This paper describes a method for selecting autopilot parameters in flight. Here, the method is applied to a two-axis PID autopilot, but the process is relevant for other control architectures as well. A parameterized controller is constructed which selects gains based on the results of an autonomous real-time global aerodynamic modeling procedure. This enables the vehicle to choose its control strategy based on the learned dynamics of the system. The process is intended to speed up the autopilot certification process by combining the modeling and control design processes while reducing the required amount of ground testing. The technique is demonstrated in simulation and on flight test data of a subscale aircraft. The nonlinear simulation results show that the system follows the desired response, and the flight test results highlight some of the benefits of using the Learn-to-Fly methodology during development.
展开▼
