Neural Network-based Direct Model Reference Adaptive Control of Quadrotor Attitude

The scope of this research is to develop a model reference adaptive control system of a quadrotor attitude. Unlike many known papers, the plant parametric uncertainties are taken into consideration, and their explicit equation is obtained for the Euler angles control loops. It is a linear regression with nonlinear elements in both the regressor vector and time-varying parameters matrix. Having analyzed such an equation, a neural network is chosen to compensate for the uncertainty, and a mathematically sounded set of its inputs is defined. The adaptive laws to adjust the network parameters are derived, and the asymptotic convergence of the tracking error is shown. A Parrot Mambo quadrotor model is used as a testbed. The proposed system has reduced the sum of squared errors between the setpoint and Euler angles compared to $k^{x}x+k^{r}r$ controller.