影像伺服及其在機器人搭乘電梯之應用

Abstract

本論文為因應日益成熟的機器人導航技術，和呼應功能導向的機器人設 計，以及突破一般機器人只有在單一樓層的平面運動限制，而研究一個以 模倣人類行為的機器人搭乘電梯之實驗，以期其可以在一棟擁有電梯系統 的現代化大樓各個樓層，執行原來其只能在單一樓層所做的功能，以擴充 其能力和使用的彈性。在本論文中，我們利用影像處理的技術，並引入了 一個即時分類的類神經網路，成功地在實驗中辨識出電梯操作面板上的各 個樓層按鍵。並以此為基礎發展出一套基於視窗追蹤的法則，用以導引機 器臂按下所辨識出的按鍵，經過實驗的驗証，我們提出的法則的確可以成 功地解決了機器人在不同模式下，不同的行為控制問題。 In this thesis, we develop a mobile manipulator system which can mimic a human being to manipulate the elevator. In spite of the recently well developed navigation technology, and task- oriented service robot, a breakthrough to the mobile robot is still required on the fact that it only can work at single floor. It is desired that it can perform all kind of services in each floor of a modem building with the elevators. This will greatly enhance its ability and flexibility. We employ the image processing technology and developed a neural network for real- time clustering. This system can successfully recognize each bottom of the elevator panel. Then we develop a new window-based tracking method for guiding the mobile manipulator to push the elevator bottom. Finally, we develop a fuzzy logic controller with two fuzzy rule bases to solve the robot system in two different control modes with two different behaviors.