The Vision of Robots
11
human perspective it is perfectly arbitrary. To Shakey, it does not “look like”
anything because Shakey is a machine. It is only the mind of a living beholder
that perceives the world in shapes. The machine merely checks whether two
objects match when compared.
A template is not necessarily a stencil or rigid pattern, but may be a set
of instructions for recognizing the form mechanically. In order to identify a
pattern mechanically, a computer must decompose the image—break it up into
simple components—and analyze how it has been assembled from elementary
parts. In contrast, animal vision grasps the image as a Gestalt, an unbroken
whole, and recognizes it from its configuration as a single undivided unit, the
entire image being taken into account simultaneously.
Let us whimsically pretend that Shakey has some form of awareness of what
it is seeing. Unlike our visual awareness, which is richly structured and admits
endless variations of form and shape, Shakey’s visual “awareness” has exactly
three settings. These three discrete options are the whole of its visual world: It
is either 1 or 2 or 3, and for Shakey there is nothing more in creation.
Let’s now imagine that a thermostat has awareness:The thermostat’s function
is to turn a furnace on when the ambient temperature falls below 70 degrees,
and off when the temperature rises above 72. The thermostat’s external world
(its world-model) consists of one bit of information—a cipher that can exists
in two possible states: For the sentient thermostat, that is the whole of reality.
Multitudes of robots like Shakey—but far more sophisticated—work in
factories everywhere. For example, there are robots that pick sundry items out
of a bin, detect their shape, then place them in a standard orientation on a
conveyor belt. In order to do this, the robot visual system has a template-like
shape detector able to correctly select among a small number of fixed shapes.
A robot might, for example, distinguish flat objects of different sizes whose
cross-section is either triangular, square or hexagonal: The robot would then
adjust a grasping hand to conform to the object. As above, the three shape
options are distinguished from one other by means of a template: It is purely
in the template sense that the machine’s vision “recognizes” shapes. But most
crucially, there is a built-in association between the shapes of the objects, and
the permissible openings of the grasping hand.
In other words, there is a correspondence which associates each shape with a
specific adjustment, or opening, of the mechanical hand. For the machine, the
“meaning” of a given shape is the hand-opening it corresponds to: Clearly, the
machine has no geometric conception of what it’s doing—and needs none: For
the machine, to see means to correctly associate a few stock shapes with settings
of its grasper.
11
human perspective it is perfectly arbitrary. To Shakey, it does not “look like”
anything because Shakey is a machine. It is only the mind of a living beholder
that perceives the world in shapes. The machine merely checks whether two
objects match when compared.
A template is not necessarily a stencil or rigid pattern, but may be a set
of instructions for recognizing the form mechanically. In order to identify a
pattern mechanically, a computer must decompose the image—break it up into
simple components—and analyze how it has been assembled from elementary
parts. In contrast, animal vision grasps the image as a Gestalt, an unbroken
whole, and recognizes it from its configuration as a single undivided unit, the
entire image being taken into account simultaneously.
Let us whimsically pretend that Shakey has some form of awareness of what
it is seeing. Unlike our visual awareness, which is richly structured and admits
endless variations of form and shape, Shakey’s visual “awareness” has exactly
three settings. These three discrete options are the whole of its visual world: It
is either 1 or 2 or 3, and for Shakey there is nothing more in creation.
Let’s now imagine that a thermostat has awareness:The thermostat’s function
is to turn a furnace on when the ambient temperature falls below 70 degrees,
and off when the temperature rises above 72. The thermostat’s external world
(its world-model) consists of one bit of information—a cipher that can exists
in two possible states: For the sentient thermostat, that is the whole of reality.
Multitudes of robots like Shakey—but far more sophisticated—work in
factories everywhere. For example, there are robots that pick sundry items out
of a bin, detect their shape, then place them in a standard orientation on a
conveyor belt. In order to do this, the robot visual system has a template-like
shape detector able to correctly select among a small number of fixed shapes.
A robot might, for example, distinguish flat objects of different sizes whose
cross-section is either triangular, square or hexagonal: The robot would then
adjust a grasping hand to conform to the object. As above, the three shape
options are distinguished from one other by means of a template: It is purely
in the template sense that the machine’s vision “recognizes” shapes. But most
crucially, there is a built-in association between the shapes of the objects, and
the permissible openings of the grasping hand.
In other words, there is a correspondence which associates each shape with a
specific adjustment, or opening, of the mechanical hand. For the machine, the
“meaning” of a given shape is the hand-opening it corresponds to: Clearly, the
machine has no geometric conception of what it’s doing—and needs none: For
the machine, to see means to correctly associate a few stock shapes with settings
of its grasper.