Implementation of a Neural Network for Nonlinearities Estimation in a Tail-Sitter Aircraft

A tail-sitter aircraft’s control is a challenging task, especially during transition maneuvers where the lift and drag forces are highly non-linear. In this work, we implement neural network (NN) capable to estimate such nonlinearities. Once they estimated, one can propose scheme these correctly feed-forwarded. Our NN implementation has been programmed in C++ within PX4 Autopilot, an open-source autopilot for drones. To ensure that does not considerably affect autopilot’s performance, coded must imply light computational load. test our approach, have carried out series of realistic simulations Software The Loop (SITL) using Autopilot firmware. These experiments demonstrate implemented be used aerodynamic improve algorithms all flight phases aircraft: hover, cruise flight, transition.