本研究是以影像為基礎的視覺伺服系統為架構，並且藉由論文中所閳述的影像轉換矩陣(即物件之影像速度與其空間速度之關係矩陣)為基礎來設計此控制器，使用的理論基礎為最小方根誤差運用遞迴運算求出每個取樣時間內系統的影像轉換矩陣，藉由影像轉換矩陣的收斂使得系統穩定且估算正確的控制命令。 接下來我們將提出一里阿普諾函式(Lyapunov function)證明此視覺伺服系統的穩定性，並且討論系統包含不確定因素(uncertainties)時的穩定性極限，經由更完整的攝影機模型推導出更接近於真實的影像轉換矩陣，由模擬的結果得知系統穩定性極限的數據並且比較在收斂性實驗的經驗數據，來提供架設此適應性控制系統的依據。 本文所提的視覺伺服系統主要的優點為避開繁雜的攝影機校正與推導機械臂的動態方程式，並且由即時的影像的回授與控制法則計算出此時的機械臂控制命令(Robot Command)正確追蹤到目標物。 In recent years, the application of robot is extensive. In order to recognize the relative position between robot and working space, CCD is applied to get the image information to orientate . The focus of our work is to achieve real-time tracking object using visual sensory feedback under non-calibrated environment. The image Jacobian, relating robot velocity commands to image feature’s velocity, is derived based on the scheme of three cameras and shown to be a function of camera focal lengths, fiducial’s depths, fiducial’s positions relative to robot reference frame and pose between camera reference frames and object reference frames. We proposed an on-line adaptive control algorithm to recursively estimate the image Jacobian for robot control. If the image jacobian matrix is convergence, the visual system will be stable. Then we will prove the stability of this visual system by Lyapunov function and discuss robust stability bounds of this linear discrete-time system with time-varying uncertainties. From the result of simulation and experimentation, we can get experience dates in mounting cameras. The main advantages of our system is avoiding tedious camera calibration, pose estimation, and ignoring the robot dynamics. Experiment results are presented to demonstrate the effectiveness of the proposed algorithm. Experiments on MITSUBISHI RV-1 robot are presents to demonstrate the effectiveness of the image-based visual servo with our algorithm.
