Operational Design Domain Exit Detection and Risk Assessment for Roadworks Scenarios in Autonomous Vehicles

Operational Design Domains (ODDs) are increasingly used as a foundation for arguing the safety of autonomous vehicle (AV) deployments, yet concrete workflows for monitoring ODD compliance and handling ODD exits in complex environments remain scarce. This paper proposes a camera-based workflow for detecting ODD exits caused by roadworks and assessing the associated risk in order to support the selection of an appropriate Minimum Risk Maneuver (MRM). A CNN-based object detector is fine-tuned on the ROADWork dataset to identify roadwork-related objects, from which three frame-wise metrics are derived: the distance to the nearest detected object, an area-weighted per-frame probability of roadworks existence, and a risk metric that incorporates both distance and existence probability information, including the presence of dynamic actors such as workers and work vehicles. To increase robustness against detection noise and misclassifications, these signals are temporally smoothed using exponential moving averages. The workflow is evaluated qualitatively on two representative scenarios, where the AV either approaches a work zone along a straight road or encounters it suddenly after a turn. The results demonstrate that the proposed framework can detect ODD exits in advance, generate interpretable risk profiles, and provide a structured basis for integrating MRM decision logic into an ODD monitoring system.