Skip to content

Spec note: Stereo + Wrist Kit

Synchronized egocentric collection with a stereo head camera and a camera on each wrist: depth from the head, close-up views of both hands, and motion data from all three — every frame and sample on one shared timeline.

A real kit recording: Stereo GS head camera (top) and two Mono wrist cameras (bottom). Depth and point cloud computed afterwards on a computer. Download it from Sample data.

Configuration

Position Camera Notes
Head Trinet Stereo or Trinet Stereo GS Two cameras, 70 mm baseline; GS = global shutter for fast motion
Left wrist Trinet Mono One ultra-wide camera
Right wrist Trinet Mono One ultra-wide camera

Every camera needs a radio-beacon adapter for wireless sync — beacons are sold separately. Kits are paired at the factory; more cameras can join the same kit. See Wrist Kit.

Kit at a glance

Cameras 3 (head + two wrists) — 4 video streams (2 head eyes + 2 wrists)
Motion data 3 IMUs — one per camera, each at 400 Hz with a magnetometer at about 100 Hz
Audio Stereo microphones in every camera, each with a hardware mute switch
Start / stop Press the button on any camera — the whole kit starts and stops together
Recording Each camera records to its own microSD card; no phone, computer or cables between cameras
Field monitoring Trinet app → Wireless status shows every camera's state over Bluetooth (firmware 0.5.9+)

Timing and sync

Measure Typical
Motion data to video, within each camera Sub-millisecond, hardware-timestamped
Frame timestamp precision (jitter) About 10 µs
Head stereo, left eye to right eye Both eyes triggered from a single clock
Camera to camera within the kit Under 1 ms — measured typically about 50 µs (validation report)
Frame rate in the kit About 29.7 fps on every camera — intentionally slightly below 30 so the cameras hold their frames in step (30 fps when a camera records on its own)
  • Frames are timestamped at mid-exposure; motion samples at the moment they're measured — all on the camera clock, and across the kit on the shared kit clock.
  • If the radio link drops mid-take, every camera keeps recording and re-aligns when the link returns.
  • For the first one to two seconds of a take, cameras still locking on can be less accurate — trim them when you need the tightest alignment.

Method and details: Timing and sync.

Per-camera specifications

Head: Trinet Stereo Head: Trinet Stereo GS Wrists: Trinet Mono
Sensors 2 × 3 MP rolling shutter 2 × 2.3 MP global shutter 1, rolling shutter
Video 1920×1080 per eye, 30 fps (≈29.7 in a kit) 1920×1080 per eye, 30 fps (≈29.7 in a kit) 1920×1080, 30 fps (≈29.7 in a kit)
Effective field of view ≈158° H × 94° V per eye (from calibration) ≈180° H × 96° V per eye (from calibration) 150° H × 90° V
Codec on card H.264 (default) or H.265 H.264 (default) or H.265 H.264 (default) or H.265
Card bitrate 10 Mbps constant, per eye 10 Mbps constant, per eye ≈15 Mbps variable (adjustable)
Frame timing Mid-exposure of the centre row; per-frame readout time Whole frame at once Mid-exposure
Motion Accel ±8 g + gyro ±2000 °/s at 400 Hz; magnetometer ≈100 Hz same same
Audio Stereo, AAC 44.1 kHz Stereo, AAC 44.1 kHz Stereo, AAC 48 kHz (default)
Calibration Per-unit factory calibration, embedded in every recording same Reference (batch) calibration for the model; per-unit calibration available as a service
Storage ≈10 GB per hour ≈10 GB per hour ≈7 GB per hour

Full details: Stereo · Stereo GS · Mono.

Storage and power

Head (Stereo / GS) Each wrist (Mono) Whole kit
Data per hour ≈10 GB ≈7 GB ≈24 GB
Recommended card 128 GB or more, V30 64 GB or more, V30 —
Hours per card ≈12 h on 128 GB ≈9 h on 64 GB —
Power USB-C, 5 V, 1 A or more USB-C, 5 V One power bank per camera (or a multi-port bank)

See Memory cards and Tested power banks.

Outputs

Each camera writes its own files; takes recorded together share a take number (grp<number>) so matching files group easily. The device tag after it says which camera wrote the file.

Camera Files per take (example)
Head (stereo) grp143024_cae8dabe_take0001_L.mp4, _R.mp4 (video + audio + calibration), _L.vts, _R.vts (frame timestamps), .imu (motion)
Each wrist grp143024_23890f5f_1.mp4 (video + audio + calibration), .vts, .imu, .json (take, kit and role)

To match takes across the three cards and play them side by side, see View a kit take in sync.

The Python toolkit reads every file, plays kit takes side by side, checks frame drops, frame rate and sync, exports to MCAP / ROS 2 and places every take on UTC. Formats: File formats.

Heat

If any camera in the kit needs to cool down, the whole kit pauses and resumes together, so takes stay matched. If one camera's card fills up or is removed, the kit stops together. See Heat and thermal protection.

Try it and order

Specifications for current hardware and firmware 0.5.9; typical values, not guarantees. Subject to change — see the firmware release notes.