As a novel biologically inspired underwater vehicle, a robotic manta ray (RoMan-II) has been developed for potential marine applications. Manta ray can perform diversified locomotion patterns in water by manipulating two wide tins. These motion patterns have been implemented on the developed fish robot, including swimming by flapping fins, turning by modulating phase relations of fins, and online transition of different motion patterns. The movements are achieved by using a model of artificial central pattern generators (CPGs) constructed with coupled nonlinear oscillators. This paper focuses on the analytical formulation of coupling terms in the CPG model and the implementation issues of the CPG-based control on the fish robot. The control method demonstrated on the manta ray robot is expected to be a frame- work that can tackle locomotion control problems in other types of multifin-actuated fish robots or more general robots with rhythmic movement patterns.