Impact of Visual Embodiment on Trust for a Self-driving Car Virtual Agent: A Survey Study and Design Recommendations

Designing trust-based in-car interfaces is critical for the adoption of self-driving cars. Indeed, latest studies revealed that a vast majority of drivers are not willing to trust this technology. Although previous research showed that visually embodying a robot can have a positive impact on the interaction with a user, the influence of this visual representation on user trust is less understood. In this study, we assessed the trustworthiness of different models of visual embodiment such as abstract, human, animal, mechanical, etc., using a survey and a trust scale. For those reasons, we considered a virtual assistant designed to support trust in automated driving and particularly in critical situations. This assistant role is to take full control of the driving task whenever the driver activates the self-driving mode, and provide a trustworthy experience. We first selected a range of visual embodiment models based on a design space for robot visual embodiment and visual representations for each of these models. Then we used a card sorting procedure (19 selected participants) in order to select the most significant visual representations for each model. Finally, we conducted a survey (146 participants) to evaluate the impact of the selected models of visual embodiment on user trust and user preferences. With our results, we attempt to provide an answer for the question of the best visual embodiment to instill trust in a virtual agent capacity to handle critical driving situations. We present possible guidelines for real-world implementation and we discuss further directions for a more ecological evaluation.