AI-Driven Real-Time Relay Optimisation in Smart Urb... | AI Research

AI-Driven Real-Time Relay Optimisation in Smart Urban NR-V2X Networks via Learning-to-Optimise Graph Neural Networks
This paper addresses the challenge of maintaining stable, low-latency connectivity for Connected and Automated Vehicles (CAVs) in dense urban environments. Because urban obstacles and limited infrastructure often block direct communication between vehicles and Road-Side Units (RSUs), the authors propose an AI-driven framework that uses neighboring vehicles as relays to extend the reach of the network. By shifting from traditional, computationally heavy optimization methods to a machine learning approach, the system can make real-time decisions to keep vehicles connected as they move through complex city landscapes. The same ai systems question is explored in TRACE, which adds a research perspective.

Modeling Networks as Graphs

The researchers represent the vehicular network as a dynamic graph where vehicles and RSUs are nodes, and the radio links between them are edges. This structure allows the system to capture both the physical location of vehicles and the quality of their wireless connections, such as signal strength and capacity. By treating the network as a graph, the AI can "reason" about the entire environment, identifying the best multi-hop paths to route data from a vehicle to an RSU while respecting constraints like traffic demand and link capacity.

Learning to Optimize

To train the AI, the authors first used a traditional mathematical method called Mixed-Integer Linear Programming (MILP) to find the "perfect" relay decisions for a large dataset of urban scenarios. They then trained a specialized Graph Neural Network (GNN) to learn these optimal patterns. This GNN acts as a "Learning-to-Optimise" (L2O) engine; instead of solving complex equations from scratch every time a vehicle moves, the trained model can predict the best relay configuration almost instantly by processing the current graph state. The same ai systems question is explored in Kernel-Managed Shared Memory for System-Wide Personalization, which adds a research perspective.

Performance and Real-Time Speed

The study shows that this AI-based approach is highly effective, recovering up to 11.3% more connectivity compared to relying on direct, single-hop communication. Crucially, the model is significantly faster than traditional optimization. While the MILP solver becomes slower as the number of vehicles increases, the GNN provides near-optimal results with near-constant, millisecond-level latency. This speed is essential for real-world applications in smart cities and Industry 4.0, where communication decisions must be made in real time to adapt to fast-changing traffic and signal conditions.

Important Considerations

While the framework demonstrates strong performance in realistic simulations across various urban layouts—such as dense city centers and arterial roads—the authors note that the current validation is based on simulated radio and mobility environments. The research has not yet been tested in physical field trials or live, operational NR-V2X deployments. Additionally, the system's ability to maintain these connectivity gains is dependent on the overall network load; as infrastructure reaches capacity, the potential for improvement naturally decreases. The same ai systems question is explored in FlashVector, which adds a research perspective. as detailed in the full paper on Arxiv