Control of the human-robot interaction system presents many challenges, which include the consideration in terms of the properties of human operators, sensor device, and linkage mechanisms of the robot. This paper presents the application of a variable mechanical stiffness control based on a human stiffness estimation. In the controller design, dual disturbance observers with respect to two operation modes, namely the common mode and the differential mode, are designed and applied for controlling wire rope tension and interaction force of human. The human stiffness estimation plays a dominant role in achieving this intelligent behavior, and smooth interaction force, by allowing a robot system to adapt the mechanical stiffness of the twin direct-drive motor system. The advantage points of the high mechanical bandwidth and low stiffness transmission are combined together. The experiment results from two separate experiments show that the above strategy was able to regulate the mechanical stiffness of the robot and provide a smooth interaction force.