Trust-Gated Reinforcement Learning-Based Secure Routing for Wireless Sensor Networks
TL;DR
Experimental results demonstrate that TG-RL-SR consistently improves routing robustness, achieving up to 20–25% higher packet delivery ratios under severe attack conditions while maintaining competitive energy efficiency.
Abstract
Wireless sensor networks (WSNs) are increasingly deployed in critical monitoring applications where secure and energy-efficient routing is essential. However, routing-layer attacks such as blackhole, sinkhole, and selective forwarding can significantly degrade network reliability. This paper proposes TG-RL-SR, a Trust-Gated Reinforcement Learning-based Secure Routing framework designed to enhance routing robustness in adversarial WSN environments. The proposed method formulates routing decisions as a reinforcement learning problem while constraining the action space using dynamically updated trust values. A lightweight linear Q-learning model and a risk-sensitive reward function are employed to balance packet delivery reliability, energy efficiency, and attack resilience. Extensive simulations are conducted under multiple attack scenarios and malicious node ratios using two baseline routing methods for comparison. Experimental results demonstrate that TG-RL-SR consistently improves routing robustness, achieving up to 20–25% higher packet delivery ratios under severe attack conditions while maintaining competitive energy efficiency. Additional analysis shows that the trust-gated action mechanism effectively reduces the selection of malicious next-hop nodes. These results indicate that the proposed approach provides a practical and lightweight solution for secure routing in resource-constrained WSN environments. Received: 19 January 2026 | Revised: 9 April 2026 | Accepted: 11 July 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement Data are available from the corresponding author upon reasonable request. Author Contribution Statement Rui Zheng: Methodology, Software, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Project administration. Dashdorj Yamkhin: Conceptualization, Visualization, Supervision. Junxu Wei: Validation. Huqiang Liu: Software, Validation, Data curation. Zagarzusem Khurelbaatar: Resources.