結合點雲影像辨識之機器手臂夾取位置與姿態之設計

Abstract

隨著自動化工廠越來越普及，人們開始追求更「聰明」的自動化系統，因此本研究建立了一套能利用點雲資訊規劃出物件之夾取位置及姿態的方法，可結合二指平行夾爪用於工業之自動夾取系統。首先，將物件之特徵向量與已儲存於資料庫中模型之特徵向量比對獲得物件的位置及姿態後，根據姿態從方向性邊界盒(Oriented Bounding Box，OBB)坐標系中選擇一個不會碰撞到桌面的軸向作為夾持軸，並將夾爪移動範圍內的點雲投影至和夾持軸垂直的指定平面上，使3D點雲資料簡化為2D來縮短運算時間。接下來，在附著於物件上的OBB坐標系中找出可行夾持位置及姿態後再根據評分及摩擦角大小選出相對於物件的最佳夾持位置及姿態，最後透過轉移矩陣將其轉換至機械手臂坐標系中。透過此方法能快速地計算出相對於物件的最佳夾持位置及姿態，並根據物件的擺放姿態調整夾持的方向使夾持結果更穩固。

With the increasing popularity of automated factories, people began to pursue the 'smart' automation systems. Therefore, this study established a method that uses the point cloud information to plan the gripping position and posture of the object for a robot. The method utilizes a two-fingered gripper and could be used in an automated system in industry. First, after the feature vector of the object is compared with the feature vector of the model stored in the database, the position and posture of the object are obtained. According to the posture, one of the axes in Oriented Bounding Box (OBB) coordinate system which could avoid gripper from crashing to the table is selected as the gripping axis. The point cloud within the moving range of the gripper is projected onto a specified plane perpendicular to the gripping axis, such that the attained 2D data can be utilized. Next, after finding the feasible gripping position and posture in OBB coordinate system attached to the object, the optimal gripping position and posture relative to the object are searched based on the gripping indecies. Finally, the gripping position and posture are converted to the global coordinate system such that the robot gripper can be moved to grip the object.