The Neural Network Fuzzy Compensator Design for the Billiard Robot

• Main Authors: Bo-Lu Chen, 陳柏檽
• Other Authors: Jr-Syu Yang
• Format: Others
• Language: zh-TW
• Published: 2003
• Online Access: http://ndltd.ncl.edu.tw/handle/92715784333023513840

碩士 === 淡江大學 === 機械與機電工程學系 === 91 === The objective of this research is to establish the error compensator of the intelligent billiard robot by neural network and fuzzy theory. The billiard robot is trained by neural network for many times in advance to make the robot possess the imitation ability of how people play billiard and improve the“pocketing” ability of it. In addition, the correction rules of compensator errors made by fuzzy theory are applied to approvethe skill and the rate of pocketing. In the learning process of imitation how people play billiard, the pocketing outcome of the robot and related information are recorded, and then a database is established. Moreover, take advantage of the neural network training of back-propagation network (BPN) and make use of its learning and memorizing features to establish the rules of pocketing errors of the billiard robot. After network training, the predicted errors are established rules of compensating errors by fuzzy theory. Hence, the goal of compensating errors could be reached without modifying the hardware of the robot. The angle of shooting balls and pocketing errors are measured by CCD camera and stored in the computer. Those are analyzed, learned, and memorized by the program of neural network, even inferred errors except the data of measuring. Before the command signals are sent to the controller, compensating signals added to command signals are sent to controller in order to eliminate the error.