Sliding-Mode Disturbance Observer-Based Robust Model Predictive Control Using Sparse Identification of Nonlinear Dynamics

Existing Sparse Identification of Nonlinear Dynamics (SINDy)–based flight-control studies for fixed-wing aircraft have offered limited treatment of longitudinal–lateral coupling and explicit disturbance estimation, while nonlinear model predictive control remains highly sensitive to aerodynamic-model fidelity. This study develops a physics-informed SINDy framework and an integrated nonlinear model predictive control (NMPC) architecture with a supertwisting sliding-mode disturbance observer for fixed-wing unmanned aerial vehicles. Known physical terms, including gravity and thrust, are retained, whereas uncertain aerodynamic components are identified from limited, noisy flight data and embedded consistently in both the NMPC predictor and the observer. High-fidelity flight simulations using a wind-tunnel-derived ground-truth aerodynamic model are used for validation. Over the disturbance-active interval, the proposed controller reduces trajectory root-mean-square errors by factors of 2–3 relative to controllers without a disturbance observer. Compared with a least-squares-model-based controller equipped with a linear extended state observer, the proposed framework reduces the average root-mean-squared error by 20.6% over the full trajectory and by 14.9% over the disturbance-active interval.