Conditional disturbance utilization-based intelligent morphing and flight control for morphing aircraft using reinforcement learning

Embodied intelligence (EI) advances beyond the traditional perception-decision-control serial paradigm by interacting with the physical environment, offering a new intelligent perception and control framework for the aerospace system development. Morphing aircraft are ideal platforms for the study and application of EI, as their adjustable aerodynamic configurations significantly influence perception and control processes. This paper addresses the control problem of morphing aircraft under wind disturbances and treats disturbances as usable resources. First, a dual-layer architecture of intelligent morphing-flight control is constructed, decoupling forewing and aft wing morphing while maintaining system stability. For wind disturbances, a nonlinear disturbance observer is used to incorporate disturbance estimation and diagnosis into the reward functions of conditional disturbance utilization (CDU). This helps guide the aircraft in utilizing environmental disturbances or generating additional thrust for disturbance compensation. In addition, this paper introduces a progressive multistage training scheme enabling the aircraft to gradually adapt to the environment and learn to exploit beneficial disturbances. This significantly accelerates the convergence of reinforcement learning based on a dueling double deep Q-network. Simulation results demonstrate that the proposed intelligent morphing-flight control framework excels in optimality and robustness. The findings also verify the morphing aircraft’s ability to utilize disturbances actively. And the effectiveness of CDU, inspired by EI, is validated through comprehensive simulation results.