On robust adaptive control of robots and unmanned aerial vehicles

Abstract

The control of uncertain robotic systems is often of a nonlinear and high order nature, requiring the use of robust adaptive controllers. The design of these controllers requires knowledge on the system order, structure, and parameters to guarantee convergence of system trajectories to the desired values. This work aims to provide new and more efficient ways to model and control nonlinear systems, paving the way to the development of a neural network controller for the output tracking of a generic nonlinear system. More specifically, the dynamics of various robotic systems were first mathematically modelled, upon which investigations were made in identifying the system parameters, the control of attitude, and the synthesis of neural adaptive controllers with enhanced stability against unmodelled dynamics. The development of a Joint Unscented Kalman Filter was made to simultaneously identify the inertial and aerodynamic constants of a quadcopter in a bifilar pendulum setup. A novel body-frame approach was presented for the attitude tracking of aerospacecrafts. A new model reference adaptive controller was developed to provide enhanced robustness against unmodelled dynamics with an intermediate reference plant. Promising results on a new neural network feedforward controller for the output tracking of a generic nonlinear plant without plant knowledge were also documented. All new findings are encapsulated in theorems with proofs of convergence where appropriate. Simulation results were obtained to substantiate the results, with the help of experimental studies on the attitude control of multiple made-in-laboratory quadcopter aircraft.