Proceedings of the ASME 2008 International Design Engineering Technical Conferences & Computers and
Information in Engineering Conference
IDETC/CIE 2008
August 3-6, 2008, Brooklyn, New York, USA
DETC2008-49389
A METHOD FOR CREATIVE BEHAVIORAL DESIGN BASED ON ANALOGY AND BLENDING
FROM NATURAL THINGS
Kazuya Tsujimoto
Department of Mechanical Engineering, Kobe
University, 1-1 Rokkodai-cho, Nada-ku, Kobe
6578501, Japan
Akira Tsumaya
Department of Mechanical Engineering, Kobe
University, 1-1 Rokkodai-cho, Nada-ku, Kobe
6578501, Japan
Amaresh Chakrabarti
Centre for Product Design and Manufacturing,
Indian Institute of Science, Bangalore 560012,
India
ABSTRACT
In the near future, robots and CG (computer graphics) will be
required to exhibit creative behaviors that reflect designers’
abstract images and emotions. However, there are no effective
methods to develop abstract images and emotions and support
designers in designing creative behaviors that reflect their
images and emotions. Analogy and blending are two methods
known to be very effective for designing creative behaviors.
The aim of this study is to propose a method for developing
designers’ abstract behavioral images and emotions and giving
shape to them by constructing a computer system that supports
a designer in the creation of the desired behavior. This method
focuses on deriving inspiration from the behavioral aspects of
natural phenomena rather than simply mimicking it. We have
proposed two new methods for developing abstract behavioral
images and emotions by which a designer can use analogies
from natural things such as animals and plants even when
there is a difference in the number of joints between the
natural object and the design target. The first method uses
visual behavioral images, the second uses rhythmic behavioral
images. We have demonstrated examples of designed
behaviors to verify the effectiveness of the proposed methods.
1
INTRODUCTION
In the near future, robots and CG (computer graphics) will be
an intrinsic part of human life. They will especially play an
important role in entertainment and art [1]. Hence, they will be
required to exhibit creative behaviors that reflect designers’
abstract images and emotions. Furthermore, it is expected that
inexperienced laymen as well as skilled workers will be able
Shinji Miura
Graduate School of Science and Technology,
Kobe University1-1 Rokkodai-cho, Nada-ku, Kobe
6578501, Japan
Yukari Nagai
School of Knowledge Science, Japan Advanced
Institute of Science and Technology, 1-1 Asahidai,
Nomi, Ishikawa 9231292, Japan
Toshiharu Taura
Department of Mechanical Engineering, Kobe
University, 1-1 Rokkodai-cho, Nada-ku, Kobe
6578501, Japan
to design creative behaviors, which will result in novel
products being introduced in some fields such as
entertainment and art. For example, you may desire to make
the ornaments in your room behave as you like and change
their behaviors every day. It is necessary for humanoid robots
and CG characters to reflect your abstract images and
emotions in their behaviors, so the behavioral pattern that you
can feel closer to them.
In order to design the behavioral pattern of an artificial being,
it is necessary to give a rough shape to abstract images and
emotions. This process involves generating ideas that reflect
the design concept [2]. Following this, the designer should
implement the design concretely and optimize it. For example,
when designing the behavior of a robot arm, a designer first
generates ideas or rough images of the type of movements
such as powerful, smooth, zigzag, etc. Following this, the
necessary calculations and drafting are done. The design of
robotic behavior has been researched extensively. However,
most of these researches include industrial robots that are able
to perform only limited tasks as their behaviors are
programmed depending on the programmer’s personal
experience. Furthermore, it is difficult to design complex
behaviors using conventional methods.
On the other hand, researches at the forefront of design
engineering have recognized that analogy and blending are
very effective in designing creative ideas [3]. Keeping this in
mind, Min An proposed an effective method of blending
behavioral criteria and then designing creative behaviors [4]
(Fig. 1). By using this method, a designer can select his
favorite behaviors and blend the criteria by which the
1
Copyright © 2008 by ASME
behaviors can be realized using the optimization method so
that the designer will reflect his tastes. However, the initial
behaviors were produced randomly, and hence the designed
behaviors were incongruous with the designer’s images and
emotions.
Obtaining behavioral criteria
randomly produced by GP (genetic
programming)
Producing behavioral patterns by GA
(genetic algorithm)
The designer selects any two desired
patterns
Combining behavioral criteria selected
by the designer by GP
Obtaining the criterion that produces
the desired behavior
Figure 1. The system proposed by Min An.
One of the methods used to design appropriate initial
behaviors is to use the motion capture data of natural objects
such as animals and plants. Nowadays image processing
technology has made remarkable progress [5]; hence, we can
easily extract motion data from movies of animals and plants.
However, there still exist differences in structure between the
design targets and the natural objects, such as the number of
their joints. Although we have developed a method to imitate
the behavior of a model robot by assuming suitable behavior
criteria when there is a difference in the number of joints
between an imitating robot and a model robot [6], it is still
difficult to reproduce the entire behavioral pattern.
In conclusion, there are no effective methods to develop
abstract behavioral images and emotions and support
designers in designing creative behaviors that reflect their
images and emotions regardless of the number of joints.
The aim of this study is to propose a method of developing
abstract behavioral images and emotions and giving them a
rough shape by developing an effective computer system that
supports a designer in the creation of the behaviors to be
designed. This method focuses on deriving inspiration from
the behavioral aspects of natural phenomena rather than
simply mimicking it. In order to give a rough shape to abstract
images and emotions, we use analogy and blending. In this
study, we have assumed that the design target has a linked
pole structure. We make this assumption based on the
following two points: first, it is possible to represent all natural
objects that have interesting motions as skeletal structures
using CG. Second, this linked-pole model is the most
fundamental model and can be extended to other models such
as a meatball, a polygon, etc.
In this study, we propose two new methods for developing
abstract behavioral images and emotions by which a designer
can use analogies from natural objects such as animals and
plants even in the case where there is a difference in the
number of joints between the natural object and the design
target. The first method involves the use of visual behavioral
images. The second one uses rhythmic behavioral images.
This paper is organized as follows. Section 2 provides an
overview of analogy and blending. Then, we point out
problems in using them in the design of creative behaviors.
Section 3 introduces Hermite interpolation. Section 4 proposes
a method to extract behavioral images independent of the
number of joints and discusses a way to develop visual
behavioral images. In section 5, wavelet analysis and the
genetic algorithm are introduced, and we propose a method to
extract rhythmic behavioral images and learn them. In section
6, examples of behaviors designed by these new methods are
demonstrated.
2
ADAPTATION OF THE CREATIVE DESIGN
PROCESS
Many studies have been conducted to analyze the
characteristics of the design thought process from the
viewpoint of creativity. As a result, it has been found that
analogical thinking and concept-synthesizing processes such
as blending are keys to creative thinking [3].
2.1 Analogical Reasoning and Metaphor:
Analogical reasoning and metaphor are known to play very
important roles in creative design (Gero and Maher 1991;
Goldschmidt 2001). For example, the “Swan chair,” a famous
example of design, was imaged using analogy [3].
The form of the chair resembles a swan: message that this
chair is soft and elegant like a swan to users. Analogical
reasoning and metaphor are methods of concept creation that
involve developing a new concept from an existing concept. In
practice, they are frequently used in the design process. In this
thought process, the design result and the base object share
some common features such as shape and function.
2.2 Concept Blending:
The main roles of analogical reasoning and metaphor are to
understand or to transfer known concepts, that is, it is analytic
rather than synthetic since its primitive process is the
extraction of some features from known concepts by analyzing
it [3]. From the viewpoint of mental cognition in the domain
of cognitive science, Finke et al. (1992) described conceptual
synthesis as an efficient means of developing creative insights
into new inventions and carried out experiments on creation
that involved mental products generated by imagery synthesis
[7]. It has been pointed out that the development of creative
thinking that is related to the transforming of concepts [8]
(Boden 2004) or conceptual spaces [9] (Gardenfors 2000) is
necessary to support human creativity.
On the other hand, in studies on cognitive linguistics,
Fauconnier (1994) focused on the construction process of
meaning in ordinary discourse. He analyzed how conceptual
integration creates mental products and how to deploy systems
of mapping and blending between mental spaces [10].
From the viewpoint of mental space theory, he showed that
conceptual integration operates on two input mental spaces to
yield a third space that is called “the blend.” The blended
space inherits partial structures from the input spaces and has
emergent structures of its own. Both mental products, imagery
and discourse, have shown emergent features and have
stimulated creativity. We can think of concept blending as a
process in which two basic concepts are blended at an abstract
level and a new concept that inherits some abstract features of
the two base concepts but concrete features of neither is
generated. For example, if one were to design something by
combining the concepts of a musical instrument and a dress,
the design result could be a wine glass that induces melody by
2
Copyright © 2008 by ASME
blending the concept of a party and that of strings [3].
Analogy
Abstraction
Base
Object
In this study, we assume that designers can extract motion data
from video data by using the latest image processing
techniques. First, they use IDEA-INSPIRE to search for
animals and plants that are congruous with their images and
emotions. Following this, they input the extracted motion data
into the methods developed in this study.
Target
Object
Blending
Object1
New
Object
Figure 3. IDEA-INSPIRE [Chakrabarti et al. 2004].
Object2
Figure 2. Analogy and Blending.
2.3 Analogy by Nature:
The importance of learning from nature has long been
recognized and attempts have been made by researchers like
Vogel (1988) and French (1998) to learn from nature with
product development in mind [11]. Systems in the natural
world and their functionality are seen as rich sources of
inspiration for the generation of ideas. Hence, we use natural
objects as base objects for analogy.
Chakrabarti developed a computational tool for supporting
designers in the generation of novel solutions for product
design problems by providing natural or artificially inspired
and analogical ideas; his aim was to initiate work in
systematic biomimetics and to develop an integrated approach
for systematic idea generation in product design using
inspirations from natural and artificial systems [11]. The name
of this system is IDEA-INSPIRE [Chakrabarti et al. 2004].
Designers can use the corpus of diverse motions that nature
provides as a potent source for solving product design
problems and as an inspiration for developing creative and
innovative novel products. The database of natural systems
has been developed with approximately 100 entries from the
plant and the animal domains [11]. The motions analyzed are
varied in both the media in which they occur. The information
collected contains details about the function, behavior, and
structure of these systems as perceived in the source materials
[11]. The types of information collected include written,
diagrammatic, pictorial, and video data about the systems and
their varied motions [11].
The SAPPhIRE model [11] is used to represent the entries in
the databases, and reasoning procedures are developed to help
browse and search for the entries that are analogically relevant
for solving a design problem.
2.4 Visual Image and Rhythmic Image:
In order to develop behavioral images and emotions and give
them a rough shape, it is necessary to define them. From the
viewpoint of physical phenomena, behaviors can be described
using physical quantities such as acceleration and velocity.
However, it is not possible to calculate physical quantities
while imitating the behavior of something. The features of
behavior should be more intuitive in nature. In this study, we
assume that behavioral images can be divided into two
groups–visual behavioral images and rhythmic behavioral
images. A visual behavioral image is literally what a behavior
looks like. This image can be represented using mimetic
words. For example, a snake’s behavior is termed as “zigzag”;
this word invokes visual images. A rhythmic behavioral image
represents a way of changing quantities such as the angles of
joints and velocities. This image can be represented using
onomatopoeic words. For example, we usually use the word
“hop” to describe a frog’s behavior; this word represents
rhythmic behavioral image. However, a part of visual
behavioral images and a part of rhythmic behavioral images
dependent on each other. One of the aims of this study is to
distinguish rhythmic behavioral images from visual behavioral
images.
In the next section, we describe the Hermite interpolation
theory, which plays an important role in this study.
Behavioral image
Visual
image
Rhythmic
image
Figure 4. Relation between behavioral images.
3
Copyright © 2008 by ASME
Hermite
polynomials
Pictures
1
+ Δt y s =Object
2
s=4
Frame
n −1
s=3
s =1
0 Object 2 x
Y
+ Δt
0
y
Blended
polynomials
xn − 1,1( s )
yn − 1,1( s )
xn − 1, 2( s )
yn − 1, 2( s ) w1 xn − 1, 2 + w2 Xn − 1, 2
Xn − 1.1( s )
Yn − 1.1( s )
Xn − 1, 2( s ) w1 yn − 1, 2 + w2Yn − 1, 2
Yn − 1, 2( s ) (0 ≤ w1 , w2 ≤ 1)
xn ,1( s )
yn ,1( s )
xn , 2( s )
yn , 2( s )
Xn ,1( s )
Xn , 2( s )
Yn ,1( s )
Yn , 2( s )
Blended shape
images
Transferred
pictures
Step 6
Step 7
X
n
+ Δt
Hermite
polynomials
Object 1
Frame
0
Y
Shape images
(interpolated at
defined points)
x
Object 2
X
0
Step 1
Step 2
Step 3
Step 4
w1 xn , 2 + w2 Xn , 2
w1 yn , 2 + w2Yn , 2
(0 ≤ w1, w2 ≤ 1)
Step 5
Figure 5. Abstracting visual behavioral images and blending.
n
(1)
k =0
gives the unique Hermite interpolating fundamental polynomial
for which [12]
P ( xi ) = f ( xi )
P′( xi ) = f ′( xi )
where
(2)
(i = 0, 1, 2,..., n)
xi denotes the points.
The polynomials hk(x) and gk(x) in the fundamental Hermite
interpolating polynomials are defined as follows [12]:
h k ( x) = {1 − 2 L′k ( x k )( x − x k )}{L k ( x)}2
g k ( x) = ( x − x k ){L k ( x)}2
where
x − xm
m =0
xk − xm
(4)
n
1
L′k ( x k ) = ∏
m =0 x k − x
m
n
P( x) = ∑ hk ( x) f ( x k ) + ∑ g k ( x) f ′( x k )
k =0
n
L k ( x) = ∏
3 HERMITE INTERPOLATION
Hermite interpolation is a method that allows us to consider
given derivatives at data points as well as the data points
themselves. The interpolation is represented as a polynomial
that has a degree less than or equal to the number of both data
points and their derivatives minus 1 [12].
(3)
(k = 0, 1, 2,..., n)
The fundamental polynomials of Lagrange interpolation are
defined by [12]
4 APPROACH 1–DEVELOPMENT OF VISUAL
BEHAVIORAL IMAGES
Since this study aims to extract and develop behavioral images
regardless of the number of joints, we capture motion as a
sequence of pictures and extract smooth images from each
picture using Hermite interpolation. A visual behavioral image
is a collection of these smooth shape images.
However, this collection cannot be directly applied to analogy
or blending since the number of terms in Hermite interpolation
polynomials differs according to the number of sample data
points. Therefore, the second set of Hermite interpolation
polynomials is obtained from the first interpolated data at the
defined points, so that the numbers of terms in these
polynomials are equal in all the samples. The following method
is used to extract visual behavioral images and transfer them to a
design target (Fig. 5).
Step 1) Motion data is divided to obtain a collection of sequence
frames (pictures).
Step 2) Hermite interpolation polynomials are obtained in each
frame (In fig. 5, s is the parameter representing the
number of points).
Step 3) Interpolation is carried out at defined points to acquire
smooth images.
4
Copyright © 2008 by ASME
Step 4) Hermite interpolation polynomials are obtained again in
each frame (the number of terms in Hermite
interpolation polynomials are the same).
Step 5) The Hermite interpolation polynomials are blended
(calculating weighted averages). The weights can be
changed by the designers.
Step 6) Interpolation is carried out at some points corresponding
to the design target
Step 7) The data of defined points is transferred to the
corresponding joints of the design target.
5 APPROACH 2–DEVELOPMENT OF RHYTHMIC
BEHAVIORAL IMAGE
In this step, we use wavelet analysis and the genetic algorithm
to transfer rhythmic behavioral images to design targets.
5.1 Wavelet Analysis:
The fundamental idea of wavelets is to analyze signals
according to scale. Wavelets are functions that satisfy certain
mathematical requirements and are used to represent data or
other functions. In wavelet analysis, the scale that one uses for
looking at data plays a special role. Wavelet algorithms process
data at different scales or resolutions. If we look at a signal with
a large “window”, we would only notice overall features [13].
However, if we look at a signal with a small “window,” we
would notice small discontinuities [13]. Using wavelet analysis,
we can observe the data in detail and analyze discontinuities in
the changes. The analysis is performed using contracted highfrequency version of the prototype wavelet. The wavelet
analysis procedure adopts a wavelet prototype function called an
“analyzing wavelet” or “mother wavelet [13].” In this study, we
use Daubechies wavelets [13] as wavelet prototype functions.
Because the original signal or function can be represented in
terms of a wavelet expansion (using coefficients in a linear
combination of the wavelet functions), data operations can be
performed using only the corresponding wavelet coefficients
[13]. Moreover, if the best wavelets that can be adapted to the
data are chosen or the coefficients below a threshold are
truncated, the amount of data can be reduced [13]. This “sparse
coding” makes wavelets an excellent tool in the field of data
compression.
5.2 Genetic Algorithm:
Genetic algorithms are inspired by Darwin’s theory of evolution.
We have developed a solution using the genetic algorithm.
In general, an algorithm starts with a set of solutions
(represented by chromosomes) called a population [14]. The
solutions from one population are taken and used to form a new
population [14]. This is done to ensure that the new population
will be better than the old one. Solutions are selected to form
new solutions (offspring) based on their fitness–the more
suitable they are, the greater the chance of reproduction [14].
This process is repeated until some condition (for example, the
number of populations or the improvement of the best solution)
is satisfied.
The genetic algorithm (GA) transforms a population of
individual behaviors into a new generation using the Darwinian
evolutionary principle and naturally occurring genetic
operations such as crossover and mutation [14]. Each individual
in the population represents a possible behavior. The individuals
in the population are usually fixed-length character strings
patterned after chromosome strings. GA attempts to find a very
good behavior or the best behavior by genetically breeding the
population of individuals.
The basic operations of the genetic algorithm are simple and
straight forward [14]:
Reproduction: The act of making a copy of a potential solution.
Crossover: The act of swapping gene values between two
potential solutions, simulating the “mating” of the
two solutions.
Mutation: The act of randomly altering the value of a gene in a
potential solution.
It is necessary to be able to evaluate how “good” a potential
solution is relative to the other potential solutions. The “fitness
function” is responsible for performing this evaluation and
returning a positive integer number, or “fitness value” that
reflects how optimal the solution is: the higher the number, the
better the solution [14].
The fitness values are then used in a process of natural selection
to choose which potential solutions will continue on to the next
generation and which will die out.
5.3 How to Obtain Rhythmic Behavioral Images:
First, we analyze the standardized changes in the angles of
points corresponding to a design target using Daubechies
wavelets and place the results into a database.
Next, in order to reflect rhythmic behavioral images in the
design targets, we use the GA. We need to decide on the
makeup of the chromosomes, which includes the number of
genes and what those genes represent. In this study, a
chromosome represents a time series of the angles of a joint. We
set up a gene to represent an angle in a frame (Fig. 6).
It seems reasonable in this study that the measure of fitness in
the GA would be a correlation between the rhythms of natural
objects and those of the design target. In GA, the system
analyzes the standardized changes in the angles of joints in a
design target using Daubechies wavelets and then evaluates the
correlation. Through this method, it is possible to design
behaviors that reflect rhythmic behavioral images even if the
behaviors are apparently not alike. The procedure is explained
as follows:
Step 1) The first three steps of approach 1 are performed.
Step 2) The interpolated data is converted to a time series and
the changes in the angles are calculated at each
interpolated point.
Step 3) The changes in the angles are standardized at each
interpolated point.
Step 4) Rhythmic features are analyzed using Daubechies
wavelets at each interpolated point (input signals are the
standardized changes in the angles and output signals are
the coefficiencies).
The first four steps are performed on two natural objects to be
blended.
Step 5) The population is analyzed using GA.
5
Copyright © 2008 by ASME
Step 6) The correlation between the rhythms of two natural
objects and those of a design target are evaluated and
certain chromosomes are selected for breeding (The
correlation is calculated at each point corresponding to a
design target).
Step 8) The procedure continues through a number of iterations
between Step 5 and Step 7 until a design target can learn
the rhythms.
Step 7) The chromosomes are evolved.
Natural object 1
Frame n
+ Δt
Natural object 2
Frame n + 1
+ Δt
Frame n
+ Δt
+ Δt
θ i ( n)
Frame n + 1
+ Δt
+ Δt
θ i ( n)
θ i (n + 1)
(i = 0, 1, 2,..., k ) i
Changes in the angles
:
θ i (n + 1)
(i = 0, 1, 2,..., k )
point number
corresponding to
a design target
Changes in the angles
Standardization
Standardization
Signal processing using Daubechies wavelets
Signal processing using Daubechies wavelets
Wavelet coefficiencies
Wavelet coefficiencies
Wavelet coefficiencies
Evaluate correlation
Evaluate correlation
Signal processing using Daubechies wavelets
Standardization
Changes in the angles
GA coding
(i = 0, 1, 2,..., k )
Chromosome of a point
θi (0) θi (1)
θi (n − 1) θi (n)
θi (2)
Frame
n－1
Frame Frame Frame
1
2
0
Frame
n
0° ≤ θ i ( j ) ≤ 360°
j : frame number
i : joint number
design target
Joint 1
Joint 0
(θ 0)
(θ 1)
Joint 2 (θ 2)
.
Figure 6. Extracting and obtaining rhythmic images by GA.
6
Copyright © 2008 by ASME
6
Extracting rhythms of natural objects
Acquiring
visual
behavioral
image of a
natural object
The
interpolated
data (visual
behavioral
images) is
converted to a
time series
The changes
in the angles
are calculated
at each
interpolated
point.
Storing the
wavelet
coefficiencies
in the
database
Analyzing
rhythms by
wavelet
analysis
EXAMPLE OF BEHAVIORAL DESIGN USING THE
PROPOSED METHODS
We designed “jumping and wavelike” behaviors for a robot arm
on CG by using the two proposed methods. The structure of the
robot arm is shown in Fig. 8. First, we designed using visual
behavioral images. Next, we designed using rhythmic
behavioral images. Designers can search for any natural systems
using IDEA-INSPIRE. In this example, we created animations
of natural objects instead of extracting motion data by image
processing. We used the movements of frogs and snakes as the
base objects for analogy in both visual and rhythmic operations.
y
Learning rhythms of natural objects by GA
The behavior
whose rhythm
is the most
similar to
those of the
natural
objects is
output
Behaviors
whose
rhythms are
similar to
those of the
natural
objects are
evolved
Evaluating
similarity
between the
natural
objects and
generated
behaviors
o
Joint 1
l1 (θ 1)
Joint 0
(θ 0)
x
0° ≤ θ i ≤ 360 °
l2
l3
Joint 3
(θ 2)
Joint 2
li = 100
(i = 0, 1, 2)
Figure 8. Structure of a design target.
Initializing
population
(Generating
many
behavioral
patterns)
The changes
in the angles
are calculated
at each joint
Analyzing
rhythms by
wavelet
analysis
calculation flow
reference
6.1 Example of approach 1–visual image:
We designed “jumping and wavelike” behaviors for a robot arm
using approach 1, which transfers visual behavioral images to a
design target.
Figure 9 shows some frames in sequence of the visual image of
a frog and some frames in sequence of the visual image of a
snake. Figure 10 shows some frames in sequence of blended
visual behavioral images. The weights for the blending are
shown in Table 1. Figure 11 shows some frames in sequence of
the designed behavior for the robot arm.
Figure 7. The procedure of learning rhythmic behavioral
images.
Table 1. The weights for blending.
Item
Weight 1
Weight 2
7
Value
0.5
0.5
Copyright © 2008 by ASME
Visual image of
frog
Visual image of
snake
Blended visual
image
+
+
Figure 10. Blended image.
+
+
Designed robot
arm behavior
+
+
+
+
+
+
+
+
Figure 11. Designed behavior of a robot arm.
Figure 9. Sequenced frames of the visual images of a
frog and a snake.
8
Copyright © 2008 by ASME
7 Example of Approach 2–rhythmic image:
We designed “jumping and wavelike” behaviors for a robot arm
using approach 2, which utilizes rhythmic behavioral images.
The extracted rhythmic behavioral images are shown in Fig. 12
and Fig. 13.
The system randomly generates angles between 0° and 90°at
each gene and calculates the changes in the angles. The
parameters of GA are shown in Table 2.
The designed behavior is shown in Fig. 14.
Designed robot
arm behavior
Rhythmic image of
frog
Figure 12. Rhythmic image of a frog.
Rhythmic image of
snake
Figure 13. Rhythmic image of a snake.
Table 2. GA parameters.
Item
Population
Number of Generations
Probability of Crossover
Probability of Mutation
Value
1000
200
0.6
0.3
Figure 14. A designed behavior obtained using rhythmic
images.
8 SUMMARY AND CONCLUSIONS
In this paper, we have proposed two new methods for
developing abstract behavioral images and emotions by which a
designer can use analogies from natural things such as animals
and plants even when there is a difference in the number of
joints between the natural object and the design target. The first
method uses visual behavioral images, the second uses rhythmic
behavioral images. We used Hermite interpolation to extract
behavioral images independent of the number of joints. We used
wavelet analysis and the GA to abstract and learn the rhythmic
images, respectively. We simulated the design of behaviors
using the proposed methods and designed creative behaviors
that reflect the images of the original objects. Creative behavior
of a robot arm should satisfy utility as well as novelty. It should
be explained how the behavior generated by the method satisfies
the utility requirement of creativity. However, in this study we
have only focused on the generating process of creative
behaviors. In the future, we intend to evaluate the creativity of
the behaviors designed by the two methods in this work from
the viewpoint of originality and utility. In addition, we also plan
to develop a searching method using rhythmic behavioral
images.
9
Copyright © 2008 by ASME
9 REFERENCES
[1] Billard, A., 2000, “Dreams and imitation in robot”,
Proceedings of the Socially Intelligent Agents-The Human
in the Loop, AAAI Fall Symposium, Menlo Park, CA, pp.
9-12.
[2] Pahl, G., and Wolfgang, B., Engineering design: a
systematic approach, Springer-Verlag, Berlin Ltd.
[3] Taura, T., Nagai, Y., 2005, “Primitives and Principles of
Synthetic Process for Creative Design”, Computational
and Cognitive Models of Creative Design Ⅵ, pp. 177-194.
[4] An, M., Taura, T., Shiose, T., 2007, “A study on Acquiring
Underlying Behavioral Criteria for Manipulator Motion by
Focusing on Learning Efficiency”, IEEE Transactions on
Systems, Man, and Cybernetics, Vol. 37 (4), pp. 445-455.
[5] Gonzalez, R., and Woods, R., 2007, Digital Image
Processing, Prentice Hall, US.
[6] An, M., Kagawa, K., and Taura, T., 2003, “A study on
acquiring model’s criterion focusing on learning
efficiency”, Proceedings of the Twelfth IASTED
International Conference on Applied Simulation and
Modeling.
[7] Finke, R., Ward, T., and Smith, S., 1992, Creative
Cognition: Theory, Research, and Applications,
Cambridge, the MIT Press, A Bradford Book, London.
[8] Boden, A., 2004, The Creative Mind Myths and
Mechanisms, Routledge.
[9] Gardenfors, P., 2000, Conceptual Space, the MIT Press, A
Bradford Book, London.
[10] Fauconnier, G., and Turner, M., 2002, The Way We Think
– Conceptual Blending and the Mind’s Hidden
Complexities, Basic Book, NY.
[11] Chakrabarti, A., Sarkar, P., Leelavathamma, B., and
Nataraju, B.S., 2005, “A functional representation for
aiding biomimetic and artificial inspiration of new ideas”,
Artificial Intelligence for Engineering Design, Analysis
and Manufacturing, Vol. 19 (2), pp. 113-132.
[12] Ralston, A., and Rabinowitz, P., 2001, A First Course in
Numerical Analysis, DoverPublications, UK.
[13] Percival, D., and Walden, A., 2000, Wavelet methods for
time series analysis, Cambridge University Press, UK.
[14] Miettinen, K., Neittaanmaki, P., Periaux, J., and Makela,
M., 1999, Evolutionary Algorithms in Engineering and
Computer Science: Recent Advances in Genetic Algorithms,
Evolution Strategies, Evolutionary Programming, Genetic
Programming and Industrial Applications, John Wiley &
Sons Ltd, UK.
10
Copyright © 2008 by ASME