|
Today, as has been the
case for decades, there is tension between those involved developing the
theory of automatic control and those applying it in the market place. This
tension is exemplified, on one hand, by the extraordinary hyperbole and over
zealous salesmanship associated with various "low tech" approaches to
control (think fuzzy washing machines) and, on the other hand, by the
unwillingness of most control theorists to modify their "high tech"
approaches so as to make them relevant to problems in which saturation,
exception handling, and rule based designs are the most effective way to get
the job done (think
m-synthesis).
This non-meeting of minds has generated considerable discussion but rather
little in the way of new ideas. The purpose of this talk is to describe a
point of view on the design of controls which can lead to principled
approaches to design, based on methodologies that do not require one to
abandon the scientific method or to corrupt the various modes of thought
that have served science and engineering so well in the past. We seek to
show that there are substantial benefits to be reaped if one includes in the
optimization process terms that reflect the cost of implementing the control
laws, and that when one does so the resulting control law bears more
resemblance to something practical. The appropriate form for the new
performance measures, and the nature of the solutions of the optimization
problems that they lead to, will be discussed and illustrated. In particular
we will consider the idea of including in the performance measure something
akin to what psychologists call attention and go on to argue that by
reducing the attention required one makes the control law easier and less
expensive to implement. There will be some brief video clips illustrating
laboratory implementations of these ideas relating to robotics.
Roger Brockett is An Wang
Professor of Electrical Engineering and Computer Science at Harvard
University. He received a B.S., M.S., and Ph.D. in Engineering from Case
Institute of Technology. He is active in the field of automatic control and
its applications, having worked in areas as diverse as nonlinear filtering,
robotics, quantum control and differential geometric methods. His current
research interests include robotic manipulation, computer vision, and
intelligent machines. At Harvard, he founded the Harvard Robotics Laboratory
and is Director for the Center for the Dynamics and Control of Smart
Structures. His awards have included the IEEE Control Science and
Engineering award, the American Automatic Control Council’s Richard Bellman
Award, and the Reid Prize from SIAM for work on Differential Equations and
Control. He was elected to the National Academy of Engineering in 1991.
|
