A ZMP equilibrium equation stating the relationship of various moments applied to a robot body of a robot, based on desirable motion data made up by trajectories of respective parts, imaginarily divided from the robot body, is generated, and moment errors in a ZMP equilibrium equation are calculated. A priority sequence of the parts, the target trajectories of which are corrected to cancel out the moment errors, is set. The target trajectories are corrected from one part to another, in a sequence corresponding to the priority sequence, to compensate the moment errors.