Industrial robots performing continuous-path operations require high trajectory tracking accuracy, yet repeatability at isolated poses does not guarantee precise path following because joint tracking errors vary with motion history, robot configuration, and gravity effects. This paper proposes a temporal convolutional network (TCN)-based trajectory tracking error compensation method incorporating gravity-torque features representing both terminal-load and robot self-weight effects. The gravitational effects of the terminal load and robot links are mapped into joint-space torques using the corresponding Jacobians, and these torque features are combined with commanded joint angles and angular velocities for six-joint tracking-error prediction. The predicted errors are then processed using Savitzky–Golay smoothing and an adjacent-sample increment constraint before being added to the original commands for pre-compensation. Experiments were conducted on a six-degree-of-freedom industrial robot to evaluate the proposed method. Compared with several other methods, the proposed method achieved the lowest residual mean absolute error of 2.815 mrad. For the composite trajectory, the mean absolute joint tracking error, mean end-effector position-error norm, and mean rotation-vector orientation-error norm were reduced by 83.42%, 82.13%, and 82.84%, respectively.

trajectory error compensation,industrial robots,gravity-torque features,temporal convolutional network,offline pre-compensation