Primal‐Dual Neural Network Based Robust Model Predictive Control for Quadrotor UAV With Polytopic Uncertainties and External Disturbances

ABSTRACT This work proposes a Primal‐Dual Neural Network (PDNN) based Robust Model Predictive Control (RMPC) framework to address trajectory tracking challenges for quadrotor unmanned aerial vehicles (UAVs) under unmodeled dynamics, external disturbances, and nonlinearities. Firstly, a polytopic uncertain model capturing system uncertainties and external disturbances is established. A cascade control strategy is then introduced to decouple underactuated dynamics and mitigate nonlinear coupling effects. An RMPC design with a model‐uncertainty compensation strategy is introduced to reject the model uncertainties. The compensation‐based RMPC problem is formulated as a quadratic programming (QP) problem, and a PDNN optimization method is developed to ensure fast computation solutions for real‐time control. The input‐to‐state stability (ISS) of the closed‐loop system under PDNN‐RMPC is rigorously proven. Experimental results demonstrate that the proposed PDNN‐RMPC framework achieves a reduction in computation time compared to the interior‐point method. Furthermore, under varying external disturbances, compared with existing tube‐based MPC and Recurrent Neural Network (RNN)‐MPC, the proposed PDNN‐RMPC respectively improves the average of 4.23% and 3.25% for position tracking accuracy, by 38.70% and 39.17% for attitude tracking accuracy. These results highlight the ability of the proposed PDNN‐RMPC in balancing real‐time performance, robustness, and energy efficiency, providing a practical solution for quadrotor UAV flight scenarios.