Using Reinforcement Learning Tue Visual Servo System Control Gain

• Main Authors: Chuang, Yi-Lin, 莊易霖
• Other Authors: 黃國勝
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/eaf294

碩士 === 國立中正大學 === 電機工程研究所 === 103 === How to set control gain is always an important issue in control system design. The control gain is also important in visual servo system. If control gain is too high, the system will become unstable. In contrast, small control gain results in long converge time. Approaches like adaptive control and cerebellar model articulation controller (CMAC) are proposed for solving this issue. But the adaptive control system is too complex that hard to design, and CMAC needs to be trained over the whole workspace of visual servo system which requires significant computational resources. In this thesis, we propose using Q-learning to tune the control gain of image-based visual servoing system, decrease the time consumption of the robot approaching to desired position.