Real-time validation of a dual neural network controller for a low-cost UAV

Abstract

This paper describes a novel indirect adaptive control technique based on neural networks for Unmanned Aerial Vehicles (UAV). Two neural networks are amalgamated to provide faster tracking of the commanded reference. A pre-trained internal model network and an online trained controller network form the core of the dual neural network (DNN) architecture. A Telemaster UAV equipped with various sensors and an onboard data-logger and control (ODC) unit forms the experimental platform. The plant model and the controller are designed in numerical simulations based on data obtained from experimental flights of the UAV. The DNN controller is validated on a real-time hardware in loop (HIL) simulation. Results from the numerical simulations and HIL validations are provided.