Abstract: For the problem of low absolute positioning accuracy in industrial robots, the Optotrak Certus HD optical motion tracking system made by NDI company in Canada is utilized as the measurement device for the robot pose, and an optimallypruned extreme learning machine (OPELM) algorithm based on the self-born weighted least squares (SBWLS) method is proposed. The algorithm realizes the pose compensation effect of the robot end-effector by mapping the robot target pose to the revised pose. In order to verify and analyze the validity of the error compensation method, the end-effector of the Epson's 6-axis robot is used to perform different motion modes, i.e., linear motion, circular motion, and discrete random motion, at a series of speeds in the experiment. The results show that the proposed error compensation method improves the pose accuracy of robot end-effector, and the total absolute positioning accuracy of the test point in X, Y, and Z axes is raised from 2 mm~3 mm to 0.06 mm~0.25 mm, which means the positioning accuracy is increased by an order of magnitude. What's more, the root-mean-square error and mean absolute error after compensation are reduced to 26.09% of the uncompensated errors. Meanwhile, the compensation method also significantly decreases the influence of the outliers, and possesses the outstanding robustness.