Developing a model of driver's uncertainty in lane change situations for trustworthy lane change decision aid systems

Inspired by the “emancipation” theory of trust, this paper proposes to develop driver's trust in assistance systems based on the assumption that driver's appropriate trust in these systems can be built, when the support of assistance systems is adapted to the drivers' uncertainty state and helps reducing their uncertainty. For example, a trustworthy lane change assistance system is supposed to provide support to the driver during a lane change maneuver by adapting to the state of the driver's uncertainty about distance gaps and closing speeds in respect to the surrounding traffic. The precondition for such a system is a model of driver's uncertainty, which can be used to recognize driver's uncertainty states in lane change situations. This paper mainly presents the development of a probabilistic model for classifying driver's uncertainty in lane change situations. Using experimental data obtained in a simulator experiment, we considered three Bayesian networks: a naive Bayesian classifier, a Tree-Augmented-Naive Bayesian classifier, and a fully connected Bayesian Network. Based on the Bayesian Information Criterion and Accuracy metrics, the Tree-Augmented Naive Bayesian classifier was chosen to predict driver's uncertainty in lane change situations.