Nonlinear controller for a UAV using Echo State Network
Bogdan Pugach, Brian Beallo, Dave Bement, Sean McGough, N. E. Miller, Justin Morgan, L C Jorge Rodríguez, Kyle Winterer, Tristan M. Sherman, Subodh Bhandari, Zekeriya Aliyazicioglu · 2017
A nonlinear adaptive controller for an unmanned aerial vehicle (UAV) has been developed using Echo State Network (ESN), which is a form of three-layered recurrent neural network (RNN). Online learning is used to train the ESN in real-time starting from randomized weights. The ESN is integrated into ArduPilot, an open source autopilot, for complex flight simulations. Software-in-the-loop and hardware-in-the-loop simulations are performed using the FlightGear Flight Simulator. The response of the UAV using the controller based on the ESN has surpassed the performance of the traditional controllers. Noise and external disturbances are added to show the effectiveness of the controllers. A UAV test platform is designed and built to gather aircraft flight data and test the ESN.