Investigating drivers’ trust in autonomous vehicles’ decisions of lane changing events

Abstract

It is potential to improve the interaction between autonomous vehicles (AVs) and drivers by calibrating drivers’ trust in AVs. In this study, we investigated drivers’ trust in AVs’ decisions of changing lanes on a six-lane highway. We derived the AV lane changing scenarios using a machine learning model. The scenarios were rated by 250 participants recruited from Amazon Mechanical Turks (AMTs) in a survey study. The study was designed as a mixed-subject design where the between-subject variable was the amount of information presented (i.e., 3, 4, 5, 6, 7 pieces of information) and the within-subject variable was the information display format (i.e., tabular or visual forms). The results showed that 1) mental demand was always lower in the visual display compared to the tabular one, 2) trust and risk seemed to be inversely proportional across conditions, and 3) 4, 5, or 6 pieces of information tended to be preferred better than others. These results provide design implications on calibrating trust in AV systems by involving the driver in the decision-making process.