TinyRoute: secure adaptive routing for intelligent transportation system using TinyML and few shot learning

Intelligent Transportation Systems (ITS) needs secure and adaptive communication mechanisms for addressing the dynamic vehicular mobility, routing instability, and increasingly sophisticated cyber-attacks in Vehicle-to-Vehicle (V2V), Vehicle-to-Infrastructure (V2I), and Vehicle-to-Everything (V2X) environments in which the recent VANET routing and intrusion detection approaches have improved routing efficiency or attack detection individually. However, it generally lack an integrated framework capable of simultaneously supporting adaptive routing, lightweight edge intelligence, and reliable zero-day attack detection under highly dynamic traffic conditions. To address these issues, this paper proposes TinyRoute, a secure adaptive routing framework that integrates Geometric Deep Reinforcement Learning (Geo-DRL), Honey Badger Optimization (HBO), TinyML, and a CNN LSTM-based Few-Shot Learning Intrusion Detection System (FSL-IDS) in which the Geo-DRL performs mobility-aware route selection but HBO refines routing paths by avoiding malicious and congested regions. The lightweight on-device anomaly inference within autonomous vehicles is enabled by TinyML. With only the limited samples, the cloud based FSL-IDS detects both known and zero-day attacks respectively. Experimental evaluation demonstrates that TinyRoute achieves a routing latency of 28.4 ms, a Packet Delivery Ratio (PDR) of 96.7%, an intrusion detection accuracy of 95.2%, and a communication overhead of only 12.4 KB/s, outperforming recent state-of-the-art VANET routing and IDS approaches. These results demonstrate the effectiveness of the proposed framework for secure, adaptive, and low-latency ITS communication.