The research team led by Professor Kwon Min-hye from the School of Electronic Engineering at SSU has developed an autonomous driving technology that can predict surrounding vehicles’ driving behaviors and proactively respond to road conditions using AI(Artificial Intelligence) reinforcement learning. This study, published in the June 2025 issue of an international journal, uses reinforcement learning as a core component of physical AI, enabling self-driving cars to make flexible, human-like decisions even in unpredictable situations. The team enhanced the practicality and scalability of this technology by integrating EFT(Episodic Future Thinking)—a method that mimics human foresight—into reinforcement learning. EFT is an AI approach that anticipates potential future scenarios and formulates response strategies in advance, first proposed by Kwon’s team at the prestigious AI conference NeurIPS(Neural Information Processing Systems) in 2024.
Additionally, the researchers adopted offline reinforcement learning based on real-world road data, allowing autonomous vehicles to acquire diverse driving strategies without the need for simulation environments. Simulation experiments demonstrated that the proposed technology outperforms conventional autonomous driving systems in various scenarios, including highway driving and congested traffic conditions.
Professor Kwon stated, “The key achievement lies in enabling autonomous vehicles to go beyond simply reacting to immediate road situations, allowing them to predict surrounding vehicles’ behaviors and make strategic decisions. Since the system is trained on realworld data, it also holds significant industrial potential.”
