From Overtrust to Distrust: A Simulation Study on Driver Trust Calibration in Conditional Automated Driving

Abstract

Conditional automated driving delegates routine control to automation while keeping drivers responsible for supervision and timely takeovers. In this context, safety and usability hinge on calibrated trust, a state between overtrust and distrust that aligns reliance with actual system capabilities. We investigated how calibrated trust relates to concurrent behavior during conditional automation in a driving-simulator study (n = 26). After a brief familiarization block, drivers completed four takeover request (TOR) exposures while performing a non-driving-related task (NDRT). Trust was assessed with a validated multi-item inventory. NDRT engagement was operationalized as successful Surrogate Reference Task (SuRT) clicks per second, and takeover behavior was indexed by TOR reaction time (TOR-RT) from TOR onset to the first valid control input. The results showed that higher trust was associated with greater ND RT throughput during automated driving, whereas TOR-RT did not change significantly across repeated exposures, consistent with familiarization. In this sample, we did not observe a systematic penalty in TOR-RT associated with higher trust; however, confidence-interval benchmarks indicate that modest delays cannot be ruled out. This suggests that, after brief onboarding, calibrated trust can coexist with timely safety-critical responses within the limits of our design. These findings tentatively support interface and training strategies that promote calibrated trust (e.g., predictable TOR policies, transparent capability boundaries, and short onboarding) to help drivers navigate between overtrust and distrust.