Ministry of Science and ICT Releases Guidebook for E2E Learning Data in Autonomous Driving

The Ministry of Science and ICT released a guidebook defining standards for building E2E (end-to-end) learning data in autonomous driving. E2E autonomous driving uses a single AI model to handle sensor input, perception, decision-making, and vehicle control, offering greater flexibility in handling unpredictable scenarios compared to traditional rule-based systems. However, it requires massive datasets. Previously, domestic entities built datasets separately due to differing sensor configurations across vehicle types, hindering data sharing. The ministry developed this guidebook to maximize data sharing and adapt to technological changes, covering data collection, processing, calibration, labeling, and storage formats. It also includes guidelines for scenario selection, position correction, spatial alignment, and dataset specifications, along with case studies. The guideline was developed through collaboration between the Ministry of Science and ICT, Ministry of Industry and Trade, Ministry of Land Transport, and National Police Agency, with ETRI leading the effort. The ministry incorporated industry feedback through events like the Korean ITS Academy session and autonomous driving industry forums, and plans to apply the guidelines in smart cities for E2E data development. The guideline is available on the KADIF website.