Agentic Beam Optimization for mmWave V2X Communications Using Fluid Antenna Systems

Abstract

Millimeter-wave (mmWave) vehicular networks are a key enabler of beyond-5G intelligent transportation systems but are highly vulnerable to fast mobility, dynamic blockages, and rapid topology variations, which often cause beam misalignment and quality-of-service (QoS) degradation. Data-driven beamforming based on artificial intelligence (AI) enables low-latency prediction but lacks robustness in highly dynamic environments, whereas optimization-based approaches require significant computation and are unsuitable for real-time deployment. To address these challenges, we propose an agentic AI model with the grey wolf optimizer (GWO) for robust beam management in mmWave vehicular networks. The model operates as an autonomous networking agent that perceives environmental dynamics, including mobility and LOS/NLOS conditions, and performs goal-driven beam selection and adaptation. In the model, a neural network provides fast beam prediction based on QoS requirements, while GWO refines backup beams to improve recovery from beam misalignment. Our scheme leverages a dual-array fluid antenna system, which provides hardware agility through adaptive frequency and radiation-pattern tuning. It also employs Hamming-window-based antenna weighting to suppress sidelobes and inter-beam interference. SHAP analysis identifies SINR, packet delivery ratio, and throughput as the most influential features. Simulation results achieve an $R^{2}$ score of 0.9878 and significantly lower MSE and RMSE compared with conventional machine learning approaches.