Walking motions of biped robots have features of humanoid and intelligent. Generally, when we human are walking, we may control the walking motions using qualitative instructions, such as linguistic "slow" or "fast," "forward" or "backward," "big step" or "small step". This motivates us to design a fuzzy controller for biped robots. Furthermore, to improve the robustness of the biped walking system, a framework of biped gait control based on a predictable type-2 fuzzy logic controller (T2FLC) is proposed to ensure the dynamic balance of the biped under the condition of complex process noises and measurement noises. Different with existing controllers for biped robots, a state estimator based on square root unscented Kalman filter (SRUKF) is incorporated in the proposed control strategy. Using the estimated biped states as inputs, the proposed T2FLC can predictably adjust the posture of the trunk timely and properly to ensure the dynamic balance of the whole legged system. Simulation results are presented to verify the effectiveness of the proposed method.