Effects of Trust in Automation and Takeover System Type on Driver Takeover Behavior in Level 3 Conditionally Automated Driving: A human-in-the-loop driving simulation study

Level 3 conditionally automated driving permits engagement in Non-Driving Related Tasks, positioning driver trust as a critical factor that modulates monitoring behavior and takeover readiness. Although graded warning strategies (e.g., two-stage warnings incorporating a Monitoring Request) show promise in facilitating takeover transitions, existing research has largely examined warning design and trust in isolation. Consequently, the interactive mechanism between advanced graded warning strategies and driver trust states remains underexplored. This study conducted a driving simulation experiment utilizing a 2 (Trust: High vs. Low) ×2 (Warning: Single-stage vs. Two-stage) within-subjects design to evaluate the effects of these variables on takeover reaction time and collision rates. Results revealed a distinct contrast between response speed and safety outcomes. While both the two-stage warning condition and low trust levels significantly accelerated takeover reaction times, the safety benefits of the warning system were strictly modulated by trust levels. Specifically, statistical analysis indicated a significant interaction: low-trust drivers achieved the highest avoidance success rates under the twostage condition, whereas high-trust drivers exhibited a significant decline in safety performance under the same condition, despite faster reactions. These findings highlight an automation paradox: excessive trust can be counterproductive specifically when coupled with advanced graded warnings. The study suggests that the performance degradation in high-trust drivers may be attributed to a confirmation bias, where the pre-warning signal is misinterpreted as a routine status update. Consequently, the study advocates for adaptive takeover systems that dynamically calibrate warning strategies based on the driver's trust state to ensure the efficacy of human-machine interaction.

Keywords: conditionally automated driving, trust in automation, takeover warning systems, takeover performance, driving safety.