A Penetration Method for UAV Based on Distributed Reinforcement Learning and Demonstrations

The penetration of unmanned aerial vehicles (UAVs) is an essential and important link in modern warfare. Enhancing UAV’s ability autonomous through machine learning has become a research hotspot. However, the current generation strategies for UAVs faces problem excessive sample demand. To reduce demand, this paper proposes combination policy (CPL) algorithm that combines distributed reinforcement demonstrations. Innovatively, action CPL jointly determined by initial obtained from demonstrations target asynchronous advantage actor-critic network, thus retaining guiding role training. In complex unknown dynamic environment, 1000 training experiments 500 test were conducted related baseline algorithms. results show smallest highest convergence efficiency, success rate among all algorithms, strong robustness environments.