Adaptive and Robust 6G IoT Networks for Assistive Systems via Graph-Augmented DRL

The convergence of sixth-generation (6G) wireless communication and the Internet of Things (IoT) is expected to fundamentally transform the development of next-generation assistive technologies by enabling ultra-reliable, low-latency, and AI-native communication services. 6G networks are envisioned to provide terahertz-band transmission, intelligent edge computing, integrated sensing and communication, and native artificial intelligence support, which collectively facilitate mission-critical applications. Assistive IoT devices—including wearable health monitoring systems, implantable biosensors, intelligent mobility aids, neuroprosthetics, and brain–computer interface (BCI) platforms—operate under stringent quality-of-service (QoS) and quality-of-experience (QoE) constraints. These systems demand guaranteed reliability, bounded latency, high data integrity, and energy efficiency while functioning in highly dynamic and heterogeneous network environments characterized by device mobility, fluctuating channel conditions, uneven computational capabilities, and limited battery resources. Traditional communication and resource management strategies based on static optimization, rule-based heuristics, fixed protocol stacks, or centralized control architectures are increasingly inadequate for the scale and complexity of 6G-enabled IoT ecosystems. Such approaches struggle to adapt to rapidly changing topologies, dense node deployments, and dynamic traffic patterns typical of assistive environments. Although Deep Reinforcement Learning (DRL) has emerged as a powerful tool for adaptive network optimization—enabling autonomous policy learning for routing, spectrum allocation, and resource scheduling—conventional DRL models typically treat network states as flat feature vectors. This representation neglects the intrinsic relational structure of communication networks, limiting scalability, robustness, and generalization in large-scale distributed systems. Performance evaluation demonstrates that the proposed GA-DRL framework significantly outperforms traditional heuristic-based approaches and standard DRL implementations. Specifically, the framework achieves notable reductions in end-to-end latency, improvements in packet delivery reliability, enhanced energy efficiency, and greater resilience under dynamic channel and mobility conditions. Furthermore, the graph-based representation enhances convergence stability and policy generalization across varying network scales. These findings highlight the effectiveness of graph-augmented learning in enabling proactive, topology-aware, and context-adaptive decision-making within AI-native 6G ecosystems. The proposed GA-DRL framework establishes a scalable and intelligent communication foundation for future assistive IoT infrastructures, supporting mission-critical healthcare and human-assistive applications in next-generation wireless environments.