MOBILITY AND SAFETY IMPACTS OF AUTONOMOUS VEHICLES

2020-02-06T12:56:42Z (GMT) by Fasil Sagir
Connected and Autonomous Vehicles (CAV) are revolutionizing the automotive space. We are at the cusp of a, once in a century, transformation in the automotive space. This work strives to understand, analyze and provide insights on the various dimensions this transition is going to impact. We begin with the exploration of the CAV landscape which is in a continuous state of flux. We attempt to examine, analyze and evaluate this space using semi-structured interviews with experts from across the whole country. The interviews are supported additionally by survey questions which further capture the expert views quantitatively. This initial exploratory study leads us to the central questions of this study which include (1) Modeling of SAE (Society of Automotive Engineers) vehicles from level 0 to level 5 using a simulation framework (2) Analysis of mobility and safety impacts of SAE vehicles. (3) Building a predictive model of the risk level of autonomous vehicles based on trajectory information.

For the modeling of AVs, the different levels of SAE were mapped to particular functionalities. Each of these functionalities were then modeled using the external driver model (EDM) and were tested on VISSIM to evaluate their performance. The mobility impacts of these models were tested on a highway and an intersection environment. The analysis were conducted for 100% penetration levels for each SAE and also for different penetration levels.

One of the most important benefits of AVs that has been touted by OEMs and DOTs alike, are the safety benefits of CAVs. Among many industries which will be affected by the safety aspects of CAVs, insurance industry is one of them. An immediate challenge that lies in front of them will be to evaluate the risk level of xiii different SAE classes of vehicles. This will be especially true as most of the SAE level data is unavailable or very scarce. To overcome this limitation, we propose a novel methodology to identify risky driver behavior for every SAE level. The framework includes the utilization of surrogate safety measures modified for SAE levels. The trajectory data created from SAE level simulation is used as the data set for model training and testing which predicts driving risk. The models evaluated are logistic regression, decision trees and neural networks. This framework provides a foundation for modeling the riskiness of autonomous vehicles in traffic networks.