Improving Automated Insertion Applications by Restoring Rigid-Body Condition in Point-Based Registration

Published: March 19, 2018

Author(s)

Geraldine S. Cheok, Marek Franaszek, Karl Van Wyk, Jeremy A. Marvel

Abstract

Accurate knowledge of an object's location is essential in automated robotic applications such as machine tending, pick-and-place operations, and part assembly. Perception systems are used to acquire the object's location, and this information is then passed on to the robot. However, the measured location of the object is in the coordinate frame of the perception system. For the robot to act on the object, the location of the object must be transformed relative to the coordinate frame of the robot. In manufacturing applications that involve industrial robots, a common way to register coordinate frames to each other or to a global frame is by rigid-body, point-based registration using at least three common points or fiducials that are measured in both frames. Researchers at the National Institute of Standards and Technology (NIST) have developed a method to improve this registration method. The new method restores the rigid-body condition (RRBC). This report presents an assessment of the NIST RRBC method by conducting practical robotic peg-in-hole experiments. In these experiments, a robot arm was used to insert a peg into a hole. A simple Pass/Fail metric was used to evaluate the effectiveness of the method. Two other metrics involving registration errors were also used. The results show that the RRBC reduces the number of failed insertions, especially with tight tolerances.
Citation: NIST Interagency/Internal Report (NISTIR) - 8198r1
Report Number:
8198r1
Pub Type: NIST Pubs

Supersedes

Keywords

peg-in-hole experiments, registration, registration error, rigid-body, target registration error
Created March 19, 2018, Updated November 10, 2018