<p>A manipulator often cannot converge rapidly within finite time and has low tracking accuracy owing to factors such as manipulator model errors and external disturbances. To address these problems, this paper proposes a time-varying second-order sliding mode control of super-twisting based on radial basis function neural networks. First, three radial basis neural networks are optimized by the gradient descent method to approximate the parameters of the dynamic model of the manipulator, thus getting rid of the dependence on the model. Second, a time-varying nonsingular fast terminal sliding mode surface is designed to achieve fast convergence in finite time. Meanwhile, an improved super-twisting algorithm is adopted in combination with the boundary layer concept, which not only eliminates the chattering but also improves the tracking accuracy. Finally, simulation experiments are carried out by simulating a 2-DOF and a 6-DOF manipulator. The results verify the effectiveness and superiority of the proposed method.</p>