Framework for Reliability-Based, Safety-Explicit Horizontal Curve Design using Naturalistic Driving Data

Abstract

The design of horizontal curves is generally based on a deterministic analysis of driver comfort using a point-mass model based on data mostly from the early 1900s and might be outdated. Furthermore, the design lacks quantitative safety evaluation. This paper proposes a new framework for horizontal curve design that provides a quantitative evaluation of safety performance. The framework uses recently collected data from the naturalistic driving study (NDS) to develop models to predict distributions of parameters that reflect contemporary driver behavior, such as speed on curve and comfort threshold in favorable weather and rainy conditions. Statistical models were also developed for more accurate evaluation of vehicle dynamic parameters than the point-mass model. The variability of driver behavior and vehicle dynamics parameters was considered using reliability analysis to develop surrogate safety measures for four design criteria: vehicle stability, comfort threshold, sight distance, and rollover. Safety performance functions were then developed to relate reliability indices to expected safety performance. The results showed that only the driver comfort criterion was not significantly related to expected safety performance. A design example was presented using the proposed framework, which showed the expected change of safety performance of the curve being considered. An optimum radius was then found to minimize the number of expected collisions.