In an aging society, most of walking assistance device has been researched in an aspect of augmentation system which assists the human walking by torque enough. However the user depends on the performance of assistance device without the independent reliance. Therefore we researched the compensation system which offers the minimum torque required during the human walking. Furthermore the stability of mutual human-device should be ensured due to being the much distribution of user's independent will. Then, the assistance system can react to the human motion through the stability every moment. Therefore we propose the walking assistance system for the walking stability by using the human motion information. The walking assistance system is based on the compensation model and the walking stability is determined by ZMP and COG information for the human motion. To verify the verification of the proposed system, the relation between the ZMP and COG information is analyzed corresponding to the environment for the human motion. And the experiment for the walking assistance system applying the compliance controller is implemented in the two situations which are safe and dangerous.