Hierarchical Goal-Guided Learning for the Evasive Maneuver of Fixed-Wing UAVs based on Deep Reinforcement Learning

作者：Yuan, Yinlong; Yang, Jian; Yu, Zhu Liang; Cheng, Yun; Jiao, Pengpeng; Hua, Liang^*

来源：JOURNAL OF INTELLIGENT & ROBOTIC SYSTEMS, 2023, 109(2): 43.

DOI：10.1007/s10846-023-01953-9

• Deep reinforcement learning
• Fixed-Wing UAVs
• Penetration decision
• Hierarchical goal-guided learning
• XSimStudio

Fixed-wing unmanned aerial vehicles (UAVs) will play a vital role in forthcoming military conflicts. Effectively avoiding threats and improving the survivability of fixed-wing UAV in dynamic hostile environments are the keys to the success of combat missions. Hence, endowing fixed-wing UAVs with the ability to autonomously generate evasive maneuver is the primary problem that should be solved. With considering the threat of air-to-air missile attacks, this paper designs a novel hierarchical goal-guided learning (HGGL) method, which combines with traditional off-policy deep reinforcement learning (DRL) algorithms and endows the agent with the ability to evade a series of air-to-air missiles. The pivotal idea of the proposed algorithm is to use the hierarchical features of the goal, it improves the availability of training data to eliminate the limitation of the convergence rate of traditional DRL algorithms owing to sparse rewards. We demonstrate the performance of our algorithm in several simulation experiments. All experiments are applied on the XSimStudio platform. The results demonstrate that the proposed algorithm improves the convergence speed and outperforms the state-of-the-art traditional algorithms.

• 单位
  南通大学