State Space LS-SVM as a Disturbance Observer in Sliding Mode Control of a Quadrotor UAV

This paper proposes the approach of employing state space least squares support vector machine (SS LS-SVM) as a disturbance observer in the sliding mode control of a quadrotor. SS LS-SVM, which was recently introduced by the authors, is adopted for the disturbance estimation task in this study. A quadrotor type unmanned aerial vehicle is considered as the system of interest to apply and assess the performance of SS LS-SVM as a disturbance observer. Quadrotor continuous time mathematical model is taken into account in a standard integrator based on Euler discritization. Both parametric uncertainties and external disturbances are lumped in a disturbance term and added to the nominal model. That term is approximated by SS LS-SVM in an output error prediction context by minimizing the state estimation error via gradient descent as the training method. The proposed disturbance observer works in collaboration with a standard nonlinear observer. It is only necessary for estimating the system states using the measured system output while SS LS-SVM performs the estimation of disturbance. SS LS-SVM enables placement of a native LS-SVM directly in a state equation. Simulation results indicates the significant performance of closed loop disturbance estimation by the SS LS-SVM disturbance observer and based on that, robustness of the employed control method is empowered.