Robust 6G IoT Communication Models for Assistive Devices Using Graph-Augmented Deep Reinforcement Learning

Abstract

It is anticipated that the incorporation of sixth-generation (6G) wireless networks and the Internet of Things (IoT) into each other could be a game-changer in the creation of next-generation assistive technologies because it will provide ultra-reliable, low-latency, and intelligence-driven communication services. Assistive IoT devices such as wearable health monitors, implantable medical sensors, mobility aids, neuroprosthetics and brain-computer interface systems are subject to strong quality-of-service (QoS) constraints with highly dynamic network conditions, uneven capabilities of devices, topology variations brought by mobility and severe energy constraints. Traditional communication and resource management methods that rely on a static optimization or fixed protocols, or a centralised control, are becoming unsuitable to address these needs in multi-faceted and extensive 6G conditions. Even though deep reinforcement learning (DRL) has become one of the most promising methods to adaptively control networks, current DRL-based models do not learn the underlying network topology and inter-device dependencies to achieve scalability and robustness. To overcome such difficulties, this paper presents a Graph-Augmented Deep Reinforcement Learning (GA-DRL) framework of robust and intelligent 6G IoT communication adapted to assistive devices. The framework proposed directly represents the communication environment as an evolving attributed graph and uses a Graph Neural Network (GNN) to discover topology-sensitive state representations, which will then be used by a DRL agent in the learning of the optimal communication policies involved in routing, power management, scheduling, and avoidance of interference. Simulation results have been taken in massive dense and mobile 6G IoT environments and it is proved that the suggested GA-DRL framework can reach significant improvement in the end-to-end latency reduction, the reliability of the packet delivery, energy consumption, and the resistance with dynamic network conditions as compared to traditional heuristic-based techniques and standard DRLs. These results emphasise the success of graph-augmented learning in facilitating proactive, context-sensitive decision-making and makes the suggested framework a scalable and intelligent basis of future support infrastructure of assistive IoT systems in AI-native 6G-based environment.