Incorporation of Robust Sliding Mode Control and Adaptive Multi-Layer Neural Network-Based Observer for Unmanned Aerial Vehicles

Abstract

The control and state estimation of Unmanned Aerial Vehicles (UAVs) provide significant challenges due to their complex and nonlinear dynamics, as well as uncertainties arising from factors such as sensor noise, wind gusts, and parameter fluctuations. Neural network-based methods tackle these problems by accurately approximating unknown nonlinearities through training on input-output data. This paper proposes an adaptive Multi-layer Neural Network (MLNN) Luenberger observer-based control for altitude and attitude tracking of a quadrotor UAV. The MLNN observer, employing a modified back-propagation algorithm, is used for the quadrotor's state estimation. The adaptive nature of the proposed observer helps mitigate the effects of parameters such as wind gusts, measurement noise, and parameter variations. Subsequently, a sliding mode controller is designed based on the observed states to track the reference trajectories. Lyapunov stability is ensured by using the modified back-propagation weight update rule for the proposed MLNN observer. Simulation results demonstrate superior tracking performance of the proposed observer compared to the Sliding Mode Observer (SMO) and a Single Hidden Layer Neural Network (SHLNN) observer, even in the presence of the aforementioned parameters.