Measurement and prediction of driver trust in automated vehicle technologies: An application of hand position transition probability matrix

Abstract

Driver trust has a great impact on the intention to accept, use, and adapt to automated vehicles. To date, driver trust in automated vehicle technologies has mostly been estimated by subjective data. Currently available objective measures of driver trust primarily rely on self-reported ratings as the ground truth, while the largely reported inconsistency between drivers’ self-reported trust levels and observed actual behavior suggests that drivers’ trust measures shall not solely rely on subjective reporting values. To address the issue, this study proposes an objective method to assess and predict driver trust in automated vehicle technologies, by using the transition probability matrix of drivers’ hand positions during automated vehicle system engagement. An on-road experiment was conducted on public roadways in real traffic. Data on use frequencies of advanced driver assistance systems (ADAS) and self-reported trust ratings were collected and combined in classifying driver trust levels: lower, medium, and higher based on the K-means clustering results. Drivers’ hand positions during ADAS engagement (i.e., during lane-departure warning and lane-keeping assist system uses) were then found closely associated with their trust levels. Differences of frequencies and transition probabilities for hand positions were further compared within and across the three trust groups. Results showed that drivers from the lower, medium, and higher trust groups were more likely to keep hands on the top, mid, and low positions of the wheel, respectively. Factors affecting driver trust in automated vehicle technologies were also explored through mixed model analyses. Middle-aged drivers placed more trust in ADAS than younger drivers, while female drivers exhibited greater trust than male drivers. The Random Forests algorithm was applied to build a prediction model for driver trust in automated vehicle technologies, by inputting the hand position transition probability matrix, age, gender, and ADAS types as independent variables. The overall prediction accuracy was approximately 80%. Findings in this study could contribute to the objective and real-time estimations of driver trust in automated vehicle technologies.