Robust robot-camera calibration

Calibrating the parameters of a vision system used in robotics is crucial for many tasks where the robot has to physically interact with the environment. This paper introduces a robust method for calibrating the relative poses between the base frame of the robot and one or more cameras. The method is based on tracking a marker attached to the end-effector of the robot and does not require manual bootstrapping. The method is robust to a large number of outliers (incorrectly detected marker positions) and provides an estimate of the uncertainty of the estimated parameters based on the variance of the observed errors, providing information on the accuracy of the estimate. Experimental results demonstrate that highly robust estimates can be obtained with relatively few measurements.