Robust Adaptive Neural Network Tracking Control for Quadrotor Unmanned Aerial Vehicle via Reinforcement Learning Strategy

Abstract

This paper proposes an adaptive event-triggered trajectory and attitude tracking control framework for quadrotor unmanned aerial vehicle (QUAV) with external disturbances. To handle unknown uncertainties in QUAV control system, we propose a dual neural network (NN) architecture: combining reinforcement learning with disturbance estimation for real-time disturbance compensation. Specifically, the Actor-NN generates compensation signals to offset uncertainties, while the Critic-NN dynamically evaluates control performance to adjust the learning process. A nonlinear neural network disturbance observer (NNDO) is incorporated to estimate the lumped total disturbances in real time. By combining backstepping control approach with event-triggered mechanism, the proposed control strategy achieves rigorous closed-loop stability with guaranteed exclusion of Zeno behavior. Experimental validation on QUAV demonstrates the effectiveness of the proposed scheme in balancing computational efficiency and control performance.OPEN ACCESS Received: 30/08/2025 Accepted: 12/12/2025

Document