ScholarWorks@성균관대학교

초록

This study proposes a novel Fail-safe mechanism for autonomous vehicles that is designed to minimize excessive emergency stops and unnecessary control interventions during sensor fault situations, while ensuring that a Minimal Risk Maneuver (MRM) is executed swiftly when actual risk accumulates. To this end, we define an exponentially accumulating risk assessment function that accounts for both the duration and the criticality of sensor failures. The decision-making logic is implemented using a Behavior Tree structure, enabling intuitive debugging of Fail-safe transitions through its modular, node-based flow. For validation, we constructed a CARLA-based digital twin environment. Using OpenStreetMap data, we generated a 3D simulation map of a complex commercial district in Korea and implemented a sensor fault interface to replicate diverse failure scenarios. Experimental results show that the proposed Fail-safe mechanism successfully satisfied the fault handling time intervals, completing MRM execution within 100 ms for critical sensors and 120 ms for noncritical sensors.

키워드