Self-Tuning Proportional Double Derivative-Like Neural Network Controller for a Quadrotor

Abstract

In this paper, a self-tuning proportional double derivative-like neural network nonlinear adaptive controller for attitude tracking control of an unmanned aerial vehicle (UAV) is presented. The proposed scheme consists of neural networks with two neural nodes in the hidden layer and includes activation feedback. The error between the desired angle set-point and the output as well as the change of error is selected as the controller input. The optimal initial weight parameters are obtained by employing an adaptive ant colony optimization. The proposed controller can online tune the weight parameters of the hidden layer with a stable learning rate based on the controller input. The effect of the learning rates on the stability of the neural network controller was analyzed. The designed controller was developed based on a nonlinear model of an UAV with quaternion representation in the presence of parametric uncertainties and external disturbances. Simulation results demonstrate the validity and effectiveness of the proposed algorithm with different reference attitude signals.