CAREER: Using Multiple Gaits and Inherent Dynamics for Legged Robots With Improved Mobility

Regents Of The University Of Michigan - Ann Arbor, Ann Arbor MI

Investigators

Abstract

The goal of this Faculty Early Career Development (CAREER) Program project is to investigate legged robots that are faster and energetically more efficient. The project capitalizes on the currently untapped possibility of using different gaits at different locomotion speeds. This idea is inspired by nature. Humans, for example, switch from walking to running as they increase speed; horses transit from walking to trotting and galloping. Switching gaits is analogous to switching gears in a car. It increases versatility and reduces energy consumption. Additionally, in nature the choice of gaits is strongly coupled to an animal's morphology. A massive elephant moves differently than a filigree gazelle. This project will investigate this complex relation of gaits, motions, and morphologies, and will transfer the underlying principles to robotic systems. The work will be conducted in simulation studies and with actual robots. In the long term, this CAREER plan aims at the development of robots that reach and even exceed the agility of humans and animals. It will enable us to build robots that can run as fast as a cheetah and as enduring as a husky, while mastering the same terrain as a mountain goat. Moreover, it will provide us with novel designs for active prosthetics and exoskeletons. The project will also leverage the fascinating topic of legged locomotion to spark the interest of K-12 students, underrepresented groups, and a broader audience for science, technology, and engineering. This project seeks to understand the fundamental principles of designing, building, and controlling legged robots that embrace and exploit their inherent mechanical dynamics. The underlying premise is that locomotion can emerge in great part passively from the interaction of inertia, gravity, and elastic oscillations. The goal of this project is to identify systems in which such dynamics can be excited in a variety of different modes. Different modes would correspond to different gaits, and would enable efficient motion in different operational conditions. The work will extend optimal control and machine-learning techniques, such as multiple shooting, direct collocation, or probabilistic direct policy learning methods, to allow the simultaneous generation of gaits, motions, and morphological parameters (including stiffness values, mass distributions, or actuator sizes). Since performance criteria such as speed, efficiency, robustness, and agility only fully manifest themselves in actual hardware implementations, the research team will study these concepts not only in simulation, but also with hardware prototypes. In particular, the methodology will be employed to examine the benefits of flexible spines in quadrupeds and of articulated ankles in bipeds, as well as to compare series elastic actuation and parallel elastic actuation concepts.