Structured Cross-Task Transfer Reinforcement Learning for Mobility Control based UAV Anti-Jamming Communications

Due to the flexible mobility capability of UAVs, mobility control has been considered as a method in UAV anti-jamming communication and resource management. While reinforcement learning (RL) has emerged as a promising solution for UAV mobility control, it faces an intrinsic learning efficiency issue. When mobility control is treated as a decision variable, frequent exploration between agents and the environment will incur an unacceptable mobility cost, which will drain the limited battery energy of UAVs. To address these challenges, this paper introduces an efficient RL approach for mobility based UAV anti-jamming communications. We first propose a quantitative framework to extract and model the underlying correlation relationships among different learning tasks. Then, a structured cross-task transfer RL (STRL) approach is proposed to quickly adapt and apply the knowledge gained in previous tasks to new ones, facilitating efficient transfer learning. The simulation results show that the proposed approach could derive the best mobility control policy without exploration in complex communication environments compared to several existing approaches.