Page 98 - Anatomy of a Robot

03_200256_CH03/Bergren 4/17/03 12:27 PM Page 83
COMPUTER HARDWARE 83
can support all the requirements of our robot without further analog design effort,
we can come out ahead. Consider Analog Devices’ mixed signal family at
www.analog.com/technology/dsp/mixedsignal/index.html.
Display systems Many robots require control panels or information displays. It
is not difficult to integrate a liquid crystal display (LCD), even a large one, into a
computer circuit these days. Many computer chips can support LCDs directly.
Low-power units The handheld personal digital assistant (PDA) market, along
with phones and pagers, has spawned a whole series of computer chips that can
operate on very low levels of voltage and power. If the power for our robot’s com-
puter system is a significant part of the power budget, then consider low-power
computer systems. Many other techniques for saving power in computer systems
can be used as well. We’ll visit power control later in the book.
Game units It’s a little-known fact, but most computers wind up in games.
That’s right. The sheer number of computers going into toys dwarfs the other prac-
tical uses. These are generally very small computers that cost next to nothing.
They’re found in toys like Furby, digital pets, talking dolls, and so on. It is not easy
to deal with the suppliers of these computers; they demand huge orders.
A look under the covers of a small robot made with such a chip is provided at
www.phobe.com/furby/. Furby and Furbies are the intellectual property of Tiger
Electronics.

Parallel Processors

Parallel processing is not new. The method stems from the realization that many com-
putational problems do not have to be executed one step at a time. Often, a computa-
tional problem can be broken down into problems that can be executed simultaneously
without fear that the work done on one problem will obviate the need for work on the
other problem. In WWII, the atomic bomb project employed dozens of people who sat
at mechanical calculators performing computations in parallel.
Most modern general-purpose processors (like those from Intel or Motorola) already
contain more than one computer within the chip. This is done because almost every
computational problem can benefit at least in some ways from parallel processing.
Consider for a moment the work done in the following software pseudo-statement: If
A, then B, else C. The serial way to process this statement is to compute A, and then com-
pute either B or C. With three processors at our command, we could compute A, B, and
C all at the same time. When this single phase of computation is complete, the computer
merely chooses (based on A), either B or C as the answer. This can save one computer
cycle. It’s true that a third of the work is wasted, but the program runs twice as fast.