From bipedal locomotion to prosthetic walking: A hybrid system and nonlinear control approach

Abstract

When modeled after the human form, humanoid robots more easily garner societal acceptance and gain increased dexterity in human environments. During this process of humanoid robot design, research on simulated bodies also yields a better understanding of the original biological system. Such advantages make humanoid robots ideal for use in areas such as elderly assistance, physical rehabilitation, assistive exoskeletons, and prosthetic devices. In these applications specifically, an understanding of human-like bipedal robotic locomotion is requisite for practical purposes. However, compared to mobile robots with wheels, humanoid walking robots are complex to design, difficult to balance, and hard to control, resulting in humanoid robots which walk slowly and unnaturally. Despite emerging research and technologies on humanoid robotic locomotion in recent decades, there still lacks a systematic method for obtaining truly kinematic and fluid walking. In this dissertation, we propose a formal optimization framework for achieving stable, human-like robotic walking with natural heel and toe behavior. Importantly, the mathematical construction allows us to directly realize natural walking on the custom-designed physical robot, AMBER2, resulting in a sustainable and robust multi-contact walking gait. As one of the ultimate goals of studying human-like robotic locomotion, the proposed systematic methodology is then translated to achieve prosthetic walking that is both human-like and energy-efficient, with reduced need for parameter tuning. We evaluate this method on two custom, powered transfemoral prostheses in both 2D (AMPRO1) and 3D (AMPRO3) cases. Finally, this dissertation concludes with future research opportunities.