Using Safety Performance Models, Autonomous Vehicle Data, and Machine Learning to Develop Contextual Complexity Criteria to Establish a Standardized Process for On-Road Evaluation of Medically At-Risk Drivers Considering Static and Dynamic Factors of the Roadway Environment
Date of Award
Doctor of Philosophy (PhD)
Dr. Jennifer Ogle
Dr. Johnell Brooks
Dr. Wayne Sarasua
Dr. Mashrur "Ronnie" Chowdhury
The field of transportation engineering has an opportunity to positively impact the medical community, specifically the clinicians who evaluate, train, and rehabilitate at-risk drivers. Driving Rehabilitation Specialists (DRSs) have an essential role in making roads safer for medically-at-risk drivers, their passengers, and other road users. DRSs conduct on-road driving evaluations, which are considered the gold standard to make fitness-to-drive decisions due to their high face validity. Most DRSs use a fixed route, meaning the exact same route is used to evaluate each client. When a DRS develops a fixed route, that clinician identifies characteristics of the roadway they think are most important (e.g., signalized intersections, unprotected left-turns, protected left-turns). While transportation engineers are trained to know that the combination of static (e.g., roadway type, median, presence of lighting) and dynamic (e.g., traffic density, traffic speed, weather) conditions together define the complexity of a driving environment, transportation engineers have not previously developed materials specifically for DRSs. On the other hand, clinicians do not receive specialized training on these engineering topics and, as a result, do not have the skill set or tools to quantify and measure critical aspects of the roadway context in which the on-road evaluation is conducted.
This dissertation sought to create a methodology to measure the contextual complexity of the driving environment considering the roadway’s static and dynamic characteristics with the long-term goal of providing DRSs the tools to design and evaluate routes using tools similar to those available to transportation engineers. This study utilized comprehensive open-source data collected by Waymo autonomous vehicles that allow for the development of models to estimate the roadway environment’s complexity considering both static and dynamic traffic characteristics. An unsupervised machine learning technique using clustering algorithms was used to measure and classify the driving environment’s dynamic characteristics (e.g., vehicle, pedestrians, bicycles) into appropriate risk categories to develop a dynamic complexity model. A static complexity model was developed utilizing safety performance models and critical variables identified in the American Association of State Highway and Transportation Officials (AASHTO) Highway Safety Manual (HSM). The dynamic and static complexity models were then combined to build an absolute complexity model that provides a comprehensive and quantitative evaluation of the roadways. The knowledge and insights gained from the models developed to quantify static, dynamic, and absolute complexity is foundational work that would enable development of the tools for DRSs to evaluate their routes to ensure the most critical roadway components from the transportation engineering perspective are considered in evaluation of driving context. This process is anticipated to revolutionize the process in which on-road driving assessments are designed and evaluated by the clinicians who assess medically at-risk drivers.
Bendigeri, Vijay, "Using Safety Performance Models, Autonomous Vehicle Data, and Machine Learning to Develop Contextual Complexity Criteria to Establish a Standardized Process for On-Road Evaluation of Medically At-Risk Drivers Considering Static and Dynamic Factors of the Roadway Environment" (2022). All Dissertations. 2983.