A Method for Inverse Robot Calibration

Abstract

A method is described for the inverse calibration of a manipulator or robot. Inverse calibration is defined to be finding the joint angles necessary to drive a robot to a desired endpoint location. The joint angles recommended by the robot controller’s internal model will not, in general, drive the robot to the desired location because of inaccuracies in this model. Inverse calibration seeks to reduce the error. Unlike previous work in calibration, the method reported here does not require modeling any specific phenomena that may cause the error; hence it is not limited in accuracy by inability to identify all the error sources. The method consists of finding approximation functions by which corrections are made to the encoder readings recommended by the robot’s internal model. These functions are found by measuring the error at discrete locations throughout a region of the robot’s workspace and then least-squares fitting third order trivariate polynomials to the error samples. A forward calibration (one which reports actual tool location from given encoder readings) based on the above method is also described. The inverse calibration is tested on a six DOF PUMA simulation. Results show that the endpoint location error can be reduced from an average of about 1.2 mm down to an average of about 0.12 mm.