Bipedal gait is the natural means of human locomotion. Nonetheless, it is still unclear how the central nervous system coordinates the whole-body segments for gait generation. We address this question based on the well-known hypothesis that the human motion is the result of an optimization process. We consider a reduced set of criteria taken from the observation of human walking and the study of the related literature, which seem to be optimized during the human gait. Differential Dynamic Programming is applied on these criteria with a 3D whole-body skeletal model involving 42 degrees of freedom to generate walking motions. Nine different skeletal models and gaits reconstructed from motion capture data are used to this end. The simulated walking motions are then analyzed and compared to the human reference to show the quality of the gait generation process. The interest of this optimization approach for human-like motion generation is finally discussed.