Ilmu Komputer & AI editorial
AI-Driven Multi-Hop Relay Selection for Smart Urban NR-V2X Networks via Learning-to-Optimize Graph Neural Networks
The core problem
Reliable and low-latency NR-V2X communications are essential for smart mobility in dense urban environments. However, limited Road-Side Unit (RSU) density, frequent non-line-of-sight conditions, and highly dynamic vehicular topologies often prevent many Connected and Automated Vehicles (CAVs) from maintaining stable single-hop connectivity. Although multi-hop relay-assisted communication can extend infrastructure coverage, selecting relay links in real time under practical flow, capacity, and connectivity constraints remains challenging.
Mixed-Integer Linear Programming (MILP) yields optimal multi-hop relay decisions, but its computational complexity scales sharply with network density, limiting real-time applicability. To address this, the authors propose a Learning-to-Optimise (L2O) framework based on Graph Neural Networks (GNNs) for real-time NR-V2X relay selection. The framework aims to enable cost-effective enhancement of urban V2X connectivity by leveraging existing vehicular assets and supporting scalable, real-time NR-V2X operation in smart city environments.
Innovation
Experiments on large-scale urban datasets generated by an integrated SUMO--GEMV2 simulation pipeline show that the proposed approach achieves connectivity comparable to that of the MILP oracle while reducing execution time by orders of magnitude. The GINE model delivers near-constant inference latency, making it suitable for real-time NR-V2X operation even as network density increases.
Key quantitative findings include:
- **Connectivity:** The GNN-based relay selection matches the MILP oracle in terms of the number of connected CAVs and overall network connectivity, with negligible performance loss.
- **Execution time:** Inference time remains nearly constant across different network sizes, whereas MILP solve time grows sharply with the number of vehicles and candidate links.
- **Scalability:** The approach scales to dense urban scenarios with hundreds of CAVs and multiple RSUs, enabling real-time decision-making.
These results demonstrate that the L2O framework can effectively replace computationally expensive optimization with a fast neural approximation, unlocking practical deployment in smart city environments.
Why it matters
The paper addresses a critical bottleneck in NR-V2X communications: the trade-off between optimality and real-time feasibility. By leveraging Learning-to-Optimize with Graph Neural Networks, the authors show that it is possible to retain the quality of MILP-based relay selection while achieving the speed required for dynamic vehicular networks.
The use of an edge-aware GINE is particularly suitable because relay selection is inherently an edge-classification problem: the decision of whether to activate a candidate radio link depends on both endpoint nodes and the link's propagation-aware features. The graph representation naturally captures the relational structure of vehicular networks, including multi-hop paths and interference patterns.
A key strength of the approach is its reliance on an offline MILP oracle for supervision, which avoids the need for hand-crafted heuristics and enables the model to learn complex, constraint-aware policies. The integrated SUMO--GEMV2 simulation pipeline ensures that the evaluation reflects realistic urban mobility and channel conditions.
Limitations and future work may include extending the framework to handle even larger-scale networks, incorporating additional constraints such as latency and reliability guarantees, and exploring online adaptation to changing traffic patterns. Nevertheless, the proposed framework enables cost-effective enhancement of urban V2X connectivity by leveraging existing vehicular assets and supporting scalable, real-time NR-V2X operation in smart city environments.
Who should read this
Opening member content…