- Updated: July 16, 2026
- 2 min read
Self-Adaptive Anomaly Detection with Reinforcement Learning and Human Feedback in Connected Vehicles
Self‑Adaptive Anomaly Detection with Reinforcement Learning and Human Feedback in Connected Vehicles
Connected vehicles are becoming the backbone of modern transportation, operating as autonomous cyber‑physical systems that must be continuously monitored for abnormal behavior. Traditional static diagnostic methods quickly become obsolete as vehicles receive over‑the‑air updates, configuration changes, and shifting workloads. To address this challenge, a new online anomaly detection framework has been developed that combines reinforcement learning, statistical drift detection, and human‑in‑the‑loop feedback.
The framework consists of three tightly coordinated mechanisms:
- Attention‑augmented Deep Q‑Network: A factorized DQN with self‑attention selects the most suitable detector from a pool for each microservice, exploiting inter‑service dependencies in the vehicle’s microservice topology.
- Ensemble Drift Detector: Three statistical drift detectors monitor input distributions and raise an alarm only when all three concur, prioritising precision over recall.
- Human‑in‑the‑Loop Retraining: A pending‑transition buffer with a 60/40 prioritized replay strategy lets operators inject expert knowledge while preserving learned responses to historic data.
In a real‑world testbed running an automated valet‑parking application across seven backend microservices, the attention‑augmented agent achieved an F1 score of 0.69, far surpassing the best static detector (F1 = 0.11). After a software update introduced concept drift, performance dropped to 0.52, but operator‑triggered retraining restored the score to 0.65 on the new distribution while retaining 0.69 on the previous one, demonstrating robust adaptation without catastrophic forgetting.
This solution aligns with the Ubos Tech AI‑driven solutions portfolio, offering a scalable, self‑adaptive security layer for connected‑vehicle ecosystems. Learn more about how Ubos Tech enables intelligent monitoring and rapid response to emerging threats in autonomous systems.
For further details, explore our blog or contact the Ubos Tech team today.
Andrii Bidochko
CTO UBOS
Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.