“Watch out!”: Prediction-Level Intervention for Automated Driving

Abstract

It seems that autonomous driving systems are substituting human responsibilities in the driving task. However, this does not mean that vehicles should not interact with their driver anymore, even in case of full automation. One reason is that the automation is not yet advanced enough to predict other road user's behavior in complex situations, which can lead to sub-optimal action choices, decrease comfort and user experience. In contrast, a human driver may have a more reliable understanding of other road users’ intentions which could complement that of the automation. We propose a framework that distinguishes between four levels for interaction with automation. Based on the framework, we introduce a concept which allows drivers to provide prediction-level guidance to an automated driving system through gaze-speech interaction. Results of a pilot user study show that people hold a positive attitude towards prediction-level intervention as well as the gaze-based interaction method.