Reinforcement Learning-Based Integrated Decision-Making and Control for Morphing Flight Vehicles Under Aerodynamic Uncertainties

This article presents an integrated decision-making and control framework of morphing flight vehicles with variable-span wings in the glide phase. The proposed framework comprehensively consists of morphing strategy, attitude control, and online aerodynamic uncertainties estimate, thus outperforms existing works at its capacity of adequately considering the interaction between morphing mechanism and control design. Furthermore, the introduction of deep deterministic policy gradient algorithm has the effect of reducing the dependence on a precise model to some extent. By introducing aerodynamic uncertainties into the training environment and employing estimate, the framework enhances decision-making adaptability. In addition, the decision-making method is designed to optimal a comprehensive performance index including lift-to-drag ratio and attitude tracking effect, thus ensuring the physical realizability. The effectiveness of decision-making and control is validated by simulation results.