Towards Adaptive Autonomous Vehicle Systems: Considering Trust and Risk Perception during Failures

As autonomous vehicles (AVs) increasingly operate in unpredictable environments, their function is shifting from simply being modes of transportation to becoming active collaborators alongside human drivers. In turn, drivers must assume a cooperative role, working effectively with autonomous systems to achieve shared goals, most importantly, ensuring human safety. Despite considerable progress, real-world usage and testing of AVs continue to highlight vulnerabilities, particularly in safety-critical situations. This study examines how vehicle failure, the type of failure (security vs. mechanical), and the scenario context influence drivers' trust and risk perception toward AVs. Initially, six categories of failure scenarios were identified using data from the California Department of Motor Vehicles' disengagement reports. Subsequently, an online experiment was conducted, where participants experienced both baseline (normal operation) and failure drives. In the failure drive, participants encountered one specific failure type and scenario. The results revealed a substantial decrease in driver trust following the failure drive across all conditions. Higher perceived risk was associated with reduced trust and lower risk-taking behaviors. Scenarios were classified into high- and low-risk categories, with control-related issues having the most pronounced effect on increasing perceived risk. The experiment found no difference in trust or risk perception whether the failure was caused by a malicious intervention or a mechanical issue. These findings suggest that AV design should incorporate mechanisms for assessing and appropriately responding to varying risk levels in critical situations. Enhancing this capability will strengthen the collaborative relationship between humans and AVs, ultimately improving safety and reliability in human-machine teaming contexts.