WiP: Concept Drift Management in Edge-AI for Autonomous Vehicles

As autonomous vehicle (AV) technology evolves, the integration of Edge-AI has become crucial to enable real-time decision-making, reduce latency, and improve operational safety. This paper reviews the current literature on managing concept drift, anomaly detection, and machine learning models in resource-constrained Edge-AI environments for AVs. Techniques such as Online Sequential Extreme Learning Machine (OS-ELM), reinforcement learning, and federated learning (FL) are discussed for their effectiveness in handling the dynamic nature of AVs. In addition to highlighting challenges related to concept drift and adversarial threats, this paper suggests potential novel approaches for managing concept drift in AVs: a hybrid drift detection mechanism combining sequential learning with deep reinforcement learning, FL frameworks with crosslayer security, and the use of neurosymbolic AI to ensure flexible, intelligent, and secure drift management. These solutions aim to enhance the robustness, adaptability, and security of AV systems, preparing them to operate safely in increasingly complex and unpredictable environments.