Designing Affective External Human-Machine Interfaces for Autonomous Vehicles to Interact with Pedestrians in Shared Spaces

Abstract

Pedestrians are increasingly encountering autonomous agents in modern cities powered by artificial intelligence and robotics. This includes autonomous vehicles (AVs) that interact with pedestrians in various traffic situations. Significant research has explored how AVs can communicate, cooperate, and coexist with pedestrians without human drivers. In particular, external human-machine interfaces (eHMIs) have shown promise in facilitating interactions between AVs and pedestrians. Although many communication strategies for eHMIs have been studied, the potential of affective interfaces -- those designed to convey or influence emotions -- remains open for exploration. Affective interfaces, known for their abilities to enhance the perception towards intelligent systems and the experience of users, have the potential to assist AV-pedestrian interactions.

This doctoral research proposes affective external human-machine interfaces (affective eHMIs) as a sub-category of eHMIs. The overarching aim is to understand how affective eHMIs can improve the interaction of AVs with pedestrians. This thesis achieves this aim by first reporting on background research, which includes a field observation of naturalistic pedestrian-vehicle interactions in urban shared spaces, a systematic review of emotionally expressive non-humanoid robots, and focus groups with domain experts to explore relevant use cases. This phase seeks to identify potential situations and methods for designing affective eHMIs. Following this, the thesis details a research-through-design phase where affective eHMIs were designed, prototyped, and evaluated through two empirical case studies related to urban shared spaces: one focusing on the emotional expressions of AVs and the other on the affective states of pedestrians. Furthermore, in advancing virtual reality test environments for eHMIs, this thesis reports on techniques and considerations for prototyping affective eHMIs in VR, validated by empirical studies.