16 Paper Presentations at ICDCN2026
16 presentations and 2 awards at ICDCN 2026 held in Nara, Japan (January 2026)
Publications & Talks Awards
International Conference · 2026
LLM-Augmented Driving Behavior Planning for Autonomous Vehicle
Abstract
Autonomous vehicles (AVs) must make reliable decisions in dense urban environments to ensure safety and efficiency. Existing learning-based autonomous driving (AD) systems struggle to interpret high-level information, lack robustness in diverse scenarios, and provide limited interpretability. This study investigates whether large language models (LLMs) can understand complex driving scenarios and perform decision-making in a human manner in urban environments with other vehicles and pedestrians. Our study evaluates the safe driving capability of this approach by comparing it against deep reinforcement learning (DRL) methods in two challenging traffic scenarios using the SUMO simulation platform: (1) a pedestrian jaywalking event and (2) a previously unseen group crossing. Our work explores how open-source LLMs manage to effectively reason driving decisions, even without pretraining on driving-specific data. And open-source base models like Gemma3 can demonstrate strong performance when adapted through lightweight fine-tuning or prompting.
Related Lab Activities
Related Projects
A Platform for Digitalizing Knowledge of Regional Communities
Japan Science and Technology Agency (JST) CREST
「基礎理論とシステム基盤技術の融合によるSociety 5.0のための基盤ソフトウェアの創出」領域
