Robotics Reference
In-Depth Information
attach it in ways that are plausible for a real body. AARON also knows
some useful things about faces, for example that a face has eyes, a nose
and a mouth, and that there are limits to where they can appear and what
size they can be.
AARON can draw or paint anything it knows about, but in fact it
knows about very little: people, potted plants and trees, simple objects
such as boxes and tables, and how to decorate the background of a pic-
ture. It knows never to put a tree in a pot, or to put a person in a
pot, though in some art movements such devices would be perfectly ac-
ceptable. Cohen could have made a policy decision to teach AARON
about more things, different things, but instead his work has concen-
trated on attempting to make AARON draw and paint better, preferring
quality to quantity. As part of this quest for quality, and spurred by a
desire to have AARON paint a portrait of one of his friends, a poet, Co-
hen started to wonder whether AARON could be persuaded to produce
a recognizable likeness of a particular person. For almost two months
he developed AARON's knowledge of the structure of heads and faces,
during which process the number of data points needed to represent a
head and face grew from a few dozen to several hundred, while at the
same time the structural development of software for face generation ad-
vanced considerably. Giving AARON the ability to create more detailed
faces and heads than had hitherto been possible, necessitated organising
the data points in AARON's representation of a face into distinct parts.
Instead of having simply a set of points for a face, there was one set for
the upper mouth, another for the lower mouth, a set for a beard, one for
the forehead, an eyelid, the lower part of an eye, and so on, in such a
way that the individual parts of a face could be scaled up or down in size
and moved around the image at will, thereby enabling AARON to create
heads and faces that showed a true perspective of the three-dimensional
image. When Cohen completed this expansion of the program's facial
knowldge, AARON knew enough to generate from it a varied popula-
tion of highly individualized physical and facial types, with a range of
haircuts to match.
AARON remembers everything it does in a drawing or painting,
building up a very elaborate internal representation of an image as it
develops. It has to remember where the lines are, partly so that each line
corresponds to an appropriate part of the final creation and partly so that
AARON can colour and fill in a creation without daubing one brush
stroke unintentionally on top of another. In order to get everything in
