August 12, 2003
A: FROM MENTOR NORRIE ROBBINS IN CA
Very good question, Alexis. You are asking a question on the cutting edge of
robotics research. Making a machine act like a human is hard. We value our
high level thinking (for example, as in playing a complex game like chess)
and ignore our ability to walk because it takes no 'thought' on our part.
Yet, it is easier to get a machine to play chess than to get it to walk.
That's because high level thought such as chess involves fixed rules and is
a question of finding a solution in a large, but finite constraint space.
Walking (and talking!) involves solving a problem in a space without
constraints. When constraints are applied (for example, a smooth surface, no
obstacles, no wind, fixed gravity, no incline), a robot can be built to
perform pretty well. Because robotics stemmed from computer programming,
algorithmic solutions have been sought to tell a robot how to walk,
including, as you mentioned, fuzzy logic. Fuzzy logic is useful because it
is not 'black and white' thinking, and decisions can be made in spite of not
knowing whether one's foot is on solid ground. Some researchers noticed that
babies learn to walk without being 'told how.' Researchers developed
'genetic algorithms' that act much like a baby's brain, to learn by trial and error.
Of course, it takes a baby two years to learn. People don't want to wait two
years to see if their million dollar 'baby' is going to learn. One way to
speed up the process is to be 'intelligent' about how one learns. When it
comes to the decision whether the current trial is going to lead to success
or failure, there is an opportunity to use any kind of approach one wants to
try, even fuzzy logic. This is the basic question, for the robot, for the
baby, even for the chess playing computer: how do you know if this choice is
a good choice? What constitutes 'good'? What looks 'bad' locally may be good
(as in heading south to get to the freeway to go north). In chess games,
when a master makes an unusual move (which probably looks initially 'bad,'
like losing three of his/her pieces, but ultimately leads to checkmate),
they mark it with an exclamation point. In robotics, everyone is looking for
those exclamation point moves that will keep the robot up and running
in spite of gravity, low batteries, etc. A human corollary is sea legs.
That's your nervous system 'operating system' trying to get your sensors recalibrated.

There are lots of universities studying these problems. Just a couple
include:
University of Massachusetts (www-robotics.cs.umass.edu/)
Carnegie-Mellon (http://www.ri.cmu.edu/)

A good page to find links to universities and companies doing robotics is
http://www.cs.indiana.edu/robotics/world.html#software

Here are some other web sites you may want to look at:
Although getting out of date, from Carnegie-Mellon, some answers to general
questions regarding robotics: http://www.frc.ri.cmu.edu/robotics-faq/

To build your own, try http://mindstorms.lego.com/eng/default.asp?bhcp=1
And for an operating system, try http://www.noga.de/legOS/
For hardware, try http://www.robotstore.com/

from Dr. Thea Iberall, computer scientist sister of Mentor
Eleanora I. (Norrie) Robbins