| description abstract | Abstract. Autonomous vehicle (AV) lane-changing decision-making is a crucial component of intelligent driving systems, requiring a balance between traffic efficiency and driving safety. Utility-based methods are a well-known theory for AV decision-making in discretionary lane-changing scenarios. Traditional methods primarily rely on cost-benefit analysis but often fail to capture human-like decision patterns and behavioral diversity. To address these limitations, this study proposes a novel lane-changing decision-making model that integrates cumulative prospect theory (CPT) for interpretable human behavior prediction and social value orientation (SVO) to dynamically adjust the trade-off between efficiency and safety based on observed lane-changing times. Unlike conventional models that assign fixed weights to decision factors, our approach dynamically adjusts the trade-off between efficiency and safety based on the observed lane-changing durations. We utilize the highD dataset to ensure robust evaluation, with one subset for parameter identification and another subset for validation. Comparative experimental analysis demonstrates that our model significantly outperforms those existing utility-based methods and a decision-making model without behavior-prediction components, achieving higher accuracy (81.87%), F1-score (79.05% for lane-changing and 74.56% for lane-keeping), and G-mean (76.41%), particularly in lane-changing scenarios. These findings contribute to advancing AV lane-changing strategies, offering a more adaptive, human-like, and safety-conscious decision-making framework for real-world traffic environments. | |
