Augmenting Policy Making for Autonomous Vehicles Through Geoinformatics and Psychographics

Data-driven policy making has been on the rise in research, industry, and government; but especially for novel areas such as Artificial Intelligence and Autonomous Vehicles (AVs). Few studies on AVs' performance and feasibility are performed through demographic, geographic, and psychographic data. In this manuscript, descriptive and predictive analytics are presented to evaluate (and partly defy) the most commonplace beliefs for AVs safety, and vehicle collision causes. The goals of this manuscript include: identifying the context for which an accident occurs, defining the factors that feed into that context, and migrating that knowledge to AVs policy making. Policy makers around the world are being forced to re-defining policies for new technologies that effect our everyday life. Multiple aspects can influence such policy making process; the challenge exacerbates when the technology hasn't been used prior (such as AVs). AVs are coming to all roads in the US and around the world. It is very important therefore to understand the factors contributing to this major change. In this paper, machine learning models are trained, tested, and developed (in R) based on locations (longitudes and latitudes) of accidents, driver-pedestrian interactions, and driver's behavior (through three studies). After analyzing demographical data, no major pointers were found that could be used in deriving rules & regulations for the mentioned upcoming change in transport. Therefore, this paper shifts the focus to geographic and psychographic data (which had stronger correlations to safety and other car accident schemes). Additionally, technologies like Vissim and the Traffic Simulator are suggested to generate driver's scenarios, and to simulate certain cases where there can be multiple outcomes. The National Highway Traffic and Safety Administration (NHTSA) is a pioneer in the car safety industry and one of the main data collectors. The work presented in this manuscript uses data (72 datasets, and almost 9200000 data points) from NHTSA. Experimental work is presented, results are recorded, evaluated, and used to define conclusions and future directions.