Data-Driven MPC for Quadrotor UAVs Using SINDy

This paper presents a Data-Driven approach to modeling and control of quadrotor dynamics combining sparse identification of nonlinear dynamics (SINDy) with Nonlinear model predictive control (NMPC) to identify an accurate model for trajectory tracking. Accurate models are essential for NMPC to effectively control complex nonlinear systems, especially over long prediction horizons. Traditional modeling based on first principles often suffers from uncertainties and human errors, leading to insufficient accuracy for precise trajectory tracking in quadrotor UAVs. The SINDy algorithm provides the solution to this problem further expanding the use of NMPC by learning accurate parsimonious nonlinear models of complex dynamics with lots of model uncertainties. The SINDy algorithm reveals the most significant terms in the equations of motion and provides a reduced-order model suitable for real-time control. The identified model is then utilized in an NMPC framework to optimize the quadrotor’s trajectory while respecting state and input constraints. Simulation results demonstrate the effectiveness of the proposed SINDy-MPC approach in accurately modeling the quadrotor dynamics even in the presence of uncertainties and noisy data, achieving precise trajectory tracking on long prediction horizons.