Date of Award
Doctor of Philosophy in Electrical Engineering
Electrical Engineering and Computer Science
Electrical Engineering, Robotics
This dissertation focuses on the control of rigid robots that cannot easily be modeled due to complexity and large uncertainties. The function approximation technique (FAT), which represents uncertainties as finite linear combinations of orthonormal basis functions, provides an alternate form of robot control - in situations where the dynamic equation cannot easily be modeled - with no dependency on the use of model information or training data. This dissertation has four aims - using the FAT - to improve controller efficiency and robustness in scenarios where reliable mathematical models cannot easily be derived or are otherwise unavailable. The first aim is to analyze the uncertain combination of a test robot and prosthesis in a scenario where the test robot and prosthesis are adequately controlled by different controllers - this is tied to efficiency. We develop a hybrid FAT controller, theoretically prove stability, and verify its performance using computer simulations. We show that systematically combining controllers can improve controller analysis and yield desired performance. In the second aim addressed in this dissertation, we investigate the simplification of the adaptive FAT controller complexity for ease of implementation - this is tied to efficiency. We achieve this by applying the passivity property and prove controller stability. We conduct computer simulations on a rigid robot under good and poor initial conditions to demonstrate the effectiveness of the controller. For an n degrees of freedom (DOFs) robot, we see a reduction of controller tuning parameters by 2n. The third aim addressed in this dissertation is the extension of the adaptive FAT controller to the robust control framework - this is tied to robustness. We invent a novel robust controller based on the FAT that uses continuous switching laws and eliminates the dependency on update laws. The controller, when compared against three state-of-the-art controllers via computer simulations and experimental tests on a rigid robot, shows good performance and robustness to fast time-varying uncertainties and random parameter perturbations. This introduces the first purely robust FAT-based controller. The fourth and final aim addressed in this dissertation is the development of a more compact form of the robust FAT controller developed in aim~3 - this is tied to efficiency and robustness. We investigate the simplification of the control structure and its applicability to a broader class of systems that can be modeled via the state-space approach. Computer simulations and experimental tests on a rigid robot demonstrate good controller performance and robustness to fast time-varying uncertainties and random parameter perturbations when compared to the robust FAT controller developed in aim 3. For an n-DOF robot, we see a reduction in the number of switching laws from 3 to 1.
Ebeigbe, Donald, "Control Of Rigid Robots With Large Uncertainties Using The Function Approximation Technique" (2019). ETD Archive. 1344.