Semantic‐Aware UAV‐IRS Control for Cooperative V2X Perception Using Label‐Derived Context

ABSTRACT

Cooperative V2X perception requires timely, reliable delivery of perception‐critical packets, especially when occlusion, dense traffic, and vulnerable road users increase scene risk. Existing UAV‐assisted intelligent reflecting surface (IRS) studies mainly optimize channel estimation, beamforming, phase control, trajectory design, or communication‐layer resource allocation, whereas the semantic urgency of perception packets is rarely coupled with aerial and reflective assistance. This paper proposes a label‐driven semantic‐aware control framework that maps DAIR V2X‐Seq‐SPD annotations into packet priority, latency budgets, and reliability targets, then uses a UAV‐IRS‐V2X simulator to select UAV position, discrete IRS assistance mode, resource fraction, and link strategy. The study is entirely simulation‐based and does not involve real RF measurements, physical IRS hardware, UAV flight trials, or deployed V2X implementation. Experiments use a 500‐sample subset with 24,338 object‐level records, seven communication‐layer baselines, two controllers, matched‐action‐space comparisons, ablation analysis, confidence intervals, and sensitivity checks. Compared with the strongest semantic‐priority allocation baseline, the optimized candidate‐search controller reduces average latency by 42.21% and semantic‐weighted latency by 41.86%, while improving reliability‐target satisfaction by 28.37%. It also reduces UAV movement by 70.17% relative to the rule‐based controller. These results show that annotation‐derived semantic requirements can improve simulated cooperative V2X packet‐delivery control.