MESTRAN, a variable stiffness actuator for energy efficient legged robots

One of the key features that enables animals and humans to perform agile, robust, adaptive yet ef- ficient locomotion is their body’s complex muscle-tendon-ligament system. Such systems provide body and limbs with the functionality that is used to efficiently absorb external shocks and ex- change of mechanical energy, e.g. kinetic and potential energy, to exploit natural dynamics during locomotion. In biology, it has been found that animals and humans adjust their limb stiffness to ac- commodate for different speeds, gaits, and terrains. On contrary, in the field of legged robots, little has been known about how to control leg stiffness to efficiently adapt to changes of speed, terrain, and gait or stride frequency at which the leg oscillates. Therefore, this thesis aims at contributing to the primary understanding of the topic. Until today, mechanical springs with fixed spring constants are still widely used as energy sav- ing mechanisms and shock absorbers for legged robots. However, the compliance of those springs is not adjustable and manual assembly is required to make a robot leg stiffer or more compliant. Motivated by this fact, we present a systematic development and evaluation of a new variable compliance/stiffness actuator, named MESTRAN (MEchanism to vary Stiffness via Transmission ANgle) in this thesis. This actuator serves as a key tool to investigate energy efficient locomotion at various stride frequencies and on surfaces with different stiffness. MESTRAN can dynamically alter joint stiffness in an unlimited range. It is also capable of maintaining the stiffness without requiring energy and offering different types of compliance, e.g. linear, quadratic, or exponential. In this thesis, we first designed and constructed an adjustable stiffness leg based on the MES- TRAN design. We then validated the design by conducting a series of experiments by using the first leg prototype. Second, in order to investigate hopping locomotion with variable stiffness ca- pability, we designed a single-legged robot, named L-MESTRAN (Linear-MESTRAN), which is an advanced version of the MESTRAN leg. We systematically analysed and demonstrated the me- chanical performance of the legged robot using the simulations and a number of real-world hop- ping experiments. As a result, we found that a proper adjustment of leg stiffness can improve the hopping energy efficiency of the robot at various stride frequencies. Third, this finding was also investigated on surfaces with different stiffness by using the L-MESTRAN robot. The simulation and experimental results indicated that, for a particular stride frequency (3 - 6 [Hz]), the adjust- ment of the knee stiffness can accommodate for changes in surface compliance, resulting in an improvement of the energy efficiency of hopping