Deep Learning Based Neural Network Controller for Quad Copter: Application to Hovering Mode

In the past few years, new advances in Deep Neural Networks (DNN) and Deep Learning (DL) has made it possible to control Rotary Unmanned Aerial Vehicles (RUAVs) with a variety of robust and intelligent techniques. In this work, a feedforward-based deep neural network is utilized to control the altitude, hovering mode, of an RUAV system. An automated search routine was developed to determine the optimum architecture of the neural network for the controller. This network was trained using the supervised learning technique, and the controller performance was compared for three different DL/DNN training paradigms; the standard feedforward method, the greedy layer-wise method, and the Long Short-Term Memory (LSTM) method in which the response of each controller was presented, where it was found that the greedy layer-wise method gives the most optimal result.