This paper proposes a bipedal locomotion controller using an optimized Neuro-Fuzzy inference structure. In which, it is based on an adaptive neuro-fuzzy inference identifier (ANFII) for modeling the inverse dynamics of the biped robot. The proposed learning scheme is based on the zero moment point (ZMP); while the modeling is accomplished, the parameters of the bipedal locomotion controller will be duplicated by those of the ANFII respectively to provide favorable control actions. As the simulation results show, the proposed controller can generate a stable walking cycle for a biped robot. Moreover, the experimental results demonstrate that the performance of the proposed controller is satisfactory whenever the robot stays in different postures or moves on a rugged surface.