Emergent Simplicity: Robophysical Template Discovery via Geometric Mechanics and Learning

Abstract

Scientists, engineers, inventors, students, and citizen scientists have been mesmerized by the efficacy at which biological systems like snakes and lizards use seemingly simple rhythmic body and limb undulations (what we will refer to as patterns of ``self-deformation") to traverse their natural terrains. Robot systems, on the other hand, do not yet possess robust real-terrain mobility, constraining their ability to operate effectively in multicomponent landscapes. We seek to develop a framework to attain robust mobility, integrating advances in geometric understanding and discovery of high level control targets, and passive mechanical properties of appropriate self-deformation schemes. We posit that such a framework should be meaningfully simple. If we can approximate complex biological (and robotic) motion with few variables, we will truly understand first-order principles, which will ultimately lead to a long-term high fidelity understanding of how self-propulsion emerges from appropriate coordination and control of bodies and limbs. Numerous biological and robotic reduce complex, many-degrees-of-freedom systems to low degrees of freedom models, but such reductions can be non-trivial. Such a reduction is necessary for analysis and control. Our prior work achieved reduction using geometric techniques where we established linear relationships between system inputs and outputs, operating in various terrains, and parameterized these relationships with as few as two variables. This work made certain assumptions about the nature of contact between the system and the environment; one such assumption is that contact remains constant, e.g., the system is immersed in a fluid or a granular medium. However, most terrestrial locomotor tasks require interaction with a diversity of terrain--making and breaking contact with different materials, etc. The proposed work will bring systems with varying environment-system contact states into reach of ``low-dimensional geometric methods, thereby allowing us to seek unifying principles as well as pose specific hypotheses about "good" control and coordination in diverse systems such as sidewinding snakes, snake robots, cockroaches, hexapod robots, centipedes, elongated long lizards, etc.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 09, 2020
Source ID
W911NF2010129

Entities

People

  • Howie Choset

Organizations

  • Army Contracting Command
  • Massachusetts Institute of Technology
  • United States Army

Tags

Readers

  • Agent-Based Social Robotics and Mobile-Assisted Learning in Virtual Environments.
  • Robotics and Automation.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - Machine Learning Algorithms
  • Autonomy
  • Autonomy - Autonomous System Control