Abstract: Noise is inevitable in the multi-task control process of redundant manipulator. Traditional control algorithms usually assume that the control process is carried out in an ideal environment without noise, and partially robust control algorithms have shortcomings in multi-task execution. The multi-task control problem of redundant manipulator is modeled as a multilayered hybrid time-varying problem, and different tasks are described with different time-varying problems of different layers. In solving the multilayered hybrid time-varying problem, an integral error function is constructed and the zeroing neural dynamics formula is used more than one times to ensure the strong robustness of the final algorithm. The robustness of the proposed algorithm under constant noise, linear increasing noise and bounded random noise is verified by theoretical analysis and numerical experiments. Besides, the performance of the proposed algorithm is compared with the classical control algorithm.