Data-Driven Attitude Modeling and Control of Quadrotor UAVs using Koopman Operator

This paper proposes a data-driven approach to attitude modeling and control of the quadrotor UAV using Koopman operator theory. First, to address the limitations of traditional modeling methods, the Koopman operator is employed to learn the nonlinearity of the attitude dynamics from real data (i.e., angles, angular velocities and control torques). Then, the original nonlinear attitude system is reformulated in a high-dimensional linear attitude model. Second, a linear secondary regulator (LQR) control scheme is designed to ensures the stability of the attitude system. Finally, a physical platform of quadrotor attitude control is set up, experimental results validate the accurate prediction of the proposed modeling and control approach. A comparison with PID confirms the superiority of the proposed approach.