Mark Colley, member of the research group human-computer interaction, will defend his dissertation of the titel Calibrating Trust in Automated Vehicles - Theoretical, Design, and Empirical Insights into Effects of Visualizations on Trust . The jury consists of Prof. Dr Enrico Rukzio (Institute of Media Informatics, Ulm University), Prof. Dr Stephen Brewster (University of Glasgow), and Prof. Dr Wendy Ju (Cornell Tech, New York). Additional members of the committee are Prof. Dr Martin Baumann and Prof. Dr Matthias Tichy (both Ulm University).
Abstract: Automated vehicles (AVs) of SAE Level 3 and above are poised to significantly alter mobility by enabling non-driving activities and potentially reducing traffic accidents. However, user acceptance hinges on trust, which must be calibrated accurately to match the AVs' true functionalities to avoid overtrust and undertrust issues.
This thesis introduces the FACT-AV framework for calibrating trust in AVs. It emphasizes quantifying AV functionalities akin to human situation awareness processes: detecting, predicting, and planning maneuvers based on sensor inputs and data interpretation. FACT-AV uses metrics like intersection over union to quantify "Situation Detection", average displacement error for "Situation Prediction", and explores metrics for "Trajectory Planning" to enable true calibration.
The thesis then reports prior findings on how visualizations of AV functionalities impact trust, proposing designs that integrate uncertainty information across Situation Detection, Situation Prediction, and Trajectory Planning.
To validate these designs, thirteen empirical studies were conducted under scenarios like urban driving with pedestrians and complex automated intersections.
The visualizations generally enhanced understanding of AV functionalities but were sometimes found distracting.
A structural equation model analyzing all studies indicated that situation awareness and perceived safety positively influence trust, whereas mental workload negatively affects it. These insights are pivotal for developing AVs that users can trust and effectively engage with.
We invite all interested persons to attend the event.
