Every dataset ships with full technical specs and auditable provenance. 2/4/6-view options, microsecond time sync and a unified Pre-Token format that plugs directly into your training pipeline.
2 / 4 / 6-view camera array. Primary view 2160×2160@60Hz, side and environment views 1920×1080@30Hz, global shutter + fisheye lenses, hardware-level time sync.
| Views | Role | Mount Position | Resolution | Frame Rate | FOV | Shutter | Lens | Purpose |
|---|---|---|---|---|---|---|---|---|
| 2-view | Primary | Head front (L) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | First-person main view |
| 2-view | Primary | Head front (R) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | Redundancy / stereo |
| 4-view | Primary | Head front (L) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | First-person main view |
| 4-view | Primary | Head front (R) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | Redundancy / stereo |
| 4-view | Side | Front ext (L) | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Side operation zone |
| 4-view | Side | Front ext (R) | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Covers main-view blind spot |
| 6-view | Primary | Head front (L) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | First-person main view |
| 6-view | Primary | Head front (R) | 2160×2160 | 60Hz | 120°×90° | Global | Fisheye | Redundancy / stereo |
| 6-view | Side | Front ext (L) | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Side operation zone |
| 6-view | Side | Front ext (R) | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Covers main-view blind spot |
| 6-view | Environment | Head left space | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Scene context |
| 6-view | Environment | Head right space | 1920×1080 | 30Hz | 150°×120° | Global | Fisheye | Enhanced scene understanding |
| Sensor | Metric | Description | Parameter | Unit | Note |
|---|---|---|---|---|---|
| Vision | Focal length (fx,fy) | Focal length along X/Y | cam0 fx=623.87 fy=623.91 | px | Determined by resolution & FOV |
| Vision | Principal point (cx,cy) | Intersection of optical axis & image plane | cam0 cx=953.47 cy=592.21 | px | Manufacturing variance |
| Vision | Radial distortion | Corrects lens radial distortion | Calibrated | — | Critical for fisheye |
| Vision | Tangential distortion | Corrects lens tangential distortion | Usually small | — | Calibrated for precision |
| Vision | FOV (horizontal) | Horizontal field of view | Mono 177.5° | deg | 270° requirement in strict specs |
| Vision | FOV (vertical) | Vertical field of view | Mono 111.6° | deg | No-gap coverage |
| Vision | Resolution | Image pixel size | 1920×1200 | px | Supports fine operation |
| Vision | Frame rate | Frames per second | 30 fps | Hz | 20–30Hz for precision |
| Vision | Shutter type | Exposure method | Global shutter | — | Avoids rolling-shutter |
| Vision | Depth accuracy | Depth measurement accuracy | ≈8mm | mm | Low error on transparent objects |
| IMU | Sample rate | Sampling frequency | 400Hz 9-axis IMU | Hz | Common config |
| IMU | Accelerometer range | Max acceleration | ±24g | g | Large range for dynamics |
| IMU | Gyroscope range | Max angular velocity | ±2000°/s | °/s | Fast head rotation |
| IMU | Gyro random walk | Bias stability | Low random walk | °/h | Lower is better |
| IMU | Output dimensions | Output data axes | 9-axis | axis | Industry standard |
| Category | Metric | Description | Parameter | Unit | Note |
|---|---|---|---|---|---|
| Camera↔IMU | Rotation matrix (R) | Relative rotation | 3×3 rotation matrix | — | Attitude offset |
| Camera↔IMU | Translation vector (t) | Optical center offset | Millimeter accuracy | mm | Millimeter precision |
| Camera↔IMU | Time offset | Exposure vs sampling lag | ≤3ms | ms | Multimodal alignment baseline |
| Camera↔IMU | Calibration residual | Online calibration residual | <0.15° | deg | Unified spatial frame |
| Camera array | Inter-camera rotation | Neighbor camera rotation | 3×3 rotation matrix | — | Basis for 3D reconstruction |
| Camera array | Inter-camera translation | Optical center offset | Millimeter accuracy | mm | Baseline affects depth |
| Camera array | Stereo matching | Triangulation accuracy | ≤3.3mm | mm | Longer baseline, higher accuracy |
| Head-hand | Spatial transform | Head–hands transform | 6DoF + 52 joints | — | SLAM world coordinates |
| Head-hand | Coordination accuracy | 3D coordination accuracy | Vision-only | mm | IR + vision |
| Head-hand | Coordination latency | Wireless latency | N/A (shared clock) | ms | Head-hand same device |
| Quality | Reprojection error | 3D→2D projection error | N/A | px | Smaller is better |
| Quality | ATE | SLAM trajectory error | ≤3mm | mm | Global accuracy |
| Quality | RPE | Relative pose error | ≤3mm | mm | Local motion consistency |
| Quality | Spatial accuracy | Fusion localization error | ≤3mm | mm | Grasp without drift |
| Category | Metric | Description | Parameter | Unit | Note |
|---|---|---|---|---|---|
| Data engineering | End-to-end latency | Capture to local disk | N/A | ms | Real-time compression & anomaly detection |
| Data engineering | Cloud processing | 15s episode processing time | N/A | s | Parse / VIO / annotation |
| Data engineering | Valid sample retention | High-quality ratio after cleaning | >95% | % | 11 quality gates |
| Data engineering | Motion retargeting | Ego to robot action | Raw + per-frame annotation | % | Completed at client layer |
| Data engineering | Device weight | Headset weight | N/A | g | Lightweight |
| Data engineering | Battery life | Continuous capture | N/A | h | Long sessions |
| Data engineering | Local storage | Device storage | N/A | TB | MicroSD expandable |
| Data engineering | Data transfer | Offline transfer interface | N/A | — | Three methods |
| Data engineering | Power consumption | Device power draw | N/A | W | Low power |
| 3D reconstruction | Scene recovery | Millimeter reconstruction | 0.5m≤3.3mm; 0.8m≈8.3mm | mm | Industry-leading |
| 3D reconstruction | Hand trajectory | Wrist 3D trajectory | 1–1.5cm | cm | Hand motion recovery |
| 3D reconstruction | Hand pose accuracy | MANO 21 keypoints | 1–2cm, 42 points/2 hands | cm | MANO model |
| Hand interaction | Keypoint tracking | MANO 21 keypoints | 21 keypoints / 22 DoF | — | Two-hand interaction |
| Hand interaction | Upper-body pose | 23 DoF pose estimation | 23 DoF | — | Motion capture |
| Hand interaction | Full-body skeleton | 2D/3D skeleton | 2D/3D | — | Full-body motion |
| Metadata | Task labels | Instruction / ID / start-end | Structured | — | Foundation model training |
| Metadata | Frame-level CoT | Chain-of-thought annotation | Frame-level | — | High-order semantics |
| Metadata | Calibration | Scene / camera parameters | Full parameter set | — | Required for delivery |
Delivered via cloud key. Batch and annual framework purchase available with a standard MOQ of 5,000 hours; every clip carries a traceable birth certificate for audit and due diligence.
2-view, 4-view and 6-view multimodal setups: primary view 2160×2160@60Hz, side and environment views 1920×1080@30Hz, with hardware-level microsecond time sync.
All data comes with provenance and licensing chain, auditable and citable for due diligence, meeting embodied AI compliance requirements.
Yes. We cover factory, hotel, home care, agriculture and logistics, and can run dedicated collection for edge cases and extreme environments.
Email 2437632665@qq.com with the scenes, views and volume you need. We reply within 24 hours with samples.