Argoverse
Argoverse (Argo AI research datasets)
Self-driving datasets pairing sensor data with detailed maps for tracking and forecasting.
Public datasets supported by detailed maps for testing and teaching self-driving vehicles to understand the world. The bucket holds Argoverse 1 for motion forecasting and tracking, Argoverse 2 for motion forecasting, lidar, and sensor data, and additional related sets.
Access
- Cost
- FreeNo charge to download; hosted in an AWS S3 bucket.
- License
- Open, non-commercial onlyCC BY-NC-SA 4.0
- Sign-up
- No sign-up
- How you get it
- API and Bulk downloadParquet, Feather, PNG
What's in it
- One row is
- One driving scenario or sensor log with map context
- Covers
- U.S. cities (see documentation)
- Years
- Not stated
- Updated
- Irregular
- Size
- Not stated
Quality
Standard motion forecasting benchmarks with maps and clear task definitions. Non-commercial license.
- Non-commercial license
- Very large downloads
Connects to
- nuScenes: Driving dataset with annotated objects in other cities
- Boreas Autonomous Driving Dataset: All-weather sequences for localization
Questions it could answer
- How predictable are other vehicles' trajectories at intersections?
- How much do HD maps improve forecasting accuracy?