Longitudinal effects of visualizing uncertainty of situation detection and prediction of automated vehicles on user perceptions

This paper explores the impact of uncertainty visualizations in automated vehicle (AV) functionality on user perceptions over a three-day longitudinal study. Participants (N=50) watched real-world driving videos twice daily, in the morning and evening. These videos depicted morning and evening commutes, featuring visualizations of AVs' pedestrian detection, vehicle recognition, and pedestrian intention prediction. We measured perceived safety, trust, mental workload, and cognitive load using a within-subjects design. Results show increased perceived safety and trust over time, with higher ratings in the evening sessions, reflecting greater predictability and user confidence in AV by the study's end. However, inconsistencies in pedestrian detection and intention prediction led to mixed reactions, highlighting the need for visualization stability and clarity refinement. Participants also desired a feature indicating the AV's intended path and options for manual intervention. Our findings suggest transparency and usability in AV visualizations can foster trust and perceived safety, informing future AV interface design. • Computer vision models visualized automated vehicles' detection, prediction, and planning. • Longitudinal study investigating the impact of visualized uncertainties. • Three morning and evening sessions regarding user trust, perceived safety, and cognitive load. • Trust and perceived safety increased over time, evening higher than morning. • Mixed reactions to inconsistencies in pedestrian detection and intention prediction.