Augmented naive Bayesian network for driver behavior modeling

The availability of a digital driver behavior model during emergency situations constitutes a major breakthrough dealing with active safety system tuning. This article presents a modeling approach based on an input-output system (initial conditions-driver's actions). The starting point of our work is a behavioral database gathered from a track experiment with common drivers. Subjects are confronted with the sudden braking of a released trailer, which they followed for a while. Our objective is to predict driver's actions following a set of initial conditions (distance to collision, speeds, and friction). The core of our model is an inference system based on augmented naive Bayesian network. This article outlines the various stages leading to the construction of this model. It discusses its robustness using another database.