AI BriefingWorld's Largest Autonomous Driving Dataset L2D Released
Key point
Hugging Face's LeRobot team and Yaak have released L2D, the world's largest open-source autonomous driving dataset at 90TB in scale.
Details
The Learning to Drive (L2D) dataset, developed through a collaboration between Yaak and Hugging Face's LeRobot team, has been released. This dataset is the world's largest multimodal dataset aimed at building Spatial Intelligence in the automotive domain.
The key features are as follows:
- Massive Scale: Includes over 5,000 hours of driving data and over 90TB of multimodal data collected across 30 cities in Germany.
- Rich Data Composition: Provides footage from 6 HD cameras, vehicle state (speed, heading, GPS, IMU), environmental state (number of lanes, road type, road surface condition), and weather conditions.
- Natural Language Instructions and Waypoints: Optimized for training End-to-End models, including natural language commands such as "once the signal changes, cross the tram tracks and enter the roundabout" and OpenStreetMap-based Future Waypoints.
- Expert and Novice Policies: Includes both Expert policy data from experienced driving instructors and Student policy data from novice drivers.
L2D is designed to fully support Hugging Face's LeRobot training pipeline and models.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Stanford Researchers Propose EXPO-FT for Universal Post-Training in Robotics