Abstract: The stability control of underactuated 3-D biped robot is still a hard nut to crack, as a result of locomotion characteristics which mix high dimension, strong nonlinearity and underactuation. Some traditional control methods, such as event-based feedback control and PD control, are poor in robustness and weak in resistance to external disturbances. Through observation, it is certain that humans adjust gaits tactically to regain stability when they are affected by external disturbances, by contrast with trying to keep the stability sustained by only one gait which is quite limited. Inspired by this, a control method based on gait transition is proposed for the underactuated 3-D biped robot. First of all, taking the minimum energy consumption as the optimization goal, a multi group of gait and step gait is designed as the reference gait to build a gait library by nonlinear optimization method. Secondly, to obtain an optimal performance in terms of the balance between the stability and input torques, a multi-objective gait transition function is established. Finally, a reference gait that minimizes the gait transition function is obtained by solving a quadratic optimization problem, and it is then used as the walking gait for the next step with the purpose of using gait library (multiple trajectories) method to reach the goal of improving robustness. In the simulation experiment, using the proposed gait transition control method, the underactuated 3-D biped robot can walk through the rough ground with the relative height varying within the range $-$20,20 mm without falling down, in contrast to the failure of previous one-gait control method. The results show the effectiveness of the method.