A longitudinal car-following risk assessment model based on risk field theory for autonomous vehicles

Abstract This paper proposes a risk assessment method based on trajectory data which are used to quantify the risk faced by drivers for application in autonomous vehicles. A risk field is derived from the field theory of traffic flow, based on which the risk repulsion indicator of car-following is determined. By describing the repulsion force perceived by drivers in the process of car-following, the risk faced by drivers is assessed. The validity of the indicator is established from crash trajectory data obtained by simulation, and a binary logit model is employed to predict the crash. The result shows that the risk repulsion indicator based on risk field theory can distinguish crash states and non-crash states significantly. The prediction accuracy of binary logit model based on risk repulsion performs better than that of crash prediction model based on loop detector data.