State-Fraction Coupled Quantile Network Based Distributional Reinforcement Learning for Uncertainty-Aware Decision-Making of Autonomous Driving

Distributional reinforcement learning methods have demonstrated strong capabilities in handling aleatoric uncertainty in autonomous driving. However, most of them model environmental states and quantile fractions independently at the input stage, which may limit training performance. To address this limitation, we propose a state-fraction coupled quantile network (SCQN) based distributional reinforcement learning approach along with an associated uncertainty-aware decision-making framework. Experimental results show that the proposed method outperforms existing baselines, yielding safer and more efficient autonomous driving policies in high-risk scenarios.