Explainable modeling of lane-changing risk assessment incorporating driving-behavior uncertainty

OBJECTIVE: Lane-changing is one of the most common driving maneuvers. Nevertheless, dangerous lane-changing can pose a serious threat to surrounding innocent vehicles. Modeling lane-changing risks is crucial for understanding dangerous driving behaviors and preventing hazardous accidents. METHODS: When assessing lane-changing risk based on the physical processes of lane-changing behavior, there are certain limitations in considering the uncertainty of driving behavior. This study proposes a lane-changing risk assessment model that explicitly incorporates the uncertainty of driving behavior. By integrating a risk attenuation function and computing the definite integral of risk values across varying acceleration scenarios, the model comprehensively accounts for diverse driving behaviors to evaluate lane-changing risk. A subset of lane-changing samples is extracted from the NGSIM and highD datasets for model analysis and evaluation. RESULTS: The results show that, when compared with four commonly used lane-changing risk assessment indicators or models, the model proposed in this study performs better in identifying instantaneous lane-changing risks and predicting the deceleration behavior of the adjacent following vehicles. CONCLUSIONS: The risk model proposed in this study provides new ideas for lane-changing risk estimation and can effectively support the construction of human-like autonomous driving systems.