Data-driven robotic assembly: Retaining accuracy through process and software techniques

Abstract

A trend in the field of robotics is the creation of links between robot work cells and mainframe-based computer aided design (CAD) systems, simulators and off-line programming tools. In an experiment in the IBM Poughkeepsie development laboratory, harness cable design data is used to calculate path information for a mock-up assembly station. The calculations are performed using a mathematical "model" representing the physical dimensions of the robot. However, the actual robot has slight asymmetries in its joints and beams that are not predicted in the model, resulting in a substantial difference in the calculated and the actual locations of the robot gripper. This paper describes the detection and measurement of the model-vs-robot accuracy problem in the Poughkeepsie experiment and a systematic approach used to resolve it.