Event-Based Risk Assessment for Alert Limits in Automotive Lane Keeping

Abstract

This paper analyzes the integrity risk allocation for automated driving applications involving lane keeping, with an emphasis on converting a risk-per-mile specification into a risk-per-event specification. The integrity risk allocation is the basis for defining a confidence bound (or protection level) for the navigation sensor error (NSE) distribution. For the vehicle to operate safely, the protection level must not exceed the alert limit, a geometric bound that specifies the worst-case tolerable error. In this comparison, the NSE distribution describes the range of values expected from the random error at a particular instant. However, safety targets are typically quoted on a per-mile or per-hour basis. Thus, research is needed to determine the frequency per mile (or hour) of independent events that cause a loss of integrity. This paper models lane keeping in two-way traffic to estimate, for that use case, the rate of occurrence of hazardous events. We combine these models with 2019 data for a particular road segment (part of MA Route 16). For that roadway, we estimate an events/mile rate of 19 in heavy traffic, 0.27 in low traffic, with a mean of 9.5 averaged over a typical weekday.