Autonomous Navigation of an Aircraft Using a NARX Recurrent Neural Network

This paper explores autonomous aircraft navigation using machine learning approach focusing on the ground run and takeoff which is one of the most critical control problems in the aircraft navigation. A controller which controls the aircraft during the takeoff is designed as a black-box method focusing on the input-output relationship between the flight data and control commands. The controller is modelled using the time-series relationship among the data realizing a recurrent neural network (RNN) with the nonlinear autoregressive network with exogenous inputs (NARX) architecture. The flight data are acquired from pilot experiences in the X-Plane flight simulator. Furthermore, the modelling of the controller is constituted using these experiences. This paper also discusses the takeoff performance of the controller network in clean weather conditions that includes additional wind layers at various altitudes. The simulation results establish a satisfactory flight performance of controlling the aircraft in a smooth and stable manner.