|
|
Foundations of Robotics Seminar, November 2, 2010
Time
and Place | Seminar Abstract
Accurate, Low-Dimensional Modeling of Fully Elastic Robots
Lael Odhner
GRAB Lab
Yale
NSH 1507
Talk 10:30 am
There is a trend in robotics toward replacing revolute joints with elastic members. Elastic robots can be designed to have few moving parts, and can be fabricated inexpensively at macro, meso, or micro scales. One of the big challenges in elastic robot design and control is in finding accurate, computationally simple models for analysis and control. One critical measure of this complexity is the number of parameters needed to accurately describe the configuration and behavior of a robot. A traditional jointed robot can be described by a low-dimensional vector of joint angles. In contrast, an elastic body has a continuum of possible configurations. Accurate finite-dimensional approximations, such as finite element models, often have a large number of degrees of freedom. Thus, a relatively simple elastic robot may require dozens or hundreds of configuration variables to perform simple Jacobian-based analysis, to compute instantaneous kinematics or to resolve a balance of forces on the robot. In this talk, I will introduce a method for representing elastic joints using a very small set of parameters, using the framework of smooth function approximation. These joint parameters describe the configuration space of an elastic-jointed robot, as well as the elastic energy associated with the joint motion. This method is accurate for large-angle deflections, and captures not only first-order effects such as deformation under load, but also second-order effects such as compliance and buckling. I will also discuss how these models are being applied to develop inexpensive elastic robot hands capable of performing dexterous tasks.
The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.