To make the chaotic gaits converge to a stable cycle gait, a control algorithm is proposed for the biped robot with knees. Firstly, the passive gaits changes with the changing of the slope are studied by using the Poincarésection method. It shows that the increasing of the ground slope will result in the appearance of period-doubling and chaotic gaits. Then, based on the biomechanics in human walking and delayed feedback control theory, bionic walking control based on adaptive constant torque and sensory feedback for the biped robot is designed. The controller parameters are regulated gradually in terms of the initial states of the current step and the first two steps. Finally, the chaotic gaits converge to the cycle gaits with this controller. Simulation results show the effectiveness of the presented control algorithm.