Adaptive controller for unmanned aircraft

Abstract

Techniques for an unmanned ariel vehicle (UAV) having a closed-loop controller that is gain scheduled based on a trained machine learning model. The closed-loop controller could be a PID controller. Real-time data pertaining to the motor being controlled is input to the trained machine learning model. Examples of the real-time data includes, but is not limited to, motor current, motor voltage, and an estimate of motor thrust. The trained machine learning model may also input an error term of the closed-loop controller. Tuning constants for the closed-loop controller are derived based on the prediction from the machine learning model. Gain scheduling for the closed-loop controller may thus be performed 'online' while the UAV continues on its mission. Controller gain scheduling may be performed to account for changes in a payload carried by the UAV.