Adaptive trajectory tracking control for quadrotors with disturbances by using generalized regression neural networks

Abstract

In this document, the development and experimental validation of a nonlinear controller with an adaptive disturbance compensation system applied on a quadrotor are presented. The introduced scheme relies on a generalized regression neural network (GRNN). The proposed scheme has a structure consisting of an inner control loop inaccessible to the user (i.e., an embedded controller) and an outer control loop which generates commands for the inner control loop. The adaptive GRNN is applied in the outer control loop. The proposed approach lies in the aptitude of the GRNN to estimate the disturbances and unmodeled dynamic effects without requiring accurate knowledge of the quadrotor parameters. The adaptation laws are deduced from a rigorous convergence analysis ensuring asymptotic trajectory tracking. The proposed control scheme is implemented on the QBall 2 quadrotor. Comparisons with respect to a PD-based control, an adaptive model regressor-based scheme, and an adaptive neural-network controller are carried out. The experimental results validate the functionality of the novel control scheme and show a performance improvement since smaller tracking error values are produced.