Skip to Main Content
Hand-eye calibration is an important component of robotic systems that perform manipulation and grasping tasks. However, calibration is often an onerous process - there are many parameters that must be estimated for the sensors and manipulators, resulting in a high-dimensional nonlinear estimation problem. While it is easy to obtain an approximately correct hand-eye calibration, reducing the error further requires increasingly greater effort. We have developed a simple method for increasing the accuracy of an approximately correct hand-eye calibration. This method does not require any external instrumentation and is unique in that it applies a transformation to sensed object locations to produce commanded end-effector locations. This method has been applied to the robot for the DARPA ARM-S program, consisting of a 7 DOF arm and a sensor head mounted atop a 4 DOF neck. We describe the theory of our approach, our implementation, and experimental results.