Online Data-Enabled Predictive Control for Quadrotor Trajectory Tracking

Data-enabled predictive control (DeePC) has been extensively studied for its ability to achieve safe control of multiconstrained unknown systems without requiring an explicit system model. Traditional DeePC constructs a Hankel matrix using historical input-output data of an unknown system to replace the system model, enabling receding horizon predictive control. This paper proposes an online data-updated DeePC framework, which combines real-time data with historical data to construct a Mosaic Hankel Matrix online, addressing the issue of data unreliability caused by external system state variations or improper data collection. Furthermore, an adaptive prediction horizon strategy is designed subject to control frequency constraints, and the penalty formulation of slack variables is improved using a sigmoid function, achieving a balance between control efficiency and optimization performance. Finally, quadrotor trajectory tracking experiments were conducted on a ROS+PX4-based software-in-the-loop (SITL) simulation platform to validate the proposed approach.