Events that matter, and the evidence.
A vision model on every device cuts a clip the moment something worth keeping happens. Collisions arrive with g-force, speed and impact timing. Open or download any sample to see the files a delivery contains.
Road samples
Real crashes, in video and data
Road video around the impact, with peak g, the speed trace and impact timing.
School bus ahead
Fired on-device, with gyro, accelerometer and GPS from the moment of the trigger.
Crowded crosswalk
Chinatown, Manhattan: a crowd in the car's path, caught by the on-device model. Video and scene read, no motion data.
Bridges, tunnels and tolls
Entry and exit on road video, matched to the gantry.
Road conditions
Snow
Falling snow, snowy curbs and slush, with the car's telemetry.
Rain and wet roads
Wet streets and wipers in city rain, cut by the on-device model.
Flooded road
Water standing across the road after heavy rain.
Lane blocked
A truck stopped in the travel lane, and the car going around it.
Ten minutes, unbroken
Back-to-back segments from Manhattan streets onto the highway, each with 10 Hz telemetry.
What a road delivery contains
Native capture, no retiming
- Road
- 1280×720 native since Jul 2026, 1920×1080 before
- Frame rate
- ~30 fps
- Units
- 2-min segments · 6-s events
- Codecs
- HEVC native · H.264 after blur
- Picture
- SDR · no audio
Position and motion at 10 Hz
- GNSS + IMU
- aligned to the video
- Motion
- speed · heading · g-force
- Labels
- trigger · scene
- Context
- time of day
Bucket-ready, manifest-first
- Format
- MP4 + JSON sidecar
- Index
- manifest.csv per release
- Hosting
- Google Cloud Storage
- Privacy
- faces and plates blurred
Tell us the event your safety case depends on.