基于深度迁移强化学习的无人机投放自主引导机动控制算法

Autonomous Guidance Maneuvering Control Algorithm for UAV Dropping Based on Deep Transfer Reinforcement Learning

针对无人机精确投放引导问题,本文提出基于深度迁移强化学习的无人机投放自主引导机动控制算法,分别建立基于马尔可夫决策过程的引导机动决策模型、引导机动评估模型等,并设计基于迁移学习和课程学习的引导机动策略训练方法,拟合基于深度学习的引导机动策略和评估网络,最后开展仿真训练和验证试验。仿真结果表明,该算法实现了无人机在任意姿态和位置条件下,能够自主规避区域威胁并自主引导至目标投放点,成功完成投放瞄准任务,有效地提升了无人机投放引导机动控制的自主性。

Aiming at the problem of long-distance guidance for precise dropping of UAVs,this paper proposed the autonomous guidance maneuvering control algorithm for UAV dropping based on deep transfer reinforcement learning.This paper established the guidance maneuvering decision-making model for UAV dropping based on Markov decision processes.Specifically,it constructed an improved reward model for evaluating the action obtained by the algorithm we proposed based on traditional aviation fire control theory.And it presented the training process of the autonomous guidance maneuvering policy for UAV dropping based on transfer learning and curriculum learning.It constructed the autonomous guidance maneuvering policy and evaluation networks based on deep learning.Finally,the simulation results show that this algorithm achieves autonomous flight of UAV in any initial position/attitude to avoid threats in the mission area and towards the target point,ultimately performing the targeting of the dropping,and effectively improving the autonomy of maneuvering control during the UAV dropping guidance process.