Abstract: To control the multi-mode motion of the 3D snake-like robot, a motion control method based on the double-layered central pattern generator (CPG) is presented. The proposed double-layered CPG network includes the rhythmical layer and pattern layer. The CPG neurons in rhythmical layer control the phase relation between pitch joint groups and yaw joint groups of the 3D snake-like robot. The CPG neurons in pattern layer determine the phase relation and trajectory of each joint in the identical group of the 3D snake-like robot. Firstly, the CPG neurons are modeled by the Kuramoto oscillator. The layered architecture and the coupling topology of the proposed CPG network are determined. Secondly, the control parameters of four kinds of locomotion of the 3D snake-like robot, e.g. lateral rolling locomotion, sidewinding locomotion, slithering locomotion, and steering locomotion, are calculated based on the constraints of serpenoid curve. Finally, the control performance of the proposed double-layered CPG network is evaluated by the co-simulation and experiments. As shown in the experimental results, the actual velocity of the 3D snake-like robot is 3.9 cm/s, 9.0 cm/s, 2.1 cm/s, and 10.8°/s respectively when the snake-like robot performs lateral rolling locomotion, sidewinding locomotion, slithering locomotion, and steering locomotion. Consequently, the proposed method can control the multi-mode motion of the 3D snake-like robot effectively and flexibly.