Online Dynamics Estimator for Adaptive UAV Control with Uncertain Model

Small, low-cost unmanned aerial vehicles (UAVs) are more prone to actuator failure and manufacturing variability than their larger, more expensive counterparts. To build a controller that was robust to this variability and these failure modes a linear quadratic regulator (LQR) was implemented to make the aircraft follow a desired trajectory, and then at regular intervals, a discrete time continuous state and action model was fit to the state transition and input histories using least squares methods. The method was successfully implemented on several simple systems as a proof of concept, but initial attempts to apply the algorithm to a more complex aircraft model failed to explore the space thoroughly enough to generate a faithful approximation of the system dynamics.