Attention-Based Multi-Objective Control for Morphing Aircraft

This paper proposes a learning-based joint morphing and flight control framework for avian-inspired morphing aircraft. Firstly, a novel multi-objective multi-phase optimal control problem is formulated to synthesize the comprehensive flight missions, incorporating additional requirements such as fuel consumption, maneuverability, and agility of the morphing aircraft. Subsequently, an auxiliary problem, employing (Formula presented.) -constraint and augmented state methods, is introduced to yield a finite and locally Lipschitz continuous value function, which facilitates the construction of a neural network controller. Furthermore, a multi-phase pseudospectral method is derived to discretize the auxiliary problem and formulate the corresponding nonlinear programming problem, where open loop optimal solutions of the multi-task flight mission are generated. Finally, a learning-based feedback controller is established using data from the open loop solutions, where a temporal masked attention mechanism is developed to extract information from sequential data more efficiently. Simulation results demonstrate that the designed attention module in the learning scheme yields a significant 53.5% reduction in test loss compared to the baseline model. Additionally, the proposed learning-based joint morphing and flight controller achieves a 37.6% improvement in average tracking performance over the fixed wing configuration, while also satisfying performance requirements for fuel consumption, maneuverability, and agility.