Traditionally, robot control has been done typically by "highly precise control algorithms": the position of each movable body part is accurately determined at any time with vast amount of computation. This, however, causes serious problems, particularly in terms of adaptability and energy efficiency. On the other hand, an extreme approach has been gaining a lot of attention recently. A good instantiation is the passive dynamic walker, driven only by exploiting the intrinsic dynamics of its mechanical system. However, the mechanical system is not everything, just as the control system is not everything; "well-balanced" coupling between control and mechanical systems should be considered. In addition, the "meeting point" between the two systems should be flexibly varied according to the environment encountered. In light of these facts, this study intensively focuses on the stiffness of robots' joints, since this effectively influences the dominance relationship between control and mechanical systems. More specifically, the aim of this study is to develop a "real-time tunable spring" that can smoothly change its elasticity without changing its natural length, allowing robot's joints to change their position and stiffness independently.