請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84333
標題: | 結合立體影像伺服及強健適應性背向步進控制器應用於三軸並聯式氣壓機械臂之研究 Development of Visual Servo System and Robust Adaptive Backstepping Control for a Three-Axial Pyramidal Pneumatic Parallel Manipulator |
作者: | 董逸棋 Yi-Chi Dong |
指導教授: | 江茂雄 Mao-Hsiung Chiang |
關鍵字: | 並聯式機構機械臂,氣壓伺服控制,強健適應性背向步進控制,立體影像定位,影像辨識,ORB特徵辨識,影像伺服控制,軌跡規劃, parallel manipulator,pneumatic servo control,robust adaptive backstepping control,localization based on stereo vision,image recognition,ORB feature detection,visual servo control,path planning, |
出版年 : | 2022 |
學位: | 碩士 |
摘要: | 本論文研究以雙眼相機利用ORB (Oriented FAST and Rotated BRIEF) 立體影像系統辨識並獲取目標物的位置資訊,發展影像伺服控制器結合軌跡規劃產生三軸氣壓機械臂的參考軌跡,並設計三軸氣壓機械臂的強健適應性背向控制器 (Robust adaptive backstepping controller) 追蹤軌跡,以達成物件抓取、放置、動態追蹤之功能。
在影像辨識部分,以ORB (Oriented FAST and Rotated BRIEF) 特徵偵測法識別在雙眼相機中特徵點,以及計算其描述子 (descriptor),並與資料庫中目標物的描述子進行Brutal-Force比對 (matching),利用Ratio test將相似易混淆的特徵點移除。得到目標物影像上的特徵點在雙眼相機中對應的位置後,利用隨機取樣篩選演算法 (RANdom Sample Consensus, RANSAC) 得到目標影像在雙眼相機影像的透視變換矩陣(Perspective transformation matrix),將目標影像中心點映射到雙眼相機的影像中,最後在三維空間座標中以立體三角定位 (Stereo triangulation),利用視差法計算目標物中心點的三維位置資訊。 因為三軸氣壓機械臂的控制取樣頻率 (1 kHz) 與影像處理頻率 (25 Hz) 不同,影像伺服採用分層控制,第一層為影像伺服控制器,第二層為三軸氣壓機械臂位置控制器。在影像伺服控制器中,設計基於目標物位置的非直接影像伺服控制器 (Dynamic Position-based look-and-move visual servo controller) 產生三軸氣壓臂末端點的參考控制速度。由於三軸氣壓機械臂控制器輸入為末端點參考位置,與影像伺服控制器的輸出(參考速度)不同,因此本論文設計以參考速度為輸入,結合針對機台運動學之限制條件,設計五階貝氏路徑規劃 (5th order Bézier curve path planning) ,輸出末端點的最佳化路徑以間接達成參考速度的追蹤。 三軸氣壓機械臂控制器設計方面,本研究採用三軸獨立控制策略。機械臂末端點的三維運動軌跡,以三軸並聯式機構之逆向運動學計算出氣壓缸個別的運動軌跡,以氣壓缸兩端壓差以及光學尺的位置訊號作為回授,進行位置追蹤控制。為了提高氣壓缸在運動轉向處的控制精準度,本論文提出基於系統模型的強健適應性背向控制器 (Robust adaptive backstepping controller),並以傅立葉函數近似法 (Fourier series based approximation) 為基底估測系統高頻率增益 (high frequency gain),對三軸無桿氣壓缸進行即時軌跡追蹤控制。 最後進行實驗驗證,分為兩部分,第一部分為三軸氣壓機械臂運動控制,包含單軸氣壓缸伺服系統軌跡追蹤實驗,以及順向運動學規畫機械臂末端點三維運動軌跡,進行末端點軌跡追蹤實驗;第二部分為影像伺服整合三軸氣壓機械臂控制實驗,包含實現目標物影像辨識、三維位置定位,以及物件抓取、放置,與物件動態追蹤實驗。 The objective of this study is to identify and localize the target object with binocular camera by ORB stereo vision system, and develop the visual servo controller combined with the path planning to generate the reference trajectory in Cartesian space of the TPPM (three-axial pyramidal pneumatic parallel manipulator). Simultaneously, the robust adaptive backstepping controller is designed to achieve the path tracking control for pick and place and dynamic object tracking. The ORB (Oriented FAST and Rotated BRIEF) feature detection is first used to identify feature points in both cameras and calculate their descriptors, and perform Brutal-Force matching with the descriptors of the target objects in the database. The ratio test algorithm is applied to remove the similar and confusing feature points. Afterwards, the RANdom Sample Consensus (RANSAC) is used to obtain the perspective transformation matrixes from target image to binocular camera images. After mapping the center point of the target image to binocular camera images, the stereo triangulation method is adopted to calculate the target three-dimensional position from disparity of center point in binocular camera images. Because the sampling rate between the TPPM controller (1 kHz) and the feature detection (25 Hz) are different, the control architecture of the visual servo is designed in two layers in this dissertation, i.e. “dynamic look-and-move”. The first layer is the visual servo controller, and the second layer is the TPPM controller. In the design of the visual servo controller, the dynamic position-based look-and-move visual servo controller is designed to generate the reference control velocity for the TPPM end-effector. Since the control input of the TPPM controller is the reference position, which is different from the output (reference velocity) of the visual servo controller, the 5th order Bézier curve path planning with the constraints of reference velocity as well as the kinematic limitations of the TPPM were applied to optimize the end-effector output trajectory of the in Cartesian space for achieving the reference velocity tracking. In this study, the three-axial independent control scheme using pressure difference feedback and the position feedback were proposed, in which the reference position of three cylinders are calculated by the inverse kinematics. In order to improve the cylinder position tracking response, especially at the moment of moving direction change, the FSB-RABSC (Fourier series-based robust adaptive backstepping controller) is proposed as the model-based controller to track the real-time trajectory of the TPPM. The Fourier series-based approximation is used to estimate the high frequency of the system. Finally, the experiments consist of two parts. The motion control of the TPPM was implemented first, including the single-axial pneumatic cylinder and the end-effector of the TPPM. After that, the visual servo integrated with the TPPM control was validated for the realization of the feature detection and 3D localization of the target object and the application of “Pick and Place” and “Object dynamic tracking”. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84333 |
DOI: | 10.6342/NTU202200826 |
全文授權: | 同意授權(限校園內公開) |
電子全文公開日期: | 2027-05-29 |
顯示於系所單位: | 工程科學及海洋工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-110-2.pdf 目前未授權公開取用 | 14.37 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。