L2D Learning to Drive releases world's largest open-source self-driving dataset (R5+ integration)

AI Impact Summary

Yaak and LeRobot announce L2D, the world’s largest open-source multimodal self-driving dataset, featuring 90+ TB, 5000+ hours across 30 German cities, six RGB cameras, sensor data, and two driving policy groups (expert and student) with natural language instructions. The release, tight integration with 🤗’s LeRobot pipeline and OpenStreetMap-based waypoint generation, lowers the data and tooling barriers for training end-to-end spatial intelligence models, accelerating experimentation and benchmarking. By enabling access to diverse scenarios including weather, road types, and policy variations, this CAPABILITY upgrade could intensify open-source competition and require governance around data licensing and usage if adopted broadly.