Grasped Object Orientation Detection for Robust Peg-In-Hole Tasks

This paper approaches the problem of detecting the orientation of an object while the object is grasped by a robot. Manipulator robots are learning new and exciting ways to interact with their environment every day. Object recognition and grasping algorithms allow a robot to identify, target, and grasped objects of importance. However, a common problem in grasping is maintaining knowledge of the orientation of the object while it is being grasped or manipulated. We have developed an algorithm that estimates the orientation of painted plastic test tubes without assistance from external markers. By initially using line fitting to localize the center axis of the tube to assist in the tube’s orientation detection, we then compare the orientation of the tube to the gripper’s approach to the test tube array. Based on the angle difference between the +Z-axis of the gripper and the center axis of the tube, the movements necessary for the robot to reposition its arm to properly approach the test tube array can be calculated leading to success in a Peg-In-Hole task.