| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 98 === This study aims to develop a four degree-of-freedom (DOF) robotic manipulator
which is constructed to emulate the upper limb structure of human beings. Four joint motors are designed to perform the motions of a 3-DOF shoulder joint and a 1-DOF elbow joint. The 3D spatial motions of the end-effector (i.e., wrist) can be desired in terms of controlling the joint motor angles of the proposed 4-DOF robotic manipulator; hence, such a configuration results in the redundancy problem. In general, Jacobian solutions are linear approximations of inverse kinematics problems with redundancy conditions. Form the viewpoints of upper limbs’ motions of human beings, limb motions may be characterized as different motion scenarios. The same wrist position can be generated from different limb postures, and these postures depend on different motion scenarios such as writing words, waving hands, shaking hands, etc. As a
consequence, Jacobian solutions are difficult to realized specific limb motion
scenarios of human beings. Therefore, this thesis proposes a supervised neural
network based robotic manipulator control system which constructs limb motion
characteristic models according to relative joint posture features with respect to different motion scenarios. The generated motion features are further used to provide an auxiliary condition for eliminating the redundancy problem of the inverse kinematics as well as to meet specific motion scenarios. The proposed control system is implemented based on the “System on a Programmable Chip (SoPC)” techniques, and the proposed system is developed based on hardware-software co-design approaches. By properly allocating hardware and software modules, the system performance can be improved. Finally, several trajectory tracking experiments are done in terms of the Jacobian and neural network approaches, respectively. In order to
verify the system performance, this study employs a motion capture system to record the experiment results. Experiment results successfully demonstrated that the proposed neural network based control system performs similar motion behaviors when compared to Jacobian approaches for the same test trajectory and motion scenario.
|
|---|
