Non-linear autopilot design for agile air vehicles with neural networks

A model based method is presented which uses supervised artificial neural networks for autopilot design of agile air vehicles with highly non-linear and rapidly time-varying dynamics. Model Reference Control (MRC) techniques applying neural networks are utilized to design a lateral controller for a generic air vehicle. The neural controller is verified in a nonlinear three degree-of-freedom (3DoF) simulation. A sequence of lateral acceleration commands and a scenario where the air vehicle system is corrupted by untrained non-stochastic disturbances demonstrates the performance and generalization properties of the designed autopilot.