Developing a Bayesian belief network to assess collision risks for connected and autonomous vehicles in urban environments: A socio-technical synthesis

Intelligent Transportation Systems (ITS) with the aim of enhancing mobility and sustainability are gaining momentum across public policy sector. Connected and Autonomous Vehicles (CAVs) constitute an integral element of ITS. The rapid advances in the realm of Artificial Intelligence (AI) and relevant disciplines have accelerated the development and evolution of CAVs which are believed to thoroughly transform the transportation landscape in coming decades or even years. There are manifold potential benefits (e.g., increased safety and accessibility, convenience, saving time and energy, reducing traffic congestion, etc.) perceived for this disruptive technology. Nevertheless, there is a considerable extent of uncertainties over the safe and secure performance of intelligent self-driving cars in urban environments. These uncertainties can deteriorate the existing driving risks and incur new risks which can undermine the functional safety and technical reliability of those vehicles. The interdependencies between risk factors have neither been yet studied within an integrative framework nor from the sociotechnical perspective. In this study, an interdisciplinary approach was adopted to construct a Bayesian Belief Network (BBN) in order to capture influential risk factors in urban settings as well as the interdependencies between them, thereby providing estimates for the risk indices under varying and volatile circumstances. This will enable us to estimate the collision risk for intelligent self-driving cars in urban environments and evaluate the impact of risk mitigation actions. Furthermore, such a model can be used to classify the urban districts based on the estimated risks and serve policymakers in allocating resources to maximise the benefits of CAVs and avoid potential safety consequences. Sociotechnical theory as an interdisciplinary approach was adopted to form the foundation of BBN model. The factors were accordingly divided into four blocks and the intersection of these blocks represent collision risk index to quantify the safety risk in urban environments. To identify the risk factors, integrative literature review together with thematic analysis (TA) were used. A new technique was formulated to populate the node probability tables (NPTs) and generate uniform distributions. Afterwards, nine domain experts assigned weights to the identified links between the nodes and influence of the probability distributions. Sensitivity analysis was conducted to examine the influence of the incorporated nodes on the collision risk index. The outcome of the model (i.e., collision risk index) showed the highest sensitivity to traffic control infrastructure, weather conditions and traffic composition, respectively. Six scenarios were also devised to investigate the fluctuations of collision risk index due to variations in input nodes. The results of this research can provide insights for policymakers in contemplating policy choices such as investing in new or upgrading existing infrastructure, introducing new legislations, imposing regulatory requirements, licensing, and technology standardisation