Guided Exploration Reinforcement Learning for UAV Pursuit-Evasion Games
摘要
This paper proposes a novel pursuit strategy for a quadrotor-type UAV pursuit evasion game. The novel guidance law considering 3-dimensional and quadrotor UAV maneuvers is proposed as a baseline and expert knowledge for UAV pursuit-evasion games. In this paper, we propose guided exploration reinforcement learning (GE-DDPG) to enhance convergence time and exploration efficiency compared to pure DDPG with Gaussian noise and behavior cloning. The simulation results show that the proposed method outperforms the baseline guidance law and other DDPG methods in capture time and performance in UAV maneuvering scenarios.