnetometer. Microsecond-level hardware synchronization between IMU and camera - purpose-built for VLA, world model, and VIO training pipelines.
- Ultra-wide fisheye camera
- 9 DoF IMU @ 400 Hz
- µs-level hardware IMU-video sync
- UVC video + IMU stream (via SDK)
- UVC or onboard storage
- Stereo audio + hardware mute switch
Wrist Kit
2 wrist + 1 head camera
Two wrist-worn Trinet cameras plus a head-mounted camera - a complete three-camera synced rig capturing your head and both hands on one shared timeline. The multi-view, bimanual data that manipulation policies and world models need. Radio sync beacons are sold separately.
- Three Trinet cameras (2 wrist + 1 head)
- <1ms wireless cross-camera sync
- Genlocked frame capture
- One-button pairing, remembered
- Wrist + head mounts included
- Pairs with more Trinet modules
Trinet Stereo
Dual-camera depth + IMU
A plug-and-play stereo camera with an integrated IMU, stereo audio, and a hardware mute switch. Both cameras are electrically triggered from a single clock - phase-locked by construction, the timing depth and 3D reconstruction need. No RPi, no Jetson - record onboard or stream over UVC. Also available as Trinet Stereo GS, with global-shutter sensors.
- Dual RGB cameras, 70mm baseline
- Single-clock trigger, phase-locked
- Integrated 9 DoF IMU
- Stereo audio (AAC, up to 48kHz)
- Per-unit factory calibration
- Hardware mute switch
- Pairs with other Trinet modules
Trinet Stereo
Depth, straight out of the box.
Two cameras triggered from a single clock, an integrated IMU, and factory calibration served straight off the device. Unlike stereo cameras that need a Raspberry Pi or Jetson strapped to them, Trinet Stereo is self-contained - record onboard, or plug into a phone and stream over UVC with our SDK.
Sample data
Real recordings. Download them.
Raw takes exactly as the cameras wrote them - per-eye video, 400 Hz motion data, per-frame timestamps and calibration - from Trinet Stereo, Stereo GS and a full Wrist Kit. Try your pipeline before you have a camera.
Stereo GS sample: rectified left and right eyes, depth, point cloud, head attitude and 400 Hz motion data from one recording. Depth and point cloud computed afterwards on a computer.
Get Trinet for your team.
Head camera, Wrist Kit, stereo camera, bulk orders, and custom configurations - contact us for pricing.
Contact us