New interpretable AI method makes aircraft control 10x more efficient

A new paper from researchers at the University of Stuttgart tackles a fundamental challenge in flight control: how to allocate commands across multiple control surfaces when aircraft dynamics become nonlinear and coupled. Traditional linear control allocators fail in aggressive maneuvers due to model mismatch, while high-fidelity onboard models are too slow and black-box AI lacks interpretability. The authors propose using Sparse Identification of Nonlinear Dynamics (SINDy) to learn an explicit, physics-constrained analytical model of control effectiveness from flight data.

The resulting mapping is compact enough for real-time execution inside nonlinear optimization solvers, and it naturally incorporates actuator dynamics. An online adaptation layer monitors prediction residuals and refreshes the model when the plant changes (e.g., after actuator failures). On a high-fidelity nonlinear aircraft benchmark performing aggressive maneuvers, the method achieved accuracy comparable to a full nonlinear onboard model while reducing computational cost by orders of magnitude relative to established baselines. This work bridges the gap between model-based and data-driven control, offering both the interpretability needed for certification and the performance demanded by next-generation overactuated aircraft.