In this paper, the authors devise a method that a biped walking robot realizes a learning process of walking stabilization with the cooperation of a human, by imitating the learning process of a baby. We have developed a biped walking robot which has a trunk, and equipped the robot with a ZMP (Zero Moment Point) measurement system, which is composed of two universal force-moment sensors, and with a mechanical interface for the cooperation. We performed walking experiments with the biped walking robot, in which a human directly taught the robot the desired ZMP, and the robot learned to stabilize its walking with the cooperation of the human. As a result, the robot acquired the taught desired ZMP by modifying its trunk motion and it achieved stable walking by itself through the process of several walking trials.