由於機器人的穩定性高及可遠端遙控，所以機器人在特殊手術上如微創手術可以協助醫生完成人手難以達成的手術。一般而言，機器人的定位精度不能達到手術的精度要求，必須校正其定位精度。本研究提出正交神經網路補償方法以校正機器人的定位精度，並將機器人應用於影像導引系統中，以驗證其校正結果。首先，以光學定位裝置自動地量測機器人在特定位置(關節角)的方位誤差，並計算出對應的補償量，再利用正交神經網路訓練出機器人的關節角和方位補償量之間的函數關係，並以正交神經網路的輸出補償機器人的定位誤差。本研究以CRS F3機器人進行校正實驗，在校正前機器人的最大位置誤差(誤差機率分佈中2個標準差內的誤差)為20.8mm，經校正後，其最大位置誤差縮小至1.7mm，方向最大誤差也從校正前的3.9度改善為校正後的0.71度，大幅地提升機器人的定位精度。在影像導引機器人定位上，以顱骨模型進行導引實驗，由實驗結果顯示平均位置誤差為2.14mm，平均方向誤差為0.91度。 Due to the reliable stability and tele-manipulability, robot is suitable to assist surgeons to do specified operations such as Minimally Invasive Surgery. Usually, the positioning error of robot is bigger than the required positioning accuracy of operation. Therefore, robot calibration is necessary to improve the positioning accuracy. This study presents an ONN compensation method for the calibration of a six-axis robot, and its performance is verified by applying to image-guided robotic positioning. First, an optical localizer is applied to measure the positioning errors and to determine corresponding position/orientation compensations of the robot at predefined positions (joint angles). The two sets of the joint angles and their position/orientation compensations are brought to Orthogonal Neural Network to establish the mapping relations. Then, the output position/orientation compensations for a given arbitrary set of joint angles are used to compensate positioning error. The CRS F3 robot is used in calibration experiment. The calibration result shows that the average positioning accuracy has been improved from initial 20.8mm to compensated 1.7mm in distance and from initial 3.9 degree to compensated 0.71 degree in orientation. Further, the positioning accuracy of the calibrated robot has been verified in image-guided robotic positioning. The experimental result shows that the average position error is 2.14 mm and orientation error is 0.91 degree.
