A Stochastic Car Following Model

This paper describes a data-driven, stochastic car-following model. From a data-base of car-following episodes, the acceleration a of the following vehicle is modeled as drawn from a distribution that is sampled directly from the data. To make this work, the input variables speed v, speed difference δv, net space headway g (gap), and acceleration A of the lead vehicle are discretized, and in each of the resulting bins a different acceleration distribution F_v,δv,g,A (a) is estimated. In most cases, the acceleration values are distributed according to a Laplace distribution. Missing data-bins are interpolated. This model is then tested; it is found, that the resulting distributions of safety surrogate measures reproduce the ones found in reality.