Stereo vision based autonomous robot calibration

Robot calibration has been demonstrated to be a useful method to decrease the absolute positioning errors of a robot. Compared to the traditional calibration methods which require expensive external measurement devices, this paper proposes a stereo vision based self-calibration procedure which only needs a stereo camera mounted to a fixed location and a planar marker attached to the robot end-effector. The procedure consists of three consecutive steps: the automatic generation of target configurations and trajectories based on the nominal geometric models of the robot; a camera and obstacles, marker poses estimated by the two stage estimation algorithm; and the kinematic parameters identification based on a local product of exponential (local POE) formulized error model. The advantage of this self-calibration method is that the whole robot camera system can be calibrated without any manual intervention, which enables robot calibration to be completely online and suitable for the fast programming of the robot and computer vision combined work cell. A set of simulations and experiments on a UR5 robot demonstrate the convenience, efficiency and robustness of the proposed calibration procedure.