Multi-Session User Experience Assessments of Computationally Optimized Automated Vehicle Functionality Visualizations

This study investigates how to optimize the design of automated vehicle (AV) functionality visualizations—such as trajectory planning, object detection, and situational awareness—to improve passenger experience. The researchers employed a Human-in-the-Loop (HITL) Multi-Objective Bayesian Optimization (MOBO) framework. This computational method iteratively adjusts design parameters (e.g., icon size, transparency, and visibility of specific data) based on user feedback to balance competing objectives like perceived safety, trust, and cognitive load.

Unlike previous studies that focus on single-session design, this research conducted a three-day longitudinal study (N=74) to observe how user preferences evolve. Participants were assigned to different design strategies, including expert-defined designs, user-customized designs, and various MOBO initialization methods (cold-start vs. warm-start). A key experimental variable was whether the MOBO algorithm remained active across all three days to continuously refine the design or if the design was "frozen" after the first day.

The results demonstrate that keeping the HITL MOBO active across multiple sessions leads to sustained improvements in performance-related metrics, including trust, predictability, and perceived safety. While initial designs were relatively similar across conditions on the first day, the iterative optimization process allowed the system to better align with individual user needs over time. The findings suggest that an initial personalization phase followed by periodic re-optimization is a scalable and efficient way to manage the diverse and evolving preferences of AV passengers.

As AVs become more common, the "one-size-fits-all" approach to dashboard design is likely to fail due to the highly personal nature of trust and safety perceptions. By automating the design process through HITL MOBO, manufacturers can create interfaces that adapt to individual users without requiring constant manual configuration, ultimately increasing the acceptance and usability of automated driving systems.