Reinforcement Learning Control for Hypersonic Morphing Flight Vehicle with Identification of Dynamic Parameter

A reinforcement learning (RL) controller with identification of the dynamic parameter of hypersonic morphing flight vehicle (HMFV) is proposed in this paper, successfully realizing the end-to-end control of attack angle in the longitudinal plane. The following improvements are made in this paper: Firstly, the dynamic parameter (rudder efficiency coefficient) of the flight vehicle is added into the state vector, so that the RL controller can understand the control ability of the rudder and generates the optimal control commands in the current state. Secondly, five instead of only one consecutive attack angle deviations are used to jointly generate the state vector, which enables the RL controller to use the model state information of the previous period of time and improve the control stability. Three simulations are set up in this paper. The simulation results show that the RL controller proposed in this paper can achieve stable and high precision attack angle control under large-scale environmental deviations and has strong generalization under different guidance commands.