Analysis of adaptive control using on-line neural networks for a quadrotor UAV

Abstract

This paper presents an unmanned quadrotor flight controller, which has its advantages in robust structure and versatile missions, by implementing the dynamic model inversion technique with adaptive neural network. The model inversion of nonlinear dynamic system is conducted via feedback linearization, and the resultant model inversion error is compensated by direct adaptive control. Overall controller adopts PD controller with 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">nd</sup> order command filter to treat state variables, and neural network is augmented in order to ameliorate the performance of PD controller. To be specific, the type of adaptive controller employed in this paper is Sigma-Pi neural network, considering its simplicity and rapid applicability to online adaptation. Furthermore, the stability of neural network output is guaranteed by Lyapunov stability function. The final designed flight controller is simulated using pre-built quadrotor dynamics. The result of simulation shows the performance of position and attitude control, and can be analyzed with comparison to classical PID controller and model-inversed controller without online neural network.