ABSTRACT The rapid integration of mobile ad hoc networks (MANETs), the Internet of Things (IoT), and 6G technologies is creating highly dynamic, decentralized communication systems that demand secure, intelligent, and adaptive routing. Traditional routing methods struggle to cope with frequent topology changes, diverse device behavior, and increasing security threats. To address these challenges, this paper proposes a novel routing framework called NC‐RADTGNN (non‐convolutional return‐aligned decision transformer graph neural network). Unlike conventional deep learning approaches, the proposed model captures complex network structures without relying on convolution operations and aligns routing decisions with long‐term performance goals rather than short‐term actions. A lightweight cleaner fish optimization algorithm is used to fine‐tune learning parameters, improving convergence and efficiency. Additionally, a multi‐criteria decision‐making (MCDM) mechanism dynamically selects blockchain consensus protocols to enhance trust and security among distributed nodes. Experimental evaluation on 50,000 routing samples demonstrates that NC‐RADTGNN achieves 99.9% routing accuracy, 0.1% model loss, and the lowest latency (22.5 ms) among recent state‐of‐the‐art approaches. It also reduces routing overhead, control packet load, and link breakage while improving throughput and delivery ratio. Overall, the proposed framework provides a robust, secure, and scalable routing solution for future MANET‐IoT‐6G environments, combining intelligence, adaptability, and trust management within a unified architecture.
Ramesh et al. (2026) studied this question.