# Building Trust Profiles in Conditionally Automated Driving

**Authors:** Lilit Avetisyan, Jackie Ayoub, X. Jessie Yang, Feng Zhou  
**arXiv:** [2306.16567](https://arxiv.org/abs/2306.16567)  
**Date:** 2023-06-28  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2306.16567v1/

## Key points

- Trust is crucial for the acceptance and adoption of automated vehicles (AVs)
- Research aims to investigate trust profiles for personalized experiences in AVs
- Study conducted using a driving simulator with three different conditions: control, false alarm, and miss
- Researchers measured drivers' dispositional trust, initial learned trust, dynamic trust, personality traits, and emotions
- Three distinct trust profiles identified: believers, oscillators, and disbelievers
- Validation model achieved impressive F1-score of 0.90 and accuracy of 0.89
- Individual factors influence trust profiles providing valuable insights into persona's perspective
- Understanding these factors can help design personalized in-vehicle trust monitoring systems
- Tailored approaches to monitor and adjust drivers' trust levels based on unique profiles can improve safety and enhance overall experience of automated driving
- Foster greater acceptance and usage of AVs while ensuring safety and security on the roads

## Summary

Trust is a crucial factor in the acceptance and adoption of automated vehicles (AVs). Without trust, drivers and the public may be hesitant to use AVs. This research aims to investigate trust profiles in order to create personalized experiences for drivers in AVs. By understanding drivers' dynamic trust from a persona's perspective, we can better design systems that cater to individual needs. The study was conducted using a driving simulator, where participants were asked to take control from automated driving in three different conditions: a control condition, a false alarm condition, and a miss condition. Each condition consisted of eight takeover requests (TORs) in various scenarios. The researchers measured drivers' dispositional trust, initial learned trust, dynamic trust, personality traits, and emotions. Using a K-means clustering model, the researchers identified three distinct trust profiles: believers, oscillators, and disbelievers. To validate this model, they built a multinomial logistic regression model based on SHAP explainer. This model selected the most important features for predicting trust profiles with an impressive F1-score of 0.90 and accuracy of 0.89. Furthermore, the study explored how different individual factors influenced trust profiles providing valuable insights into the dynamics of trust from a persona's perspective. Understanding these factors can help in designing personalized in-vehicle trust monitoring and calibrating systems that adjust drivers' trust levels accordingly. The findings of this research have significant implications for improving safety and enhancing the overall experience of automated driving by developing tailored approaches to monitor and adjust drivers' trust levels based on their unique profiles which will foster greater acceptance and usage of AVs ultimately contributing to the widespread adoption of AV technology while ensuring safety and security on the roads.