Safety impact of connected and autonomous vehicles on motorways: a traffic microsimulation study

Connected and Autonomous Vehicles (CAVs) promise to improve road safety greatly. Despite the numerous CAV trials around the globe, their benefit has yet to be proven using real-world data. The lack of real-world CAV data has shifted the focus of the research community from traditional safety impact assessment methods to traffic microsimulation in order to evaluate their impacts. However, a plethora of operational, tactical and strategic challenges arising from the implementation of CAV technology remain unaddressed. This thesis presents an innovative and integrated CAV traffic microsimulation framework that aims to cover the aforementioned shortcomings.

A new CAV control algorithm is developed in C++ programming language containing a longitudinal and lateral control algorithm that for the first time takes into consideration sensor error and vehicle platoon formulation of various sizes. A route-based decision-making algorithm for CAVs is also developed. The algorithm is applied to a simulated network of the M1 motorway in the United Kingdom which is calibrated and validated using instrumented vehicle data and inductive loop detector data. Multiple CAV market penetration rate, platoon size and sensor error rate scenarios are formulated and evaluated. Safety evaluation is conducted using traffic conflicts as a safety surrogate measure which is a function of time-to-collision and post encroachment time. The results reveal significant safety benefit (i.e. 10-94% reduction of traffic conflicts) as CAV market penetration increases from 0% to 100%; however, it is underlined that special focus should be given in the motorway merging and diverging areas where CAVs seem to face the most challenges. Additionally, it is proven that if the correct CAV platoon size is implemented at the appropriate point in time, greater safety benefits may be achieved. Otherwise, safety might deteriorate. However, sensor error does not affect traffic conflicts for the studied network.

These results could provide valuable insights to policy makers regarding the reconfiguration of existing infrastructure to accommodate CAVs, the trustworthiness of existing CAV equipment and the optimal platoon size that should be enforced according to the market penetration rate.

Finally, in order to forecast the conflict reduction for any given market penetration rate and understand the underlying factors behind traffic conflicts in a traffic microsimulation environment in-depth, a hierarchical spatial Bayesian negative binomial regression model is developed, based on the simulated CAV data. The results exhibit that besides CAV market penetration rate, speed variance across lanes significantly affects the production of simulated conflicts. As speed variance increases, the safety benefit decreases. These results emphasize the importance of speed homogeneity between lanes in a motorway as well as the increased risk in the motorway merging/diverging areas.