Advancing our understanding of Human-Autonomy Teams (HATs) is critical for evaluating the safety of Automated Driving Systems (ADS) in real-world operations. From a Human Reliability Analysis (HRA) perspective, Performance Shaping Factors (PSFs) such as trust, attention, and task load influence how human drivers and automated agents jointly perform Dynamic Driving Tasks (DDTs), particularly during control transitions. This paper focuses on the longitudinal evolution of trust in automation within a broader research effort aimed at developing a causal HAT model representing the relationships between PSFs and safety outcomes in conditional driving automation.

Driving simulator-based experiments were conducted in a mid-fidelity setup to collect both subjective and objective data on driver-ADS teaming during takeover scenarios. Scenarios involved silent automation failures characterized by tailgating behavior, with and without warnings, under low- and high-traffic conditions. Questionnaires were administered before and after exposure to the simulated scenarios to capture driver profile characteristics, perception of the driving environment, and trust in driving automation. These subjective ratings were combined with objective performance measures, including vehicle kinematics, reaction times, crash rates, and frequency of hard braking events, to assess the impact of agent- and scenario-level PSFs on safety performance.

Longitudinal trust in automation was measured using a 12-item questionnaire adapted from the Trust in Automated Driving (TiAD) scale, employing a 7-point Likert response format. Changes in trust were analyzed using two complementary methods: (1) Wilcoxon Signed-Rank tests with effect size estimation, and (2) Cumulative Link Mixed Models (CLMM) to account for the ordinal nature of the data and subject-level variability. Results indicate statistically significant decreases in trust following exposure to automation exhibiting failure behaviors, particularly in relation to contextual complexity and system reliability. Among the 50 valid experiments collected, participants were less likely to select higher trust responses post-exposure, even when accounting for individual differences and item-level effects. Not all trust dimensions shifted equally, suggesting heterogeneous sensitivity across trust components. These findings highlight the dynamic nature of trust in driver-ADS HATs and its measurable association with safety-relevant outcomes, supporting the integration of trust-related PSFs into probabilistic risk assessment frameworks for automated driving systems.