Chapter 5 - Adaptive PID Controller Using a Multilayer Perceptron Trained With the Extended Kalman Filter for an Unmanned Aerial Vehicle

Abstract

Conventional PID controllers use constant coefficients obtained empirically. Although these controllers are widely used for their simpleness they are not designed to control complex nonlinear systems, mainly because of their lack of adaptability. With this in mind, the following chapter presents an adaptive controller based on the PID principle using a neural network, which have shown adaptation capabilities. A multilayer perceptron is trained with the extended Kalman filter and the output of the network represents the system control input. In order to show its adaptability and effectiveness, simulations are presented on a quadrotor due to uncertainties, time delays, and unmodeled dynamics, typical of this kind of systems.