REAL-TIME CONTROL AND IDENTIFICATION OF DIRECT-DRIVE MANIPULATORS (ROBOTICS)

This dissertation addresses the area of model-based control of direct-drive manipulators. The manipulator control problem revolves around the computation of the actuating torques that will cause the manipulator to follow the desired trajectory. The model-based schemes accomplish this objective by incorporating a complete dynamics model in the computation of the actuating torques. The main hindrances in the real-time implementation and evaluation of the model-based control schemes have been the computational requirements of the inverse dynamics and the assumption that the model is accurately known. In this dissertation, we present our research that removes these hindrances and paves the way for evaluating the real-time performance of the model-based control schemes.To ease the difficulties associated with deriving symbolically the identification equations, we also propose a numerical version of the identification algorithm. We have experimentally implemented this algorithm to estimate the dynamics parameters of the six degrees-of-freedom CMU DD Arm II.Further, to increase the robustness of the identification algorithms we have also developed a procedure to categorize the dynamics parameters of a manipulator. The categorization procedure is based on the reformulation of the Newton-Euler algorithm and the customization procedure.Finally, we have implemented the model-based control schemes at a sampling rate of 2 ms. and evaluated their performance. Specifically, we compare the computer-torque, feedforward compensation and the independent joint control schemes. Our experiments have conclusively established the need for including the velocity dependent nonlinear Coriolis and the centrifugal terms in the dynamics model even at low speeds of operation of the manipulator. (Abstract shortened with permission of author.)To obtain an accurate dynamics model, we have developed symbolic and numerical identification algorithms. To synthesize these identification algorithms, we outline the fundamental properties of the Newton-Euler and the Lagrange-Euler dynamics formulations. The nonlinear (in dynamics parameters) Newton-Euler model is transformed into an equivalent linear (in dynamics parameters) Newton-Euler model through a nonlinear transformation. The notion of a torque/force error model is then introduced and cast into series and parallel identifier structures for on-line and off-line dynamics parameter estimation. Our approach is illustrated by identifying the dynamics parameters of the cylindrical robot and the first three degrees-of-freedom of the CMU Direct-Drive Arm II.