Finding the Optimal Pose of 2D LLT Sensors to Improve Object Pose Estimation.

Dominik Heczko,Petr Oščádal,Jan Bém,Ivan Virgala,Tomáš Kot,Václav Krys,Zdenko Bobovský,Adam Boleslavský

doi:10.3390/s22041536

Abstract

In this paper, we examine a method for improving pose estimation by correctly positioning the sensors relative to the scanned object. Three objects made of different materials and using different manufacturing technologies were selected for the experiment. To collect input data for orientation estimation, a simulation environment was created where each object was scanned at different poses. A simulation model of the laser line triangulation sensor was created for scanning, and the optical surface properties of the scanned objects were set to simulate real scanning conditions. The simulation was verified on a real system using the UR10e robot to rotate and move the object. The presented results show that the simulation matches the real measurements and that the appropriate placement of the sensors has improved the orientation estimation.

Highlights

Improve Object Pose Estimation.In industrial manufacturing, accuracy and precision requirements in assembly and manipulation of objects keep increasing
We focused on the appropriate placement of line triangulation (LLT) sensors with respect to a scanned object to provide relevant input data for pose estimation using RANdom SAmple Consensus (RANSAC)
Diffuse light reflection is important for LLT sensor scanning, which is the scattering of reflected light into the surroundings and into the charge-coupled device (CCD), which detects the laser line on the scanned surface

Summary

Introduction

Accuracy and precision requirements in assembly and manipulation of objects keep increasing. For a precise assembly performed by an industrial robot, it is important to define the object’s picking point and orientation so that the robot can accurately grasp the object and perform the assembly. This is done using process pallets, jigs, and other equipment. There are systems that perform assembly based on automatic guidance for large objects [1], where the robot end-effector must have a sensory subsystem to ensure sufficient precision. In the study [3], the authors achieve significant assembly accuracy with an industrial robot to hundredths of a degree by on-line movement compensation

Methods

Results

Conclusion