An adaptive control method of input saturation command filtering based on radial basis function (RBF)neural network was proposed to solve the problems of limited control input, nonlinear friction, flexible deformation and unknown disturbance torque in the dynamic model of the robot joint. Based on the instruction filter backstepping method, the saturation function was used to constrain the amplitude of control input and the RBF neural network was used to approach the unknown disturbances. All the errors of the closed-loop system were proved ultimately uniformly bounded by using the Lyapunov stability theory. The simulation results show that the proposed control algorithm not only makes the control input amplitude of the system strictly constrained within the specified range, but also completes the high-precision tracking of the target trajectory (the tracking error is about ±0.003 rad). It can also resist the adverse effects of external step disturbance torque and modeling error on the control system, ensuring high accuracy and strong robustness of the system. The performance of the proposed control method is better than PID (propotional integral derivative) control and ordinary CFBC (command filter backstepping control) methods. It is of great value to the application and intelligent control of the robot joint in high precision field.