Designing for Appropriate Trust in Automated Vehicles

Licentiate thesis, 2020

Automated vehicles (AVs) have become a popular area of research due to, among others, claims of increased traffic safety and user comfort. However, before a user can reap the benefits, they must first trust the AV. Trust in AVs has gained a greater interest in recent years due to being a prerequisite for user acceptance, adoption as well as important for good user experience. However, it is not about creating trust in AVs, as much as creating an appropriate level of trust in relation to the actual performance of the AV. However, little research has presented a systematic and holistic approach that may assist developers in the design process to understand what to primarily focus on and how, when developing AVs that assist users to generate an appropriate level of trust.

 This thesis presents two mixed-method studies (Study I and II). The first study considers what factors affect users trust in the AV and is primarily based on a literature review as well as a complementary user study. The second study, a user study, is built upon Study I and uses a Wizard of Oz (WOz) approach with the purpose to understand how the behaviour of an AV affects users trust in a simulated but realistic context, including seven day-to-day traffic situations.

The results show that trust is primarily affected by information from and about the AV. Furthermore, results also show that trust in AVs have primarily four different phases, before the user’s first physical interaction with the AV (i), during usage and whilst learning how the AV performs (ii), after the user has learned how the AV performs in a specific context (iii) and after the user has learned how the AV performs in a specific context but that context changes (iv). It was also found that driving behaviour affects the user’s trust in the AV during usage and whilst learning how the AV performs. This was primarily due to how well the driving behaviour communicated intentions for the users’ to be able to predict upcoming AV actions. The users’ were also affected by the perceived benevolence of the AV, that is how respectful the driving behaviour was interpreted by the user. Finally, the results also showed that the user’s trust in the AV also is affected by aspects relating to different traffic situations such as perceived task difficulty, perceived risk for oneself (and others) and how well the AV conformed to the user’s expectations. Thus, it is not only how the AV performs but rather how the AV performs in relation to different traffic situations.

Finally, since design research not only considers how things are, but also how things ought to be, a tentative explanatory and prescriptive model was developed based on the results presented above. The model of trust information exchange and gestalt explains how information affecting user trust, travels from a trust information sender to a trust information receiver and highlights the important aspects for developers to consider designing for appropriate trust in AVs, such as the design space and related variables. The design variables are a) the message (the type and amount of information), b) the artefact (the AV, including communication channels and properties) and c) the information gestalt, which is based on the combination of signals communicated from the properties (and communication channels). In this case, the gestalt is what the user ultimately perceives; the combined result of all signals. Therefore, developers need to consider not only how individual signals are perceived and interpreted, but also how different signals are perceived and interpreted together, as a whole, an information gestalt.