Quadrotor altitude and attitude control with neural network

Abstract

Over the years, the controller design for an unmanned aerial vehicle has been attracting considerable amount of attentions and interests. In this thesis, a quadcopter has been studied intensively, and an adaptive backstepping nonlinear controller, along with altitude control using sonar sensor, has been proposed to ensure the stable flight of a quadcopter. A quadcopter consists of four motors acting as its control means. The proper thrust and air drags produced by the propellers completes the tasks to stabilize the quadcopter in the pitch, roll and yaw directions. An attitude estimation technique, the Mahony filter, was firstly implemented to yield the accurate Euler angles for the quadrotor. To understand such a complex system, the Euler-Lagrangian equations has been introduced to develop a dynamic model. Following such a mathematical model, using backstepping design technique, an adaptive nonlinear controller is designed for and implemented onto a real quadcopter. The derived mathematical model has its own uncertainties, such as changes in mass and the location of the center of the mass. Due to such a nature, an adaptive controller based on the neural network has been proposed to estimate certain nonlinear terms in the mathematical model. Successfully design and implementation of such a controller can improve the accuracy of the dynamic model. The unit quaternion representations was also used to eliminate the Gimbal Lock of the Eular angles. A different mathematical model based on Newton equations was also used to further consider the air drag effect of a quadrotor while flying in the air. Proceeding to the attitude controller design, an altitude controller is also designed to lock the flying height of the quadcopter. The design concepts are based on a traditional nonlinear backstepping method. Matlab simulation results suggest a possible implementation of the proposed controller for the quadcopter's altitude and attitude control. A couple of real time experiments have been conducted. The results was recorded and analysed.