In most existing pedestrian dynamic models, agents are modeled as sliding or jumping objects. As a kind of bipedal creature, however, this assumption makes it difficult to include some important human walking characteristics in the models, such as the periodicity of stepping, the adjustment of step frequency and step length, and the change of personal space within the same step cycle. To take these characteristics into consideration, a "Biped Model" is developed to simulate pedestrian locomotion by physical step rather than by time step. The model consists of two components: the stepping simulator and the adjustment of velocity for each step. Simulation results show that the velocity-density relations are consistent with the empirical data. Stable stop-and-go waves are simulated with a critical density. Furthermore, with the considering of biped movement, the lock-step phenomenon is reproduced. The model provides new possibilities to study crowd behavior while considering personal bipedal mechanics.