| Visual servoing is an important research area that has important theoreticalsignificance and broad application prospects. In recent years, with thedevelopment of computers, robotics, image acquisition and processing, controltheory and technology, robot visual servo has been widely applied to theindustrial, aerospace, autopilot and other fields, and it has become a branch ofintelligent robots.This thesis focuses on uncalibrated image based visual servoing system. Bycombining the advantages of the support vector regression (SVR) to withcontrollers, some effective control strategies have been proposed. The mainresearch work is summarized as follows:(1) Based on open source Robot toolbox and Machine toolbox, graphicalinterface simulation software with a highly visual, easy to operate featuresis developed. Considering the practical needs, an interface function isdesigned flexibly based on a framework of object-oriented programming.Therefore it can be directly transformed into the PC control software, whichbuilds up a good basis for subsequent visual servo study.(2) Based on analysis of visual mapping model, a support vector regressionbased PI controller is designed. Depending on its excellent learninggeneralization ability, an SVR is used to learn image Jacobian matrix offline;a flat and a three-dimensional space motion tracking are achieved via usingreal-time motion planning.(3) Considering the problems of difficulty in designing the membershipfunction and extracting fuzzy rules of a fuzzy controller design, a fuzzycontroller is designed via using SVR to extract fuzzy rules. Firstly, the fuzzybasis functions of a fuzzy controller are taken as the kernel function of anSVR, and an equivalence of the fuzzy controller and the SVR is constructed. Then, support vectors are learned by the SVR. Finally, the support vectorsare used to establish the fuzzy rules of the fuzzy controller. This method hasbeen successfully applied to a robot control. The simulation experimentsshow its effectiveness. |
PDF Full Text Request