Automated workpiece localization for robotic welding

Abstract

Precise knowledge of a workpiece's position is essential to robotic manufacturing. This often requires design and use of special purpose fixtures and programming between manufacturing operations. Our goal is to remove these requirements and automate the discovery of position (localization) of the workpiece. For quick setup and localization we have augmented the end-effector of the robot with a laser projector and a laser displacement sensor. The laser projector guides the worker in initial fixture and workpiece placement and the laser displacement sensor acquires a 3D point cloud of the workspace. The point cloud collected after scanning the workspace is processed to provide a sparse outline of the workpiece. This outline is then compared to the computer aided design (CAD) data of the workpiece to estimate a transformation between the actual and planned position. This estimation is made by using an iterative closest point algorithm. Multiple searches are run using different seed points to improve the chance of finding the best fit while maintaining low run times. In this paper, we have used flat weldments as test cases. In our experiments we were able to localize and weld workpieces with significant time savings against current practices in manual welding.