06.1 - Flight Dynamics and Control (Control & Modelling)
LEARNING-BASED FULL-ENVELOPE AUTOPILOT DESIGN BASED ON THE KOOPMAN OPERATOR THEORY
Y. Lee¹, J. Kim¹, Y. Kim¹; ¹Seoul National University, South Korea
A data-driven approach to designing an autopilot for missile is proposed based on the Koopman operator theory. The proposed approach employs neural networks to approximate the Koopman embedding function. A linear quadratic integral controller is designed for the embedded system, yielding an optimal controller capable of addressing a broad operational spectrum with only a few gain tunings.