RI: Small: A New Mechanical Coupling Metric to Enable Effective Biped Locomotion Control

University Of Notre Dame, Notre Dame IN

Investigators

Abstract

Legged locomotion has evolved to provide the primary means of locomotion for land-based mammals, including humans. While highly-structured or artificial environments may lend themselves to alternative means, legged locomotion still provides the most effective way to traverse many locales. These include uneven natural terrains, human-designed environments featuring steps and staircases, and disaster rubble consisting of natural and/or human-made elements. Because they mimic the human structure, biped robots are of particular interest for operation in human-designed environments and performing human-assistive tasks alongside their human counterparts. Several decades of research efforts have sought to move biped robots out of laboratories and enable them to perform a wider range of practical tasks in real-world environments. A fundamental challenge that remains before this vision is fully realized, though, is how to enable robots to move in the very dynamic and efficient manner of humans while still maintaining the excellent stability that humans exhibit. Most modern biped robots either move relatively slowly and thus expend an excessive amount of energy or else move more rapidly and efficiently, but are too susceptible to falls caused by only minor disturbances. This project seeks to enable the development of biped robots that move more like humans, in terms of both efficiency and robustness. It formulates and applies a new, more general metric of dynamic robot control and stability that exploits the underlying mechanics of human-like locomotion, demonstrating its utility in laboratory robot experiments. The broader impact centers around encouraging K-12 students to pursue higher education in the STEM disciplines by highlighting engineering connections between robots and children¹s own primary means of locomotion -- walking. This project seeks to formulate and experimentally validate a novel approach to biped locomotion control based on an analytical measure of the control authority over the unactuated degrees of freedom. Legged locomotion systems are inherently underactuated because there are no actuators at the points of contact between the feet and the ground. The approach in this project exhibits the four characteristics of a universal dynamic control authority metric that would enable the design of controllers to produce gaits that are both robust and efficient, characterized by 1) generality of application to a range of movements; 2) flexibility for use with a variety of mechanical designs; 3) consistency with the natural dynamics to yield energy efficient gaits; and 4) analyticity to enable systematic mechanical and control design. The basic approach is, at any state, to decompose system velocities into directions that are aligned with the inputs, termed the controlled velocities, and directions orthogonal (with respect to the kinetic energy metric) to the inputs, termed the orthogonal velocities. A computable measure of coupling between the controlled and orthogonal velocities quantifies the control authority of the system over the unactuated DOFs and is the basis for the new dynamic control metric. Configurations where there is complete decoupling between the controlled and orthogonal velocity directions are referred to as dynamic singularities. Strong coupling is advantageous for the system to have good disturbance rejection properties, while weak coupling is desirable for the system to have good disturbance isolation properties. Thus, the project examines how a thorough understanding of the nature of the disturbances can be used to exploit dynamic singularities in the control of dynamic biped gaits.