Bayesian-Based Traffic Safety Evaluation Study for Driverless Infiltration

Abstract

Although driverless technology belongs to the frontier of science and technology, there is no sufficient actual data. From the lack of a comprehensive systematic evaluation method of traffic safety under driverless penetration, considering the impact of the three core systems of driverless perception, decision-making, control, and the complex road factors on the safety of driving, we review the main risk causal factors through the analysis of the accident causal model STAMP and put forward the fusion of the Leaky Noisy-OR Gate and Bayesian network model. The Bayesian network professional analysis tool GeNIe 2.0 was used to simulate, analyze, and evaluate the driverless traffic risk Bayesian network model, which accurately assessed the traffic safety risk under driverless penetration and diagnosed and identified the sensitive risk factors. The results of this study concluded that, in order to effectively deal with the future traffic safety risks of driverless vehicles, vehicle enterprises, research institutions, software and hardware suppliers in the field of driverless driving should strengthen the research and development and manufacturing of key components such as perception, and enhance the depth of research and development of AI decision-making software, which provides a new way of thinking about the management of the safety risk of driverless traffic and a theoretical basis for the development and implementation of risk control measures.