The article focuses on evaluating real-time safety metrics for automated driving systems. It highlights the necessity of these metrics in assessing risk and aiding decision-making in driving situations. The study proposes an evaluation framework using logged vehicle trajectory data, which can help eliminate prediction errors caused by behavioral assumptions. The evaluation framework is tested on three representative real-time safety metrics- time-to-collision, PEGASUS Criticality Metric, and Model Predictive Instantaneous Safety Metric. Results show Model Predictive Instantaneous Safety Metric has the highest recall and PEGASUS Criticality Metric has the best accuracy. The framework can help researchers, practitioners, and regulators to select appropriate metrics for different applications.
Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.01501
