08 Juin – Thesis defense - Clarisse Lawson-Guidigbe

10 h Salle 0108 - ENSC-Bordeaux INP (Talence)

Anthropomorphic virtual assistant for trust in automated driving.

This thesis work was devoted to exploring the usage of two technologies that will eventually transform our daily lives: virtual assistants and autonomous cars.
Virtual assistants already have an important place in our lives and are revolutionizing the way we interact with systems through voice interaction. With a simple sentence, a virtual assistant can get us information about the weather or play music. Autonomous cars, although not yet available to the public, hold the promise of improving driving comfort, reducing accidents, and improving traffic flow on the roads. However, the adoption of such technology requires trust from users. It seems that virtual assistants, by the very nature of their anthropomorphic interface, can play a role in this context. We therefore explore the potential of virtual assistants to increase trust in autonomous driving.
The main questions addressed in this work concern, on one hand, the design choices for a virtual assistant so that it is perceived as anthropomorphic and trustworthy. On the other hand, we address the impact that such an interface can have on the perception of anthropomorphism and user trust when included in an autonomous car HMI. To answer these questions, we first chose the visual appearance of the assistant by evaluating the impact of different visual representations on the perception of anthropomorphism and trust. We chose a Mechanical-Human representation. Then, we implemented this representation in three dimensions and integrated the result in a driving simulator as a hologram. To evaluate the virtual assistant, we conducted an experiment which compared a baseline interface without virtual assistant with two interfaces integrating two versions of the virtual assistant. The results show that the perception of anthropomorphism does not increase with the level of anthropomorphism. A significant correlation confirms the impact of perceived anthropomorphism on trust. More surprising results concerning the influence of the virtual assistant on users’ performance or the impact of acquired experience on trust are discussed.

Event localization