Reducing pedestrian mental workload: the effectiveness of anthropomorphic AR eHMI in mixed traffic scenarios

Abstract

Due to the lack of implicit communication clues provided by traditional human drivers, pedestrians face higher uncertainty when interacting with self-driving vehicles (AV). The anthropomorphic external human-computer interface (eHMI) has become a promising solution to bridge this communication gap. This paper studies how different degrees of anthropomorphizing (including personalization) affect the psychological load and decision-making of pedestrians when crossing the road in a mixed traffic environment. This study has carried out three studies: Study I is a scope definition overview, which aims to identify outstanding issues about the degree of anthropomorphism and the value of personalization, and to distill the design dimensions that need to be tested. The second study is a formative collaborative design experiment (N1=30), which uses artificial intelligence-assisted interfaces inspired by the most popular tag-based interactive tools to generate and evaluate the eHMI concept. While providing verbal reasons, participants choose the most intuitive and cognitively efficient UI-label combination to provide qualitative insights for personalized design. Study 3 (N=28) tested four conditions in virtual reality (VR): no external human-machine interface (C1), detailed avatar (C2), simplified face icon (C3) and interface designed by each participant in study 2 (C4). The mixed-effect logical regression model shows that after controlling the arrival time (TTA), C3 (β=0.24, p=.021) and C4 (β=0.86, p<.001) both significantly improved the acceptance of the gap, while C2 had no significant difference. The NASA-TLX score shows that anthropomorphic display reduces the perceived workload, among which personalized design has always been rated as the lowest demand. Results indicate that simplified anthropomorphic cues—rather than highly detailed avatars—reduce mental workload most effectively, and personalized AR designs further strengthen clarity without increasing unsafe crossing behavior.

Date of Award	15 Jul 2026
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Xu Sun (Supervisor)