Neural network-based observer combined with nonsingular terminal sliding mode control for QUAV tracking: Experimental...

This study integrates the adaptive neural network technique, state and disturbance observer, and nonsingular terminal sliding mode control framework to achieve rapid trajectory tracking for a quadrotor under state uncertainty, dynamic disturbances, and unknown dynamics. The control methodology begins by modeling the quadrotor's dynamics, including inherent uncertainties and disturbances. A state observer is designed to estimate unknown states while rejecting uncertainties and disturbances, with exponential convergence of observer errors demonstrated via Lyapunov theory. The tracking error between the observed quadrotor's states and the desired trajectory is subsequently defined, and a nonsingular terminal sliding mode surface is employed to stabilize this error. Using Lyapunov-based analysis, it is proven that the sliding surfaces converge to zero at an exponential rate. Furthermore, an adaptive neural network strategy is developed to approximate unknown components, including state variables, uncertainties, and disturbances. Simulation and experimental results using a quadrotor with realistic parameters validate the proposed control strategy, confirming its robustness and effectiveness.