This paper introduces a class of model predictive safety performance metrics for driving behavior evaluation. Through formulating the interactions of various traffic agents in the dynamic system context, a class of operational safety performance metrics can be formulated as an optimization problem with various components. The proposed framework provides a unified theory for understanding the differences and correlations among various state-of-the-art safety performance evaluation metrics both qualitatively and quantitatively. Moreover, methods that do not necessarily exist in the literature are also derived through the same framework. Various metrics are directly determined by assumptions made regarding factors such as traffic aggressiveness, participant capabilities, dynamic properties, and regulations, to name a few. Regardless of the metric design, all existing metrics and variants of the described metric induce theoretical guarantees of performance in terms of the time to an undesirable outcome per the definition of the metric. Under standard assumptions, the majority of the variants come with real-time tractable solutions formulated in a certain form of a quadratic optimization problem. A series of synthesized examples and simulated testing cases are adopted to empirically illustrate the effectiveness of the proposed framework. Discrepancies and correlations among variants of the described metric are also identified numerically.
A Class of Model Predictive Safety Performance Metrics for Driving Behavior Evaluation
2021-09-19
2852616 byte
Conference paper
Electronic Resource
English
Driving Safety Performance Assessment Metrics for ADS-Equipped Vehicles
British Library Conference Proceedings | 2020
|Transportation Research Record | 2024
|Driving Safety Performance Assessment Metrics for ADS-Equipped Vehicles
British Library Conference Proceedings | 2020
|Driving Safety Performance Assessment Metrics for ADS-Equipped Vehicles
SAE Technical Papers | 2020
|