June 25, 2012 12:33 PSP Book - 9in x 6in 05-Junichi-Takeno-c05
Vehicles 7 Through 10 with Associative Concept 81
Figure 5.6. Idea of an evolutionary robot.
connecting the intermediate unit of the a-d module representing
“red” and the intermediate unit of the b-d module representing
“rapidly approaching” is slightly lower than the other wires.
As a result, even when only the color information of “red” is
given, the robot behaves evasively because the “red” information
stimulates the b-d module representing “rapidly approaching” via
the wires of slightly lower resistance. Braitenberg interpreted
this as the “judgment of ‘red’ provoked the judgment of ‘rapidly
approaching’ by association.”
Concepts arising from association are called associative concepts.
It is also possible, conversely, that the judgment of “rapidly
approaching” invokes the judgment of “red” by association. By the
June 25, 2012 12:33 PSP Book - 9in x 6in 05-Junichi-Takeno-c05
82 Professor Valentino Braitenberg’s Vehicles
Figure 5.7. Vehicle 7 with associative concepts.
above statement, Braitenberg admits the distinct existence of the
concept and the representation of “red” and “rapidly approaching”
in the intermediate units, i.e., in the robot. He also explains that
wires M created a new concept of a “red and rapidly approaching”
object.
According to Braitenberg, if a robot that has learned an asso-
ciation between “green and rapid approaching” behaves evasively
by chance for a “blue” moving object (a color that the robot has
not yet learned), it is possible for the robot to abstract colors
(make a distinction between colored and colorless robots) and the
accompanying generalization.
Abstraction is a “mental action to extract and grasp certain
aspects and properties of things or representations” (Kohjien).
June 25, 2012 12:33 PSP Book - 9in x 6in 05-Junichi-Takeno-c05
Robot with a Sequential Concept 83
Generalization is the “forming of general concepts or laws by
shaking off specifics and saving what is common.”
Braitenberg introduced his vehicles 8 and 9 to describe
techniques to conceptualize a robot’s environment, such as space,
objects, motion, and form, within the robot using vision sensors.
The discussion of a robot’s learning of its environment is omitted
here because in Chapter 7 it is explained that artificial neural
networks are capable of such learning and, further, this topic is
irrelevant to the problem of the mind.
Braitenberg’s vehicle 10 was designed to study the problems
encountered when forming a new concept while keeping past
associations. He proposes to constantly circulate stimuli for the
associations that were memorized in the past.
5.5 Robot with a Sequential Concept
Braitenberg’s vehicle 11 features new wires (E) that he calls
“Ergotrix.” These wires E are used to memorize the order of the
occurrences of representations “a” and “b.” For example, one may
wish to have the robot remember to turn right when “an obstacle
pops up in front of the robot immediately after the red lamp
lights” (order of occurrence of two events). Wires E connect the
intermediate nodes “d” to one another as shown in Fig. 5.8. For
any two intermediate units n
1
and n
2
that are connected by wire
E, signals are transmitted from n
1
to n
2
when n
1
fires and then n
2
is ignited just a short time later. Importantly, the connections using
this wire are different from a simple artificial neural network. The
directionality of signal flow is assigned to each of these wires.
The robot shown in Fig. 5.8(1) comprises two sensor drive
modules, a-c and b-c. Like the previous robot, sensory sensors “a” are
color sensors and sensors “b” distance sensors. The difference from
the previous robot is how the intermediate units “d” are connected
to one another. New pathways are added to allow recurrence to the
intermediate units themselves.
An example of a circuit network with wires E is shown in Fig.
5.8(2). Here, d
1
is the intermediate unit for the color sensor and
d
2
for the distance sensor. In the circuit network, a single input
June 25, 2012 12:33 PSP Book - 9in x 6in 05-Junichi-Takeno-c05
84 Professor Valentino Braitenberg’s Vehicles
(1)
(3)
(2)
Figure 5.8. Vehicle 11 with a sequential concept.
circulates through representations d
1
and d
2
, just like a series
of variations of state occurring in an automaton. The robot is
eventually expected to reach a final state as shown in Fig. 5.8(3).
In Figure 5.8 (3), the robot memorizes the environmental changes
as a temporal change of representations: d
11
,d
22
,d
23
,andd
12
.
Specifically, the robot saw a red lamp, felt a wall on the right, then
felt another wall in front, and, lastly, saw a blue lamp.
An automaton is defined as “a general term for mathematical
models of various computing mechanisms including computers. A
finite number of internal states are considered, connecting inputs
and outputs. The internal state varies with inputs, or the output is a
function of the internal states” (Kohjien).
This robot is capable of memorizing events occurring in the
environment and the order of their occurrence. We then add wires M
to wires E in Fig. 5.8. This robot shows us that a new concept can be
June 25, 2012 12:33 PSP Book - 9in x 6in 05-Junichi-Takeno-c05
Vehicle 12 85
formed by chaining representations in an orderly manner. A concept
created by time-series events is called a sequential concept.
5.6 Vehicle 12
Vehicle 11 was improved, at one point, to become vehicle 12. A new
type of wire C was added to control the threshold function of all
artificial neurons that make up the logic unit of the robot.
Wires C run from the control box of the logic unit comprising the
brain of the robot to all neurons in the brain to modify the threshold
functions (Fig. 5.9).
“Modify” here means to alter the threshold values of all neurons
according to the number of currently firing artificial neurons in
the logic unit. If the number of currently firing artificial neurons
Figure 5.9. Threshold Control Circuit.
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.