Multi-Agent Reinforcement Learning-Based Resource Sharing in Multi-UAV Wireless Networks

This paper investigates the resource sharing problem in a multi-unmanned aerial vehicle (UAV) wireless network by utilizing the multi-agent reinforcement learning (MARL) method. Specifically, the considered multi-UAV system involves two transmission modes, i.e., UAV-to-Device (U2D) mode and UAV-to-Network (U2N) mode, in which the U2D mode is allowed to reuse the spectrum of U2N mode to improve the spectrum efficiency. Then, we formulate an optimization problem to maximize the throughput of U2D links by jointly optimizing the channel allocation, power level selection, and UAV trajectory, while ensuring the communication quality of U2N links. Due to the highly complex and dynamic nature, as well as the challenging non-convex objective function and constraints, the resulting problem is hard to address. Accordingly, we propose a novel Multi-Agent Deep Deterministic Policy Gradient (MADDPG)-based resource allocation and multi-UAV trajectory optimization policy. Simulation results illustrate the efficacy of our method in improving the system transmission rate.