A Gain Scheduling Attitude Controller With NN Supervisor for Quadrotor UAVs

Abstract

The use of quadrotor helicopters is highly demanded in both military and commercial fields. Depending on the application area, special requirements may be needed for the maneuverability of these unmanned aerial vehicles. Traditional controller structures cannot meet these demands, leading developers to seek innovative solutions. In this study, a gain-scheduling controller is proposed to maintain the performance requirements for the attitude control of quadrotor helicopters. In this context, different operating points are selected based on a proposed rule in the working region of the quadrotor helicopter, and a controller design is carried out after a necessary transformation of the linear models. Furthermore, the robustness of quadrotors against time-varying disturbances is ensured through the implementation of a supervisor neural network (NN) controller. Specifically, a radial basis function NN (RBFNN) is employed, trained using gradient descent-based supervised learning. The response of the proposed controller is compared numerically with the traditional proportional-derivative controller's response, where the proposed controller, which is not so different from conventional controllers in terms of processor power, provides the desired performance criteria in a wide operating region. Finally, the gain-scheduling controller improved with the feedforward NN controller is tested under time-varying disturbances, resulting in a significant reduction in errors.