Möhlenhof, ThiesThiesMöhlenhofJansen, NormanNormanJansen2025-05-222025-05-222024https://publica.fraunhofer.de/handle/publica/48789010.1109/MILCOM61039.2024.107739442-s2.0-85214588978This research explores the application of policy gradient methods in multi-agent reinforcement learning, augmented with offline reinforcement Learning techniques. The goal is to leverage these methods to improve communication of military command and control information systems (C2IS) in real-time emulated radio networks by the use of agents deployed on each network node in a decentralized manner. These agents have the task of improving the performance of the network by learning to adapt the local C2IS and communication services on their designated nodes to the prevailing dynamic network conditions. The proposed method is assessed in an emulated environment, where agents are trained to augment the network performance by controlling the transmission rate of a set of Blue-Force Tracking Services within an emulated ad-hoc radio network in real-time.enfalseAd-Hoc Radio Network EmulationMulti-Agent Reinforcement LearningNetwork OptimizationReal-Time Reinforcement Learningconference paper