AI development platform Hugging Face has partnered with startup Yaak to expand its open-source robotics initiative, LeRobot, with a large-scale dataset aimed at training autonomous vehicles and robots, the companies said on Tuesday.
The dataset, named Learning to Drive (L2D), consists of more than a petabyte of data collected from sensors installed in German driving school vehicles. It includes camera footage, GPS information, and vehicle dynamics data from both instructors and students navigating a range of real-world conditions, such as construction zones, intersections, and highways.
See also: Everything You Should Know About Geely G-Pilot Assistance System
Hugging Face and Yaak said L2D is designed to advance “end-to-end” learning, a method that enables AI models to predict driving actions directly from sensor inputs. “The AI community can now build end-to-end self-driving models,” Yaak co-founder Harsimrat Sandhawalia and Remi Cadene, a member of Hugging Face’s AI for robotics team, wrote in a blog post.
Unlike existing datasets from firms such as Alphabet’s Waymo and Comma AI, which emphasize specific planning tasks like object detection, L2D offers diverse driving scenarios to support broader AI training, the companies said.
See also: Mobileye Unveils Modular ECU Series for Scalable Autonomous Driving
Hugging Face and Yaak plan to conduct real-world testing of models trained using L2D and LeRobot this summer. The trials will take place in a vehicle with a safety driver, and developers are encouraged to submit AI models for evaluation on tasks such as roundabout navigation and parking.