Data-Driven MPC for Quadrotors

Aerodynamic forces render accurate high-speed trajectory tracking with quadrotors extremely challenging. These complex aerodynamic effects become a significant disturbance at high speeds, introducing large positional errors, and are difficult to model. To fly feedback control must be able account for these in real-time. This necessitates modeling procedure that is both efficient evaluate. Therefore, we present an approach model using Gaussian Processes, which incorporate into Model Predictive Controller achieve precise real-time control, leading up 70% reduction error speeds. We verify our method by extensive comparison state-of-the-art linear drag synthetic real-world experiments speeds of 14 m/s accelerations beyond 4 g.