*Limitations*

A limiting factor in this study may be seen in its virtual reality nature within a simulator setting. This may provide an impression and sensation of safety which would not be experienced in reality and result in a more comfortable and less critical driver state in the simulator. We sugges<sup>t</sup> that the direction of the effect is the same though the effect size may be smaller.

We would expect the NASA TLX questionnaire to lead to a higher level of workload when facing the same situation in a vehicle in a real-world scenario like on the road or a test track. This influence is expected to be lower for the Expression of Trust questionnaire due to the research focus on the difference between pre- and post-evaluation. In both cases, the participants are aware of the virtual character in the simulator. Therefore, the participants' rating focuses on the complete simulator session rather than on a single scenario.

Moreover, we wish to point out that the number of participants was not completely balanced between all groups analysed in the calculations. This results in a small number of participants within some specific groups and may lead to a limited generalisation for specific results while not harming the big picture of the findings.

Measuring the effects of AD on human perception was concluded on a highly advanced system, which may lead to a limited generalisation regarding the behaviour of the system within a high situational variance in the real world.

## **5. Conclusions**

While the driving simulator offers a chance to immerse in AD functionality, it comes with the limitation posed by the virtual environment. The missing risk of harm can support the participant to gain sufficient impressions of the AD behaviour and therefore significantly increase the participant's trust in the system. When comparing the analysis of participants' impressions before and after the simulator session, the conducted experiment shows that this increase of trust holds for all demographic subgroups on a similar level. For example, the mean increase in trust resulting from the described experience in the driving simulator is measured 3.8% for study participants with and 13.7% without previous ADAS experience. Concerning the gender diversion an increase of trust of 5.4% for the male and 14.5% for the

female participants was measured. Group analyses do not reveal significant differences. Hence, we sugges<sup>t</sup> a reduced need for precise and equally balanced demographic groups. Single significant differences within the subgroups are only recognised when using single inquiry lines (specific aspects within the questionnaires).

According to the participants' subjective perception, relying on the system usability scale, the system was able to provide sufficient performance (77.49%) for the conducted evaluation in a reliable virtual environment, substantiated by a homogeneous distribution of the demographic group mean ratings on the questionnaire on a high level.

For SAE level 4 AD in a simulator, participants' subjective workload appears to be low on all singular aspects of the NASA TLX. This reflects the definition of the automated driving mode with no objective task for the driver in the operational design domain. The factors age and driving experience have a significant influence on the participant's workload. Further, the measurement provided evidence that there is a significant relationship between the trust in the system and the participant's workload. In essence, the collected experimental data consistently indicates that the desired increase of trust (as a proxy to improve user acceptance) could be supported by reducing the need for user workload in vehicles.

At present, the integration of the highly automated and autonomous operation of safety-critical systems is still in an early stage. Therefore, an industry-agreed state of the art for measuring trust in autonomous or AI-based systems does not ye<sup>t</sup> exist. Standards addressing the specific aspects of autonomous and/or AI-based technology for safety-critical systems are under development, providing essential further steps in such directions. Although this paper is not directly focusing on standardisation, the presented measurements addressing the establishment of trust and definition of measurable parameters create a foundation for further support for designers and engineers in the conception of safety-critical autonomous or AI-enhanced applications.

Future investigations into the complex combination of human drivers in an AD personal transportation system are required, using statistical analysis in combination with AI to understand and improve the future user trust in the AD technology.

**Author Contributions:** Conceptualization, P.C.; methodology, P.C.; software, H.D. and M.P.; validation, P.C. and C.K.; investigation, P.C., H.D. and M.P.; resources, P.C. and O.V.; writing—original draft preparation, P.C., C.K. and O.V.; writing—review and editing, P.C., C.K., O.V., H.D. and G.M.; visualization, P.C.; editing, A.E.; supervision, A.E.; project administration, P.C.; funding acquisition, O.V. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the European Union's Horizon 2020 research and innovation program under gran<sup>t</sup> agreements No 723324 (TrustVehicle) and No 871385 (TEACHING). The publication is funded by the Open Access Funding by the Graz University of Technology.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of the University of Surrey protocol code 353003-352994-38756154 and date of approval 27 August 2018.

**Informed Consent Statement:** Informed consent was obtained from all participants involved in the study.

**Data Availability Statement:** The scenarios conducted during the study are available at [57], the datasets of the results are not available in public as they are restricted to the informed consent and the GDPR rules provided to and signed with all participants. They are stored in a restricted data area of AVL List GmbH as stated to the subjects during the study for no longer than is necessary for the purposes of the study.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.
