Reinforcement Learning Based Topology Control for UAV Networks

The paper addresses efficient topology control for UAV swarms that provide wireless networking in environments lacking infrastructure. Multi-hop connectivity and UAV control present scalability challenges as fleet size grows. The authors propose a topology control system that analyzes relative UAV positions, optimizes connectivity considering interference and energy consumption, and reshapes the logical network by selecting per-UAV neighbors and mapping data flows. The key is adaptive connectivity optimization under dynamic conditions (e.g., user density, UAV power consumption), for which they employ reinforcement learning using DDPG so each UAV adjusts its connectivity. The system accelerates learning by flexibly changing the number of steps used for parameter learning when new UAVs are deployed. Performance is verified via simulations and theoretical analysis over multiple multi-UAV topologies, demonstrating improved energy efficiency and throughput while maintaining connectivity.