SNU ECE Prof. Jun Won Choi’s Team Develops Homegrown End-to-End Autonomous Driving AI Model, Selected as CVPR 2026 Highlight Paper
Recognition at the world’s leading AI conference underscores the global competitiveness of Korea’s autonomous driving technology


▲ (From left) Jungho Kim (Interdisciplinary Program in AI, SNU), Jiyong Oh (Department of Electrical and Computer Engineering, SNU), Seunghoon Yu (Department of Electrical and Computer Engineering, SNU), Hongjae Shin (Interdisciplinary Program in AI, SNU), Dong Hyuk Kwak (Interdisciplinary Program in AI, SNU), and Prof. Jun Won Choi (Department of Electrical and Computer Engineering, SNU)

Seoul National University College of Engineering announced that a research team led by Prof. Jun Won Choi of the Department of Electrical and Computer Engineering has independently developed SafeDrive, an end-to-end (E2E) autonomous driving AI model aligned with the latest global trends in autonomous driving technology. The work has been selected as a Highlight Paper at the prestigious Conference on Computer Vision and Pattern Recognition (CVPR) 2026.

Highlight papers at CVPR represent approximately 3% of all submissions and about 10% of accepted papers, recognizing a small group of highly impactful studies. The achievement by Prof. Choi’s team is regarded as a significant milestone demonstrating that Korean researchers can independently develop world-class autonomous driving AI technologies.

Recent advances in autonomous driving have increasingly shifted toward Physical AI-based approaches to improve safety and handle edge cases. In particular, end-to-end learning methods—where large-scale driving data are collected, refined, and used to emulate human driving decisions—have emerged as a core technology for building autonomous driving foundation models.

In this context, Prof. Choi’s team proposed a novel method called Fine-grained Safety Reasoning, which evaluates multiple candidate driving trajectories generated by the AI model by combining them with perception results and quantitatively scoring their safety. This approach enables the system to select the optimal path while addressing key limitations of existing end-to-end autonomous driving methods, particularly in safety and interpretability.

Notably, this marks the first time a domestically developed end-to-end autonomous driving research paper has been accepted and selected as a Highlight Paper at CVPR, one of the world’s top conferences in AI and computer vision. The result underscores the growing global competitiveness of Korea’s autonomous driving technologies. It also reflects a broader trend in which cutting-edge AI research is being actively produced in Korea, in parallel with recent developments such as NVIDIA’s presentation of autonomous driving models at NeurIPS.

The model developed by Prof. Choi’s team has been incorporated into EAD (Evolutionary Autonomous Driving), a reference model for commercializing End-to-End autonomous driving, currently being developed by an SNU-led consortium with support from the Ministry of Trade, Industry and Energy. In addition, validation studies are underway in collaboration with Korean autonomous driving companies to deploy the model in real vehicles.

Prof. Jun Won Choi stated, “We will continue to advance this technology and build an open ecosystem that enables collaboration and knowledge sharing with industry, ultimately leading to real-world commercialization. Going forward, we plan to further enhance the performance of the EAD model using larger-scale datasets and achieve full commercialization of end-to-end autonomous driving through the use of our own collected data.”

▲ End-to-end autonomous driving model and trajectory generation results developed by Prof. Jun Won Choi’s research team at SNU

[Contact Information]
Prof. Jun Won Choi, Artificial Intelligence & Autonomous Driving (ADR) Lab, Department of Electrical and Computer Engineering, Seoul National University / junwchoi@snu.ac.kr