Neural network-based active disturbance rejection control and its application in quadrotor unmanned aerial vehicle trajectory tracking

This study proposes a novel neural network-based active disturbance rejection control (ADRC) scheme, enhanced with the adaptive moment estimation (ADAM) optimizer for precise trajectory tracking of quadrotor UAVs under disturbances. A cascaded ADRC architecture is designed for attitude control, featuring an inner loop that enables rapid attitude response and an outer loop that handles low-frequency disturbances and uncertainties. This study implements the online adaptive tuning of extended state observer (ESO) parameters via a radial basis function neural network (RBFNN), which dynamically adjusts observer gains as disturbances evolve. The integration of the ADAM optimizer accelerates RBFNN training compared to traditional backpropagation by leveraging gradient moment estimations for adaptive learning rates. This approach enables real-time rejection of time-varying disturbances and eliminates the need for manual parameter recalibration. The theoretical stability of the proposed system is rigorously proven using Lyapunov analysis. Hardware-in-the-loop experiments validate the superior performance of the proposed scheme in three scenarios.