Skip to main content
Open Access Publications from the University of California

UC Berkeley

UC Berkeley Electronic Theses and Dissertations bannerUC Berkeley

Data-Driven Approaches for Sensing and Control of Robot Manipulators


In a sensing rich system, a large amount of data can be obtained over time and utilized to improve the performance and functionality of a robotic system. Data-driven approaches emphasize on the utilization of auxiliary sensors, sensor fusion, and data learning. Real-time control systems of robotic systems often run at kilo-Hertz sampling frequencies. New data is obtained from a variety of feedback sources every one or a few milliseconds. Auxiliary sensors provide additional feedbacks and enable sensor fusion. This dissertation presents a series of data-driven approaches to improve the sensing and control of robot manipulators from several aspects, including sensor fusion for motion sensing, statistical learning for feedback compensation, nonparametric learning control, and intelligent modeling and identification.

In regard to the limited sensing capability of conventional indirect drive-trains of industrial robots, a sensor fusion approach based on auxiliary optical and inertial sensors is introduced for direct motion sensing of robot end-effectors. The approach is especially useful to applications where high accuracy is required for end-effector performance in real-time. Meanwhile, for the scenarios where auxiliary sensor are not allowed, a statistical learning algorithms is developed for sensing compensation so that control of systems with limited feedback capability can be significantly improved. A major application of the approach is vision guidance of industrial robots. The proposed learning approach can significantly increase the visual tracking bandwidth without requiring high-speed cameras.

Besides improving the sensing capability of robots, nonparametric learning control is developed to control systems with complex dynamics. A major motivation of the approach is robotic laser and plasma cutting. Furthermore, to obtain high-fidelity models more efficiently, planning and learning algorithms are discussed for intelligent system modeling and identification. The applications of the proposed approaches range from vision guided robotic material handling to precision robotic machining. Various tests are designed to validate the proposed approaches.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View