A prediction and compensation method of robot tracking error considering pose-dependent load decomposition

•A prediction and compensation approach of robot tracking error is developed by using a TCN model.•The disturbance of terminal load decomposed on each joint is considered as input feature of the error prediction model.•A data enhancement method is developed by splitting and integrating the position, velocity and the effect of load.•The predicted errors are utilized to pre-compensate commands to improve the tracking accuracy of joints.