I. Introduction
Controlling the stiffness of the joints in a robot has a major influence on its versatility, safety and capability to adapt [1]. Stiff robots are traditionally used in industrial pick-and-place applications with highly specialized control strategies and often large payloads, whereas soft robots are starting to be established in rehabilitation, prostheses, manipulation, wearable and bio-inspired robotics [2], often with lower payloads. If a robot is to perform tasks from both regimes, a strategy for varying its joint stiffness is required. On the one hand, stiffness variability can be achieved purely by the control of the robot. On the other hand, such potentially complex control can be offloaded to the mechanical design of the robot by introducing new materials or mechanisms that inherently possess compliance. Focusing on the latter option, many prototypes with mechanical designs of variable stiffness joints have been the subject of research, ranging from pretensioning of springs [3], Series Elastic Actuators (SEA) [4] [5], tightening tendons [6] to Shape Memory Polymers (SMP) [7] and magneto-rheology [8].