To Err Is Human: The Role of Human Derived Safety Metrics in an Age of Automated Vehicles 2021-01-0875
As industry races to complete technical development of automated driving systems (ADS), important questions are being raised about how to measure the safety of such systems and the overall safety of Automated Vehicles (AVs). Traffic safety engineers have for decades utilized metrics to assess the safety of human drivers and measurements such as Time To Collision (TTC) and Time Headway (THW) have proven to be a useful indicator of increased risk of an accident for human drivers.
But what if we can do better with AVs? Are human driving derived risk metrics meaningful for a self-driving vehicle? Recently, the Institute for Automated Mobility (IAM) published a set of metrics defined specifically for self-driving vehicles that provide a thorough assessment of the safety of an AV. While humans must use estimation and cautious judgement to make decisions, AVs can use precise measurement techniques via sensors and correlate multiple sources of data in real time. Utilizing information such as the reaction time of the ADS, the braking capability of the AV and more, the IAM proposed metrics allow for the assessment of the safety of an AV to be accurately measured, not as a notion of approximated risk, but as a binary calculation of safety.
In this paper we analyze, compare and contrast human driving, risk-oriented safety metrics with the more definitive metrics proposed for AVs. We answer important questions about the necessary evolution of human derived metrics to ensure they are meaningful in the assessment of the safety of an AV, as well as whether novel metrics proposed for AVs can be used to better understand and assess the safety performance of AVs when compared to historical safety measures. Our research proves that AV-based assessment metrics can provide better insight into the safety of both AVs and human drivers.