Implementation Details of an Adaptive Flight Controller
Aaron D. Kahn · AIAA Guidance, Navigation, and Control Conference and Exhibit · 2005
The U.S. Naval Research Laboratory has implemented a novel neural-network adaptive flight controller architecture for helicopter-based unmanned aerial vehicles (UAVs) originally presented by Eric Johnson and Suresh Kannan of the Georgia Institute of Technology. Johnson and Kannan utilized on-line adaptation of a neural network to cancel model errors present in an approximate dynamic inversion of the plant. Due to the advantages of risk reduction during controller development and the increased operational flight envelope resulting from the adaptation of the neural-network, this controller was utilized on a number of dierent rotorcraft at the U.S. Naval Research Laboratory. This paper will describe several interesting implementation details of the controller, as well as modifications made to the basic controller architecture during implementation on dierent vehicles. Eects of scale, rotor system design, and sample time will be discussed. Simulation and flight test data will be presented showing the results of these modifications.