| description abstract | A learning-based guidance method is proposed to address the problem of continuously avoiding multiple online-detected no-fly zones for hypersonic cruise vehicles. Compared with previous research on the no-fly zone avoidance problem, this paper further considers the challenges posed by non-global information and the variation in the number of no-fly zones. The method comprises two components: the approach for the design and offline training of a reinforcement learning agent with heading decision-making capabilities, and the cruise guidance framework based on a multiagent coordination strategy. Firstly, considering the adaptability to a variety of tasks and training efficiency, the Markov decision process for solving the no-fly zone avoidance problem is designed. On this basis, by setting up training environments with progressive difficulty, the agent interacts with environments to complete multistage training and gradually improves the heading decision-making ability for the no-fly zone. During the guidance process, each detected no-fly zone is assigned to a trained agent to make independent heading decisions, and these agents form a coordination committee to determine the final heading command through the coordination strategy. Then the cruise guidance framework implements the commands of heading, altitude, and velocity. A series of training and testing experiments are conducted. The theoretical analysis and simulation results demonstrate the proposed method’s efficacy, robustness, and adaptability. | |
