Dynamical Model Learning and Inversion for Aggressive Quadrotor Flight

Abstract

Quadrotor applications have seen a surge recently and many tasks require precise and accurate controls. Flying fast is critical in many applications and the limited onboard power source makes completing tasks quickly even more important. Staying on a desired course while traveling at high speeds and high accelerations is difficult due to complex and stochastic aerodynamic effects, poorly modeled dynamics, and unreliable state estimation. This thesis seeks to design control strategies that enable quadrotors to track aggressive trajectories precisely and accurately in the presence of external disturbances, unmodeled dynamics, and imperfect state estimation. We first introduce a model learning strategy that allows efficient compensation of learned acceleration disturbances using the differential flatness paradigm. Then, we extend our learning approach to the feedback linearization controller and show that feedback linearization is a viable strategy for aggressive quadrotor flight. We also show that learning attitude dynamics models improves attitude control loop performance, which in turn improves position trajectory tracking performance. Finally, we validate our model learning approach in extensive outdoor experiments at high speed, under realistic disturbance conditions, and with imperfect state estimation.