Congestion-Aware Traffic and Beamforming Control for UAV Swarms in Structured Low-Altitude Corridors

Abstract

Coordinating Unmanned Aerial Vehicle (UAV) swarms in structured low-altitude airspace is difficult because traffic is confined to layered corridors with lane-level motion, finite inter-layer ramps, time-varying demand, and communication hotspots around merging zones. Here, large-scale coordination refers to high-density operations where hundreds to thousands of heterogeneous UAVs share limited routing, safety, and communication resources. This paper proposes a macro-micro collaborative framework that jointly optimizes traffic regulation, collision avoidance, and communication resource allocation. At the macroscopic level, a Residual Network (ResNet)-based global router selects congestion-aware corridor routes, while an intelligent topology controller activates dynamic ramps to relieve traffic peaks and guide Massive Multiple-Input Multiple-Output (MIMO) beamforming toward active transfer regions. At the microscopic level, layer-specific physics-constrained Dueling Deep Q-Network (D3QN) agents perform lane-changing and speed regulation under safe-headway guardrails. Simulations with 1,000 UAVs demonstrate reduced near-miss incidents and improved communication channel gains compared with conventional baselines, providing a scalable coordination paradigm for high-density Advanced Air Mobility (AAM) operations.