EEG-based assessment of driver trust in automated vehicles

Tingru Zhang,Jinfeng Yang,Milei Chen,Zetao Li,Jing Zang,Xingda Qu

DOI: https://doi.org/10.1016/j.eswa.2024.123196

IF: 8.5

2024-01-19

Expert Systems with Applications

Abstract:Effective collaboration between automated vehicles (AVs) and human drivers relies on maintaining an appropriate level of trust. However, real-time assessment of human trust remains a significant challenge. While initial efforts have delved into the potential use of physiological signals , such as skin conductance and heart rate, to evaluate trust, limited attention has been given to the feasibility of assessing trust through electroencephalogram (EEG) signals. This study aimed to address this issue by using EEG signals to objectively assess driver trust towards AVs. A simulated driving experiment was conducted, where driver trust was manipulated by introducing different types of AV malfunctions. Self-reported trust ratings were collected and used to classify driver trust into three levels: low, medium, and high. A total of 420 time- and frequency-domain EEG features were extracted, and nine machine learning algorithms were applied to construct driver trust assessment models. Additionally, to explore the potential of developing cost-effective models with reduced feature inputs, this study developed trust models using features solely from single brain regions: frontal, parietal, occipital, or temporal. The results showed that the best-performing model, utilizing features from the whole brain and employing the Light Gradient Boosting Machine (LightGBM) algorithm, achieved an accuracy of 88.44% and an F1-score of 78.31%. In comparison, models based on single brain regions did not achieve comparable performance to the comprehensive model. However, the frontal and parietal regions showed important potentials for developing cost-effective trust assessment models. This study also performed feature analysis on the best-performing model to identify features highly responsive to changes in trust. The results showed that an increased power of beta waves tended to indicate a lower level of trust in AVs. These findings contribute to our understanding of the neural correlates of trust in AVs and hold practical implications for the development of trust-aware AV technologies capable of adapting and responding to driver's trust levels effectively.

computer science, artificial intelligence,engineering, electrical & electronic,operations research & management science