Based on the framework of repetitive motion planning (RMP), this paper extends the existing repetitive motion planning methods. By employing dynamic neural networks and integrating gradient descent with velocity compensation techniques, the theoretical feasibility of the proposed approach is rigorously validated. Given the necessity of maintaining sufficiently low errors for redundant robotic manipulators performing tasks on precision instruments, improving the control strategy is of critical importance. In this paper, a tanh activation function \(\sigma = \tanh (t/2)\) is introduced to achieve a time-dependent adjustment of the feedback intensity, effectively suppressing joint deviations and ensuring global stability. Finally, simulations conducted on various redundant robotic manipulators compare the applicability of the RMP framework, highlighting the effectiveness of the dynamic neural network-based solution.