Drive
The vehicle behind
every measurement.
An 11-device multi-modal collection system including 360° 4D Imaging Radar. Every output passes through standardized calibration and precision time-sync (PTP/gPTP), delivered ready to use with no alignment work.
01 — In the Wild
What the vehicle sees, in real time.
The RIDE vehicle's real-world driving and, at the same moment, the multi-sensor stream flowing inside it — where every frame of this dataset begins.
02 — Collection Pipeline
From road capture to dataset publish.
From pre-collection calibration checks to the Dataset DB update — an operator checks and initiates collection, alignment/conversion/GT generation run automatically, and final quality checks and publish are automated, then expert-reviewed.
Calibration Check & Update
Operator checks extrinsic calibration before collection, updating it if needed
Recording Start
Operator initiates collection after confirming GNSS LOCK and PTP/gPTP sync_state=LOCKED
Capture
Simultaneous capture of Radar · LiDAR · Camera · GNSS/IMU
Hardware-level Time-sync
Precise synchronization of every sensor via PTP/gPTP
Storage
Lossless local recording · split into 60 s clips
Format & Calibration
Consistency check + standard-format (Parquet/MP4) conversion + extrinsic applied
Meta Enrichment & GT Generation
VLM-based meta inference/enrichment + automatic multi-modal GT generation
Quality Check
Automatic checks of file completeness · timestamp monotonicity · sync accuracy · ODD-match fitness, then final expert review
Dataset Publish & DB Update
Dataset finalized/uploaded · Dataset DB updated — automated, then reviewed and approved by an administrator
03 — Stack
The vehicle, the sensors, and the logging tool — together.
The three pillars behind data consistency and reliability — a dedicated collection vehicle, an 11-device multi-sensor stack, and a lossless logging tool.
Hyundai Santa Fe TM
Roof-rack sensor mount · in-house onboard compute. A dedicated platform that guarantees consistent collection across scenarios.
11 Sensor Devices
Centered on AIR4D (AFI920) as the reference sensor — 360° 4D Imaging Radar × 4 · LiDAR × 1 · Camera × 5 · GNSS/IMU × 1.
Every sensor calibrated into a shared base_link frame.
Multi-sensor Logger
Hardware-level time-sync · lossless local recording · raw and refined/synchronized data emitted together.
| Item | Detail |
|---|---|
| 4D Imaging Radar | AFI920 × 4 (front · rear · left · right) — range · azimuth · elevation · velocity · 10 Hz · SR FOV 120° / LR FOV 60° · CAN-FD / Ethernet Spec sheet ↗ |
| 3D LiDAR | Hesai OT128 × 1 (top center) — 360° spinning · 128 ch · 10 Hz · 100 Mbps Ethernet |
| Camera | SG3S-ISX031C-GMSL2F (SENSING) × 5 — 1920×1080 · 20 FPS · GMSL2 → CoaxCapture · front stereo HFOV 60° × 2 + side/rear HFOV 120° × 3 |
| GNSS + INS | Novatel CPT7 × 1 — RTK-grade GNSS · 100 Hz IMU · front · rear dual antenna · sync reference clock for this dataset |
| Onboard Compute | High-performance in-house onboard compute — migrating to NVIDIA DRIVE AGX Orin |
| Sync Switch | Dedicated sync infrastructure (PTP ↔ gPTP conversion) |
| Sync Accuracy | Master ↔ sensor offset ≤ 100 μs (Max ≤ 1,000 μs) · cross-sensor alignment ≤ 25 ms (Max ≤ 50 ms) |
Need a stack like this for your fleet?
Whether it's the same collection setup, a custom collection vehicle, or a dataset tailored to your environment — we'll work through it with you.