The stabilization of most artificial systems has been achieved by sensor based state feedback control with high signal transmission speed and high computational power, and stiff structures. In contrast, many biological systems can achieve similar or superior stable behavior with low signal transmission speed and low computational power via nervous system, and flexible structures. In order to explain this phenomenon, our research group focused the concept of self-stabilization of musculoskeletal system. Self-stabilization is defined as the ability to restore its original state after a disturbance with feedforward control. In our previous research, we analytically investigated the self-stabilizing condition of biological musculoskeletal system using the Lyapunov stability criteria and come to a conclusion that stiffness and viscosity of the joint play significant role in self-stabilization. Particularly, there exist two types of stiffness in biological muscle; one is spring-like passive stiffness and the other is active stiffness that is proportional to muscle activation. We believe that active stiffness plays a significant role in self-stabilization for dynamic movement. In this paper, we develop an active stiffness mechanism that can assign self-stabilizing function to a robotic arm. As a result, theoretically predicted self-stabilizing function is experimentally verified and explains why biological musculoskeletal system can be stabilized with feedforward control.