Warning:
This page is
provided for historical and archival purposes
only. While the seminar dates are correct, we offer no
guarantee of informational accuracy or link
validity. Contact information for the speakers, hosts and
seminar committee are certainly out of date.
We have built a three degree of freedom robot that bats two balls into simultaneous stable periodic vertical trajectories that commonly persist in excess of an hour. The juggling algorithm underlying this behavior relies on continuous estimates of ball position and velocity. This presentation will examine both the practical and theoretical issues involved in generating these estimates from a stereo camera system that views brightly illuminated white balls against a dark background. Despite this structured visual environment there remain a number of significant issues surrounding the efficient acquisition and use of camera data to generate the required information. Specifically, we have developed both a rational ``attention management'' scheme and a novel ``triangulating observer'' that ensures a stable flow of information even in the presence of unavoidable transient losses of data.
At a time when many in the robot vision community are exploring the benefits of ``visual servoing'' or have found the need for including ``attention mechanisms'' in their camera architectures, this account serves as both documentation of a particular system which incorporates most of the essential features of an ``active sensor'' yet remains simple enough to permit some formal analysis. Developing an architecture amenable to analysis yields a system whose run time behavior can be understood within a simple paradigm.
Developing and reasoning formally about this specific system interests us more generally in view of the apparent need to develop a theory and practice of ``dexterous robots.'' This term, as we understand it, denotes an autonomous machine capable of interacting with a dynamical world. The strategies of general interest to us are feedback algorithms which specify the manipulator's actions at each instant in time as a function of its current state and that of the world. For a juggling machine, the world's state reduces to the current position and velocity of one or two balls and the task of estimating this state forms the narrow focus of the presentation. It is our belief that a much larger range of dynamically dexterous tasks (of which juggling is but a simple example) will necessitate the ability to generate timely and accurate estimates for the state of a dynamical environment independent of the specific control algorithm.
Dr. Rizzi received his Ph.D. and M.S. in electrical engineering from Yale University, and his Sc.B. from MIT.
He has recently begun an NSF funded Post-Doctoral fellowship in the Microdynamic Systems Laboratory at CMU, working on issues of distributed real-time control and sensing in a mini-factory with Dr. Ralph Hollis.