TY - GEN
T1 - VLM-driven Risk-Adaptive HUD Interactions for Trust Calibration in Automated Driving
AU - Ren, Mengyang
AU - Zheng, Pai
AU - Colley, Mark
N1 - Publisher Copyright:
© 2026 Copyright held by the owner/author(s).
PY - 2026/4/13
Y1 - 2026/4/13
N2 - Automated driving trust depends on timely and intelligible feedback about driving risk. However, static “one-size-fits-all” Head-up display (HUD) designs struggle to match rapidly changing road contexts. We propose a vision-language model (VLM) approach that infers driving-scene risk and adapts HUD visualization parameters within the parameterized design space introduced by OptiCarVis. We implemented and compared four HUD parameterization strategies (static baseline; LLM-generated static; VLM-driven dynamic HUD (one-step); and VLM-driven dynamic HUD (two-step) with explicit intermediate risk tags) in an online study (N=22). It indicated significant differences in cognitive load, trust, and perceived safety across conditions. In particular, the two-step dynamic condition yielded higher trust and perceived safety and lower cognitive load than the one-step dynamic condition, suggesting that separating risk interpretation from parameter generation can produce more interpretable adaptations.
AB - Automated driving trust depends on timely and intelligible feedback about driving risk. However, static “one-size-fits-all” Head-up display (HUD) designs struggle to match rapidly changing road contexts. We propose a vision-language model (VLM) approach that infers driving-scene risk and adapts HUD visualization parameters within the parameterized design space introduced by OptiCarVis. We implemented and compared four HUD parameterization strategies (static baseline; LLM-generated static; VLM-driven dynamic HUD (one-step); and VLM-driven dynamic HUD (two-step) with explicit intermediate risk tags) in an online study (N=22). It indicated significant differences in cognitive load, trust, and perceived safety across conditions. In particular, the two-step dynamic condition yielded higher trust and perceived safety and lower cognitive load than the one-step dynamic condition, suggesting that separating risk interpretation from parameter generation can produce more interpretable adaptations.
KW - AR-HUD
KW - Automated Driving
KW - Multimodal Explanation
KW - Proactive Interaction
KW - Trust Calibration
KW - VLM
UR - https://www.scopus.com/pages/publications/105038085421
U2 - 10.1145/3772363.3798997
DO - 10.1145/3772363.3798997
M3 - Conference article published in proceeding or book
AN - SCOPUS:105038085421
T3 - Conference on Human Factors in Computing Systems - Proceedings
BT - CHI 2026 - Extended Abtracts of the 2026 CHI Conference on Human Factors in Computing Systems
A2 - Oliver, Nuria
A2 - Shamma, David A.
A2 - Candello, Heloisa
A2 - Cesar, Pablo
A2 - Lopes, Pedro
A2 - Artizzu, Valentino
A2 - Draxler, Fiona
A2 - Lopez, Gustavo
A2 - Reinschluessel, Anke V.
A2 - Tong, Xin
A2 - Toups Dugas, Phoebe O.
PB - Association for Computing Machinery
T2 - Extended Abtracts of the 2026 CHI Conference on Human Factors in Computing Systems, CHI 2026
Y2 - 13 April 2026 through 17 April 2026
ER -