PhysicalAI Autonomous Vehicles
NVIDIA
About 1,700 hours of multi-sensor driving data from diverse geographies for end-to-end driving research.
Multi-sensor recordings for training end-to-end autonomous driving systems. Data was collected and labeled automatically from sensors. Access is gated and the license is NVIDIA's own agreement, which the card says allows commercial and non-commercial AV use.
Access
- Cost
- FreeFree but gated; accept NVIDIA's license on the Hub.
- License
- Provider's own termsNVIDIA custom license agreement (tagged other)
- Sign-up
- Free account
- How you get it
- API and Bulk downloadunknown
What's in it
- One row is
- One driving clip with multi-sensor data
- Covers
- Geographically diverse (countries not listed in the record)
- Years
- Not stated
- Updated
- Irregular
- Size
- About 1,700 hours of driving
Quality
Very large and from a major vendor, but only the headline details could be confirmed from the record.
- Labels are automatic
- Custom license
- Sensor and schema details not confirmed here
Connects to
- nuScenes: Smaller, hand-annotated multi-sensor driving benchmark
- Waymo Open Dataset: Another large sensor-rich driving dataset with human labels
Questions it could answer
- How are hours distributed across regions and conditions?
- How accurate are the automatic labels against human labels in smaller sets?