Trajectory Tracking Control of Quadrotor UAV based on Parallel Cooperative PID-MPC Architecture

Abstract

Quadrotor UAV trajectory tracking control, PIDMPC parallel cooperative architecture, feedforward-feedback fusion mechanism, event-triggered strategy, fuzzy adaptive parameter tuning, MATLAB/Simulink simulation validation. To address the problem that the advantages of Model Predictive Control (MPC) cannot directly assist the Proportional-IntegralDerivative (PID) controller in the traditional cascade PID-MPC control structure, this paper proposes a PID-MPC parallel cooperative control architecture. This architecture employs MPC as the feedforward channel for trajectory prediction and lookahead optimization, and PID as the feedback channel for rapid error correction, synthesizing the outputs of both into the final control input through a dynamic weighted fusion mechanism. To reduce the online computational burden of MPC, an eventtriggered mechanism based on the state error norm is introduced, which activates the MPC solver only when the error exceeds a threshold. Meanwhile, a fuzzy inference system is designed to achieve online adaptive tuning of PID parameters. On the MATLAB/Simulink platform, using a quadrotor UAV nonlinear dynamics model as the plant, the proposed method is compared with three baseline controllers-pure PID, pure MPC, and traditional cascade PID-MPC-under four scenarios: figure-8 complex trajectory tracking, pulse and continuous wind disturbance rejection, model parameter mismatch robustness, and computational efficiency. Experimental results demonstrate that the proposed parallel cooperative architecture achieves a root mean square tracking error (RMSE) of 0.047m, representing reductions of 78.1% compared to pure PID, 52.5% compared to pure MPC, and 41.3% compared to the traditional cascade structure; the recovery time under pulse wind disturbance is only 0.826s, and the steady-state error under continuous wind disturbance is 0.031m; the MPC trigger rate is reduced to 38.7%, with an average single-step solving time of 2.134ms, satisfying real-time requirements.