GRASPING AND REPOSITIONING OBJECTS USING INVERSE KINEMATIC METHOD FOR ARM ROBOT BASED ON PIXEL POSITION REGRESSION

Abstract

This paper presented a method of recognizing objects' positions using a vision sensor to identify the grasping point of the arm robot. The arm robot used an embedded color vision Pixy CMUcam5 camera (Pixy camera) to capture and process the colored object image. Pixy camera is low in cost and easily programmed with high-speed FPS in real-time data processing. However, it still has the fisheye effect problem when recognizing the object's position. Thus, the pixel position regression algorithm was chosen to transform the colored objects' positions in the real coordinates. This method was implemented in the arm robot for grasping, picking, and repositioning tasks. The transformation results were employed to calculate the kinematics of the arm robot's joints. The experimental results showed that the Pixy camera could identify objects in the real world and transform the objects' positions. The error value at the real-world position was no more than 1.67 cm, and the experiment showed that the success rate of doing the task was up to 80%.