The “Second Workshop on Trust in the Age of Automated Driving” was held at the Bahen Center for Information Technology/University of Toronto on September 23rd 2018. We were proud that so many researchers (nearly 30) from all around the globe participated, altogether we had a great atmosphere and good vibrations!

 

Six break out groups worked intensively in two breakout sessions. In the first breakout session, we encouraged participants to create new (or adapt existing) definitions for trust in automation that fit the domain of automated vehicles. Participants were instructed to choose one of the layers of trust as provided by Hoff and Bashir [1] (Dispositional Trust, Situational Trust, and Learned Trust) provide a definition, as well as how trust in this context/layer could eventually be measured.

 

As a result, the groups came up with different taxonomies and ideas (see images below). For dispositional trust, participants emphasized the importance of personality traits (such as motivation to drive manually, religion, country, etc.), which could be measured with questionnaires or behavioral measures. Considering situational and learned trust, personal risk behavior was identified as one of the most important factors, while objective measurements could use eye tracking, or physiological measurements. One breakout group thought of modelling subjective trust from former experiences similar to a reward-function known from reinforcement learning, where the discount-factor defines for a given user, to what degree experiences in the past influence trust in the present.

 

 

We used considerations of workshop participants to revise some parts of our ongoing survey especially addressing trust in automated vehicles, which can be found at: www.automationtrust.net

 

Research on trust in automation is receiving more and more attention as it is becoming one of the key issues that could hinder the success of automated vehicles (AVs). Since users of AVs will be everyday consumers who are not necessarily domain experts, new methods must be elaborated to prevent both disuse (emerging from undertrust) and misuse (resulting from overtrust), and to calibrate trust to a level appropriate for a safe use of the technology. The importance of coping with varying levels of trust is thereby also dependent on the level of automation. For example, recurring incidents with the Tesla Autopilot, or the crash of the Uber self-driving taxi are suspected to be (at least partly) a result of overtrust. Thus, overtrust becomes a key issue especially in lower levels of automation, where human users are the main control authority responsible for monitoring the system, and apparently, some put more trust in the capabilities of the automated system than they should for safe operation.

Conversely, as soon as large numbers of fully automated vehicles are pervading the streets, overcoming distrust will

likely become more important. Since exploiting the full benefit of AVs and C-ITS technology (increased throughput,

less congestion, etc.) demands a high ratio of automated vehicles, technology skeptics must eventually be convinced

of the advantages. To deal with both under- and overtrust, researchers have proposed various strategies. However, it is currently unclear how all these findings can be combined together to provide trust calibration for a great variety of different users and environments. Clearly, methods aiming to increase and decrease trust cannot be utilized simultaneously, and further it must be determined which (if any) method can support a specific driver for accomplishing the potentially best result.

 

This workshop will address contemporary issues surrounding trust in technology in the challenging and constantly changing context of autonomous vehicles. In particular, this workshop focuses on three main aspects: (1) formulation

of a comprehensive set of definitions for trust in automated vehicles; (2) Development of interface approaches for mitigating overtrust and undertrust issues; (3) Identification of an appropriate timing of trust-related cues. The outcome of this workshop will provide a benchmark for future work in the field and is also intended to inspire joint publications among the participants.

 

When and Where?

Sunday, September 23

9:00AM - 1:00PM

 

AutoUI 2018

Toronto, Canada

Process

Key topics will be presented to the audience by the workshop organizers, followed by an introduction round of all participants (including accepted position papers). Afterwards, four break out groups discuss the proposed workshop goals.