Optimal Experiment Design for Elasto-Geometrical Calibration of Industrial Robots

Abstract

Inaccuracy of the kinematic model used in robot controllers and deflection of robot joints are two main sources of positioning errors in current industrial robots. We propose an elasto-geometrical calibration method to address these problems. The elasto-geometrical calibration identifies the accurate kinematic model and joint elasticities of any industrial serial robot by measuring the robot tool position at multiple design points. Each design point indicates a unique combination of a robot configuration (set of joint values) and an external load on the robot tool. Proper selection of the design points could significantly improve the calibration accuracy and reduce the experiment time. We propose an optimal design of experiment to find the D-, A-, and E-optimal designs from a large pool of candidate design points. Unlike the existing approaches, we use a semidefinite convex programming that can find a suboptimal design of experiments. The efficiency of the proposed elasto-geometrical calibration is evaluated on an ABB IRB 1600 robot. For this experiment, a cable-driven parallel robot is employed to apply multidirectional external loads on the tool of the ABB robot. Experimental results show that the proposed calibration method significantly improves the robot's accuracy in comparison with a regular kinematic calibration method. In addition, the D-optimal design results in less positioning error than A- and E-optimal designs.