Hybrid Deep MARL and Swarm Intelligence for Multi-Role UAV Swarms Serving 6G in Transiently Crowded Areas

Future 6G networks supposedly provide both highcapacity and reliable connectivity in extremely dense and dynamic environments such as stadiums, concerts, squares, parks, and emergency gatherings. Simply over-provisioning static terrestrial infrastructure to accommodate these transient traffic peaks is economically inefficient. Furthermore, such sites usually demand local services and supports for the crowd's activities. This paper therefore proposes a multi-role UAV swarm architecture in which unmanned aerial vehicles (UAVs) embedding 6G base-station. They jointly deliver aerial access and venuelevel services. Realistic examples may be lighting, real-time crowd safety analytics, and on-demand edge processing. We construct a hybrid control framework that combines deep multi-agent reinforcement learning (MARL) and swarm-intelligence (SI). The former optimizes high-level decisions related to 3D UAV placement, load balancing, and resource allocation. In parallel, the latter governs local emergent behaviors including formation keeping, collision avoidance, and fast re-clustering around mobile user hotspots. Channel models for A2A (air-to-air) and A2G (air-to-ground) links are concretely described and integrated into the learning scheme. Extensive simulations demonstrate that the proposed hybrid scheme significantly reduces service latency and user outage probability. Especially, it effectively improves user throughput, while considerably enhancing safety of operations.