Description:
AbstractThe combination of artificial neural networks with advanced control techniques has shown great potential to reject uncertainties and disturbances that affect the quadrotor during trajectory tracking. However, it is still a complex and little‐explored challenge. In this sense, this work proposes the development of robust and intelligent architectures for position control of quadrotors, improving flight performance during trajectory tracking. The proposed architectures combine a robust linear quadratic regulator (RLQR) with deep neural networks (DNNs). In addition, a comparative study is performed to evaluate the performance of the proposed architectures using three other widely used controllers: linear quadratic regulator (LQR), proportional‐integral‐derivative (PID), and feedback linearization (FL). The architectures were developed using the robot operating system (ROS), and the experiments were performed with a commercial quadrotor, the ParrotTM Bebop 2.0. Flights were performed by applying wind gusts to the aircraft's body, and the experimental results showed that using neural networks combined with controllers, robust or not, improves quadrotors' flight performance.