NEURAL NETWORK CONTROLLER DESIGN FOR QUADROTOR IN REAL TIME

Abstract

In this study, a neural network assisted real-time controller has been designed for highly nonlinear quadrotor orientation control. As controlled dynamic system, a highly nonlinear quadrotor model were used in this study. The controller has been designed with neural networks for Quadrotor attitude and trajectory control, and this controller’s performance was compared against the classical PID controller’s performance. These controllers for dynamic systems performance evaluations were made with the help of real-time experimental setup. Attitude and trajectory controls have done for designed controllers performance evaluation. Neural network performance was compared against the classical PID controller which was designed in the same experimental setup. Designed controllers robustness was tested with pulse inputs. We observed that neural network assisted controller was more succesful than PID controller in more complicated nonlinear dynamic systems, both in attitude and trajectory control performance. We comment that with today’s microcontroller technology, neural networks are faster and more simple for designing controllers for highly nonlinear dynamic systems.

Keywords

• Artificial Neural Networks
• Quadrotor
• Real Time Hardware in the Loop Simulation