In this paper, we introduce the bin-picking robot system we developed for the manufacturing industry. The system, that is composed of RGB-D cameras and a dual-arm robot, should locate the target object inside a bin, pick it, and place it onto the assembly jig. As target objects, three types of pipes and two types of flat plates (all of them made of metal) were used. To estimate the position and pose of an object inside the bin, the depth image captured by the RGB-D camera is used. And we describe three evaluation methods; accuracy, reproducibility and availability, those applied for an object's position and pose estimation algorithm, too. The accuracy and the reproducibility of the estimation are evaluated using an XYθ stage. The object is transferred in X and Y directions and rotated about Z axis using the stage, and the estimated transfer matrix of each location is compared to the real one. The availability for the estimation algorithm is measured by simulating the manipulation task with human operation. The results of the estimation agree with the manipulation experiments performed on the robot.
抄録全体を表示