In the operation of autonomous connected vehicles, continuous monitoring of their behavior is essential to detect anomalies before deviations from normal operations propagate into failures.
As systems evolve—due to over-the-air updates, configuration changes, and shifting workloads—the definition of normal behavior also changes, leading to the gradual degradation of static diagnostic methods.
Existing approaches typically address either automated model adaptation or operator integration in isolation, rather than as a coordinated supervisory loop.
This paper presents an online anomaly detection framework that integrates three coordinated mechanisms:
First, a factorized deep Q-network with self-attention selects the most suitable detector from a candidate pool for each monitored service, leveraging inter-service dependencies in the microservice topology.
Second, an ensemble of three statistical drift detectors monitors the input distribution and raises an alarm only when all three concur, prioritizing precision over recall.
Finally, a human-in-the-loop retraining mechanism, based on a pending transition buffer and a 60/40 prioritized replay strategy, allows operators to incorporate expert knowledge while preserving the system's learned response to prior data distributions.
The framework is evaluated on a connected-vehicle testbed running an automated valet parking application across seven backend microservices.
The attention-augmented agent achieves an F1 score of 0.69, compared to at most 0.11 for any single detector applied uniformly.
Following a real software update that induces measurable concept drift, F1 drops to 0.52; after operator-triggered retraining, performance recovers to 0.65 on the new distribution while remaining at 0.69 on the prior one, demonstrating sustained adaptation without catastrophic forgetting.
Blogger's Review: This paper offers an innovative anomaly detection solution by integrating multiple mechanisms, showcasing the complexity and challenges of modern intelligent vehicle systems, especially in dynamic environments. The combination of reinforcement learning and human feedback enhances the system's responsiveness and accuracy, making it a valuable approach for real-world applications.