Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Building cellular self-organizing system (CSO): a behavior regulation based approach
(USC Thesis Other)
Building cellular self-organizing system (CSO): a behavior regulation based approach
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
BUILDING CELLULAR SELF-ORGANIZING SYSTEM (CSO):
A BEHAVIOR REGULATION BASED APPROACH
by
Chang Chen
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERISITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(MECHANICAL ENGINEERING)
May 2012
Copyright 2012 Chang Chen
ii
Dedication
A Tribute to the Life and Memory of
My Mother
Ling Wang
iii
Acknowledgements
Firstly I would like to express my utmost gratitude and appreciation to my helpful and
knowledgeable Ph.D. advisor Dr. Yan Jin, whose continual guidance, support and
patience influenced and changed me during this wonderful journey in my life. Yan
always respected my ideas and tried to improve and research deeper with me in this
challenge topic, we did fight with each other in the research but the conflict resulted in
breakthrough for this study. He always encouraged me to strive for deeper, more
fundamental logic and the clarity for new concept. I really appreciate the days we argued
and refined the concepts and views on research matters. Other than the academia, I have
learned great deal of life from Yan, e.g. attitude to presence, future and other people, and
that knowledge has already changed my future not only as a Ph.D. but as a people in this
world.
I would like to acknowledge my committee members: Dr. Geoffrey R. Shiflett, Dr.
Henrick Flashner, Dr. Marijan Dravinski and Dr. Pin Wang. Their constructive comments
and criticisms allowed me to push this research into more challenging areas and exciting
directions. I would like to particularly thank Dr. Dravinski, I worked as teaching
assistance in his class for several years. He continuous supported me in the academia and
his striving for preciseness really helped me to perfect my work.
Next I would like to thank other USC Aerospace and Mechanical Engineering school
faculty who have had helped me on my academic career. I am thankful for the diligence
iv
and availability of Samantha Graves and Silvana Martinez-Vargas who enabled me to
perform and finish all my experiments as described in this work.
I also would like to acknowledge three of my Ph.D. friends for their kindness and great
help to support me during this journey, Yeh “Eric” Lin, Dr. Thanapat “Sumo”
Wanichanon, Harshavardhan “Harsha” Mylapilli, and Hung Fu “Aaron” Chang. I would
like to express my special thanks to Eric, who had been one of my best friends and
teachers when I felt most lost in life, he guided me and helped me clearing a lot of doubts
and provided me chances to be someone more than I am. I really appreciate Sumo,
Harsha, and Aaron as my best friends for supporting and showing me what a great man is.
This 5-year journey with those friends will become the pearl in my life and this
experience is something that I have and will continue to hold dear to me.
I would like to acknowledge my colleagues from the Impact Laboratory as well, who,
through discussions have helped me to refine this work: Dr George Zouien, Dr Majid
Yahyaei, Winston Chiang, Qianyu “Cherry” Liu, Jonathan Sauder, Newsha Khani,
Robert Fletcher, and Kai Ning. I would especially like to acknowledge the contributions
which James Humann has made to the thesis of this work. The selfless generosity of his
time and effort helped me finishing this thesis.
Finally I would like to thank my family for their unconditional love and support. I would
especially thank my girlfriend Yuan Zheng for supporting me during this journey and
helped me to locate my weakness and changed my personality.
v
To my dad Dr. Xiaoan Chen, I would never be the person I am without your showing me
the great love and the freedom in my life you guaranteed me to explore without any other
concern. Your kindness, humbleness, calm and confidence helped me and guided me to
be more patient and more hardworking. Thank you for your help and love not just during
the Ph.D. journey, but throughout my life as well.
Lastly to my Mon Ling Wang, my greatest supporter, my extraordinary mentor and
always my hero. You were a great woman for helping everyone around you in the entire
life. The way you taught me is not only by words, but by actions and the great
accomplishment you had achieved. It means great to me to have you as my mom for the
most difficult days we have experienced. Your strength, kindness, and the ultimate love
for your family and friends showed me what the meaning is to be someone great,
someone of value. The one I am today and will become in the future will always have
your footprint and reflection. You were there with me for every step I have taken in this
long journey not only as Ph.D. but as a human being. Although you have passed away
long ago, I know that you are always with me in spirit, and make me strong, confident
and truly kindness to others throughout this life. The true meaning of life is not only
taking care of oneself but be ready and willing to take care of others, and with you, I
accomplished this Ph.D. and you will live on in our hearts, forever.
vi
Table of Contents
Dedication .......................................................................................................................... ii
Acknowledgements .......................................................................................................... iii
List of Figures ................................................................................................................... ix
List of Tables ................................................................................................................... xii
Abstract ........................................................................................................................... xiii
Chapter 1: Introduction ................................................................................................. 1
1.1 Background and Motivation ................................................................................. 1
1.1.1 The rising of Complex System ................................................................ 1
1.1.2 Design for high adaptability vs. design of system state ........................... 5
1.1.3 Multi-agent system study and Biological system .................................... 7
1.2 Research Questions .............................................................................................. 8
1.3 Our approach ........................................................................................................ 9
1.3.1 Complex adaptive system and Biological system .................................... 9
1.3.2 Introduction of Behavior and Regulation idea into CSO design ........... 12
1.4 Academic Objectives .......................................................................................... 14
1.5 Thesis Organization ............................................................................................ 15
Chapter 2: Related Work ............................................................................................ 18
2.1 Introduction ........................................................................................................ 18
2.2 Engineering Design Theory and Methodology .................................................. 20
2.2.1 Systematic Design .................................................................................. 20
2.2.2 Axiomatic Design .................................................................................. 22
2.2.3 General Design....................................................................................... 23
2.2.4 TRIZ ....................................................................................................... 23
2.2.5 Design for Adaptability.......................................................................... 24
2.3 Self-Organizing Systems .................................................................................... 25
2.4 Fields of biology inspired product development ................................................ 26
2.4.1 Field 1: Bio-mimicking engineering ...................................................... 27
2.4.2 Field 2: Bio-utilizing engineering .......................................................... 28
2.4.3 Field 3: Bio-information inspired research and product development .. 29
2.4.4 Summary ................................................................................................ 40
2.5 Conclusion .......................................................................................................... 44
vii
Chapter 3: Basic Idea and Approach ......................................................................... 45
3.1 Introduction ........................................................................................................ 45
3.2 Motivation .......................................................................................................... 46
3.3 Design problem for a CSO system ..................................................................... 50
3.3.1 A Traditional Approach ......................................................................... 50
3.3.2 A Natural Approach ............................................................................... 54
3.3.3 Our Approach......................................................................................... 56
3.4 Field driven behavior regulation (FBR) for CSO Framework ........................... 58
3.4.1 Hypothesis.............................................................................................. 59
3.4.2 The Framework ...................................................................................... 60
3.5 Field driven behavior regulation (FBR) and CSO ............................................. 62
3.6 Behavior based approach to design DNA representation ................................... 64
3.7 Modeling and represent the current state (task field) ......................................... 66
3.8 FBR (behavior field) Model ............................................................................... 68
3.9 Conclusion .......................................................................................................... 69
Chapter 4: Behaviors, Fields and FBR modeling ...................................................... 71
4.1 Introduction ........................................................................................................ 71
4.2 BDA: from requirement to behaviors ................................................................. 71
4.3 Behavior Design Challenges .............................................................................. 73
4.4 Mechanical Cell (mCell) and Individual behaviors ........................................... 76
4.5 Definitions and Models ...................................................................................... 81
4.6 Behavior design in CSO ..................................................................................... 84
4.7 Model of behavior based of design DNA (B-dDNA) ........................................ 87
4.7.1 Design DNA ........................................................................................... 87
4.7.2 Behavior based design DNA (B-dDNA) Modeling ............................... 88
4.8 Field driven Behavior Regulation (FBR) ........................................................... 95
4.8.1 Initial Goal ............................................................................................. 95
4.8.2 Emergence of CSO systems Behaviors .................................................. 97
4.8.3 FBR Definition....................................................................................... 99
4.8.4 Current Approaches ............................................................................. 101
4.8.5 Field driven Behavior Regulation (FBR) ............................................. 106
4.9 Conclusion ........................................................................................................ 113
Chapter 5: Case Study and CSO Simulation ........................................................... 114
5.1 Introduction ...................................................................................................... 114
5.1.1 Objective .............................................................................................. 114
5.1.2 Multi-Agent system simulation and MASON ..................................... 115
5.1.3 Expected result ..................................................................................... 117
5.2 Simulation ........................................................................................................ 118
5.3 Case Study 1: Single Exploration Cell ............................................................. 120
5.3.1 Problem statement ................................................................................ 120
5.3.2 Task Field............................................................................................. 121
viii
5.3.3 Behavior regulation .............................................................................. 123
5.3.4 Simulation Result ................................................................................. 125
5.4 Case Study 2: CSO Mover System ................................................................... 135
5.4.1 Problem statement ................................................................................ 135
5.4.2 Task Field............................................................................................. 137
5.4.3 Behavior Regulation ............................................................................ 138
5.4.4 Simulation result .................................................................................. 142
5.5 Conclusion ........................................................................................................ 148
5.6 Limitations ....................................................................................................... 150
Chapter 6: Contributions and Future Direction ..................................................... 151
6.1 Contributions .................................................................................................... 151
6.2 Future directions ............................................................................................... 153
Bibliography .................................................................................................................. 156
ix
List of Figures
Figure 1.1: Complexity vs. the depth of D&C .................................................................... 3
Figure 1.2: Rigid design from traditional design appraoch ................................................ 4
Figure 1.3: Normal Control of System State ...................................................................... 6
Figure 1.4 Ignorance of possible system states ................................................................... 6
Figure 1.5: Comparison of the design of Biological system and CAS system .................. 10
Figure 2.1: The CSO research and related fields .............................................................. 18
Figure 2.2. The three major fields of Biology Inspired Engineering ................................ 27
Figure 2.3: Categories of the Bio-Information Inspired Product Development (BIPD) ... 31
Figure 2.4: Three major approaches for representing a CSO system ............................... 32
Figure 2.5: Comparison of complexity in different approaches towards CSO system
design ................................................................................................................................ 43
Figure 3.1: Conventional Design Process ......................................................................... 46
Figure 3.2: Natural design process demonstration ............................................................ 47
Figure 3.3: The "Feedback Loop" of Mechanical and Natural System ............................ 49
Figure 3.4 The tradition thinking framework of CAS system design ............................... 50
Figure 3.5: The natural approach of "designing" a system ............................................... 54
Figure 3.6: Our approach of designing a CSO system ...................................................... 57
Figure 3.7: Field driven behavior regulation for Cellular based Complex
Self-organizing System Design Framework ..................................................................... 61
Figure 3.8 The CSO behavior emergence in changing environment ................................ 65
Figure 3.9: The limitation of local self-organizing process .............................................. 67
x
Figure 3.10 The Network Motifs: Inner interactive control in DNA ................................ 69
Figure 4.1 Overall demonstration of Translation, Mapping, and Emergence .................. 73
Figure 4.2: Mechanical Cell Model vs. Biological Cell ................................................... 78
Figure 4.3: The model of Individual Behaviors ................................................................ 80
Figure 4.4 Four components of DNA representation ........................................................ 89
Figure 4.5 Example of Actions ......................................................................................... 90
Figure 4.6 The related entities of FBR ............................................................................ 100
Figure 4.7: Signaling cascade for photophobic response for Microbial rhodopsins ...... 104
Figure 4.8: Demonstration of a Protein Network in Biological Chemotaxis .................. 105
Figure 4.9: Example of Task Field ................................................................................. 107
Figure 4.10 Example of Behavior Field .......................................................................... 108
Figure 4.11: An Illustration of Field driven Behavior Regulation (FBR) in CSO
Systems ........................................................................................................................... 111
Figure 4.12: The summary of two fields in CSO system ................................................ 112
Figure 5.1: Task field for mCell m in single exploration case study .............................. 122
Figure 5.2: Simulation Results of a Single mCell exploring in a Random Obstacle
Field Simulation Results ................................................................................................. 126
Figure 5.3: Comparison of “Success Rate” of "Select the Best" (FBR
DM-B
) and
"Select from Top 40% randomly" (FBR
DM-G
) ................................................................ 128
Figure 5.4: Comparison of “Failure Rate” of "Select the Best" (FBR
DM-B
) and
"Select from Top 40% randomly" (FBR
DM-G
) due to collision to obstacle .................... 129
xi
Figure 5.5: Comparison of “Failure Rate” of "Select the Best" (FBR
DM-B
) and
"Select from Top 40% randomly" (FBR
DM-G
) due to reaching the limitation of steps ... 129
Figure 5.6: Comparison of “Number of Steps” of "Select the Best" (FBR
DM-B
) and
"Select from Top 40% randomly" (FBR
DM-G
) when success in task .............................. 131
Figure 5.7: Comparison of “Success Rate” and “Number of steps in successful runs”
of different FBR
DM-G
in the environment with 90 obstacles ........................................... 132
Figure 5.8: Comparison of “Success Rate” and “Number of steps in successful runs”
of different FBR
DM-G
in the environment with 120 obstacles ......................................... 133
Figure 5.9: Tasks Field for mCell m in CSO Mover case .............................................. 137
Figure 5.10: Demonstration for the bField of behavior 1 in ideal situation ................... 138
Figure 5.11: Demonstration for the bField of behavior 2 in ideal situation ................... 139
Figure 5.12: Demonstration for the bField of behavior 3 in ideal situation ................... 140
Figure 5.13: Simulation for design case 2, CSO Mover simulation results .................... 142
Figure 5.14: Illustration of the dynamic bField of the CSO Mover in the simulated
field of obstacles ............................................................................................................. 145
Figure 5.15: Resilience Test by Deactivating 4 of 12 mCells at Step 400 ...................... 147
Figure 6.1: Proposed future direction of this research .................................................... 153
xii
List of Tables
Table 2.1: The comparison of different approaches in CSO system design ..................... 41
Table 4.1: Example of Translate from Behavior Design to Serial Code .......................... 94
Table 4.2: The comparison of FBR and Multivariable Optimization ............................. 102
xiii
Abstract
Complexity of a system grows as more functionality is required by customers and more
unintended component interactions are added to the system by designers as they make
design decisions. The increasing level of both intended and unintended complexity of
systems has made it extremely difficult, if not impossible, for designers to ensure
reliability of, and instill adaptability into, their designed systems. As demands for
adaptive systems increase in areas such as space and ocean exploration and rescue and
military missions, how to guarantee reliability and increase flexibility and/or adaptability
of complex engineered systems is a major challenge. While research in biological
systems has advanced our understanding of how these systems have been designed and
developed and revealed their fundamental properties of adaptation through
morphogenesis, there has been little research in exploiting the biological design process
for the development of engineered flexible and/or adaptive systems. A new engineering
approach is needed that can overcome the limitations of conventional design methods by
applying the concepts and processes found in the development of biological systems.
In this dissertation, we develop a framework for understanding the limitations of the
conventional design process for designing complex adaptive systems. Based on the
previous work on mechanical cell (mCell) based system formation, we propose a novel
biology inspired system representation called Behavior-based design DNA (B-dDNA) for
the development and operation of our Cellular Self-organizing Systems, or CSO systems
for short. Based on the B-dDNA representation, a mechanism called Field driven
xiv
Behavioral Regulation (FBR) is proposed that implements and synthesizes system
Designing, Formation, Operation, and Adaptation processes. FBR of a CSO system is a
mathematical and selection model that is shared by all mCells and specifies cellular
behaviors corresponding to functional, system level, operational and adaptation
requirements. Our research results have demonstrated the feasibility and advantages of
the B-dDNA representation and FBR based mechanisms. Two case studies along with
more detailed computer simulations are provided to demonstrate the power of B-dDNA
and FBR for designing and developing complex engineered adaptive systems that possess
inherent capabilities to cope with increasing level of system complexities and to exhibit
high level flexibility/adaptability in response to both task and environmental changes that
occur in mission situations.
1
Chapter 1: Introduction
1.1 Background and Motivation
Engineering Design is often understood of a process of designing systems to meet the
needs or requirements which are preset by customers. In order to achieve these design
goals, multiple techniques have been developed to help engineers to generate a quicker,
better and more reliable design. The following sections will discuss several problems of
current design methodologies.
1.2 The Rise of Complex Systems
In the last 200 years, the growth of human desire and the manufacturing development
simultaneously led to the current society of concrete and steel. Divide and Conquer
(D&C), as the most effective design methodology, has been used widely in any
engineering field. No matter what kind of design methodology is proposed, the basic idea
is D&C. D&C starts from a general problem definition, goes through a decomposition
process to lower the design difficulty to smaller, understandable problems. Systematic
design suggests three aspects, energy, material and signal, to decompose the general
design problem. The efficiency of D&C is proved to be very high due to the experience
from the previous solutions to the smaller design problem, and the lower difficulty of
solving the decomposed problem, and D&C defines the way engineers think.
Manufacturing skills have been developed dramatically during the last 200 years, and the
artificial systems which humans build have also achieved a lot. More and more complex
2
systems have been developed purposefully or un-purposefully, such as space ships, space
stations, and traffic systems. A complex system is not an easy system with a single
purpose and limited behavior, but a system composed of interconnected parts that as a
whole exhibit one or more properties, and those properties are not easy to predict from
components’ properties (Jsslyn and Rocha, 2000). The difficulty in understanding the
system arises due to both the human desire and lack of knowledge. Some systems are so
complex that they are beyond the understanding of every engineer involved in building
them. For example, the reading and research of the black box from a crashed airplane
normally takes half of a year to find out which part or parts might have gone wrong and
caused the disaster. How to understand such complex systems’ design and how to
measure the difficulty to understand the system are two major problems.
The introduction of complexity in engineering design, which is done by system
engineering researchers, is prerequisite. Complexity can be understood by the chance of
failure to perform the pre-defined functions. As a result, the complex system sometimes
cannot perform as what was predicted, and when the system fails, it is also a big
challenge to find out which part or parts went wrong. In a system view, the complex
system is a system composed of interconnected parts that as a whole, will exhibit a
behavior property or behaviors among possible properties. In most cases, because the
complexity is the measurement of the unpredictability or uncertainty of the system
performing within the desired range, most design methodologies or theories are trying to
lower complexity as much as possible. Particularly, in Axiomatic Design (Suh, 1990), the
3
two axioms, independence axiom and the information axiom, are introduced to lower the
complexity of the system. System engineers also introduced two kind of complexity in
design, one is the inborn complexity, and the other is the acquired complexity. No matter
what you do to limit the complexity, the inborn complexity is not avoidable (Suh, 1990).
Traditional divide and conquer (D&C) approach to deal complex system may not be the
solution to the problem of complexity. The Figure 1.1 shows the relationship between
D&C and complexity. The deeper the D&C goes, the more complex the system will be.
Figure 1.1: Complexity vs. the depth of D&C
Complexity is very difficult to prevent, even if we gain all the knowledge of systems
engineering, as the growth of human requirements sometimes leads to systems with
multiple functions and multiple purposes. As Figure 1.2 indicates, from a traditional
approach, since the functions are distributed and assigned from a very early stage, when
the design constraints change or the system changes, it is almost impossible for one part
to adapt other parts’ functionality or behavior. Much effort has been spent to solve such
problems For example, modular design is used widely in electrical and electronic systems
4
for reuse and integration, and platform-based design (Baily, 2005) is proposed to
decompose the electrical system to platform design and platform use for the current
electronics and mechatronics. Those methods provide a way of solving the design of
complex systems in a more integrated way by choosing the right objects to use. Instead of
designing from parts, they proposed a way to use higher complex components such as
modules together with platforms. Although in this way, the difficulty to design and
understand a system is much lower, but the lack of knowledge of how to design the
interactions between those modular and platforms is still a problem to be solved.
Figure 1.2: Rigid design from traditional design appraoch
Complexity leads to low adaptability. The higher complexity that a system has means the
interaction and interconnection between components is more difficult to understand, in
other words, if one part goes wrong, it may result in a “butterfly effect” inside the system
which leads to failure of other parts. The adaptability is the ability for a system to change
itself to fit to occurring changes in the environment or system itself. A highly complex
5
system may result in a lowly adaptable system, which means any slight design error may
result in a redesign in a lot of interconnected parts. If it is possible to design a system
with qualified adaptability, the system may be the solution for future problems such as
exploiting other planets or real-time operation robots in battlefield since in those extreme
environments, adaptability is highly required. Also, a highly adaptable system may result
in less redesign, which may lower the cost of maintenance. The problem is, through what
process we can guarantee that a system can hold as much adaptability as we need.
1.3 Design for high adaptability vs. design of system state
Based on all the previous problems, if we can develop a design methodology for systems
with high adaptability, most of the problems will be minimized. An adaptive system can
modify itself to meet the changes of the environment and system itself. In this way, a
redesign from beginning is not required if new functional requirements or new
environmental impacts influence the system performance. But there is a problem with
current design; no matter how complex the system is, the designer always tries to push
the system to a “steady state” that the system is fully understandable and functional, and
these states are predefined by the designer. When the problem or exceptions happen, the
control method will push the system behavior from a previous state to another defined
system state, sometimes, it is applicable but for some of the current complex systems, the
next state is defined vague or the next state of the system is not yet fully predictable. In
those situations, the system will present complex behavior.
6
Figure 1.3: Normal Control of System State
Figure 1.3 shows a normal system design result. It divides a complex system state into
understandable and non-understandable. For those states which are understandable, the
designer will use the control method to ensure the current state is at one or some of the
predesigned state or it has the orientation towards that state. In this way, the system is
fully under control. However, when the system is in states which are not understandable,
the current control method may push the system to the understandable states, or it will
never reach any of those predefined states, and this prevents adaptation.
Figure 1.4 Ignorance of possible system states
7
Some of the designs use a way to solve the problem by shutting down the whole system
when an exception happens following by a full examination and repairing. As shown in
Figure 1.4, it is possible for those system states which are ignored during design to help
to solve the problems which are created by exceptions. The limitation of current design in
defining enough states always causes this problem, and it is one of the reasons why
complex systems are very difficult to understand.
1.4 Multi-agent system study and Biological system
Multi-agent system (MAS) is a system which arises out of the overall behavior from
multiple interacting intelligent agents. Such a system is often used to solve problems
which are very difficult for an individual agent to solve (Wooldridge, 2002). Most MAS
systems are used in software engineering, cooperation and coordination simulation,
organization and communication study, distributed problem solving, etc. and there is very
limited study in engineering design for mechatronics because for most current systems,
the intelligence is still centralized, and the system behavior depends on all components’
functions. On the other hand, by studying the biological systems, such as cell-tissue-
organ formation, the way of designing a complex system in nature involves distributing
the overall design and behaviors to all the agents involved in a normal biological
system’s components, such as cells through the encrypted series of DNA (Alon, 2005).
No current synthetic system is as complex, adaptive and evolutionary as biological
systems, although we can build very sophisticated engineered systems based on what we
have already learned from the nature. It is very obvious that in a natural way, almost
8
every cell is repeatable, replaceable and reproducible, and this promises that the loss of
part of the components of a biological system will not impact the overall system, and
even if some of those components go wrong, the overall system still can perform
acceptable performance. And there is research into how we can mimic the natural
systems in the way that all the components have similar properties as biological cells and
how we can generate a system which has global functions emerged from local agents.
The MAS system is similar in the way that the overall system has distributed action
components but there is still a gap between an MAS with a biological system in the way
they formed and the way they act. However, MAS provide us a chance to look into how
to design such complex systems, and it provides us a chance to simulate some problems
in lab to guide the later highly adaptive system design.
1.5 Research Questions
To summarize the above, the understanding of the complex system and the adaptability of
a system to environment change and requirement change are the two major challenges for
complex system design. In responding to these challenges, the following questions need
to be addressed:
1. Is there a better way of modeling the complex system design and operation
information than formation procedure / edge connection information in order to
provide a better understanding?
9
2. What kind of control mechanism can complex adaptive systems adopt to have the
adaptability?
3. How to fulfill those requirements in order to design complex adaptive systems?
1.6 Our approach
1.6.1 Complex adaptive system and Biological system
Complex Adaptive System (CAS) was defined and coined by J Holland and M Gell-Mann
from the Santa Fe Institute. For any adaptive system, it should be composed of
interconnecting or interdependent parts, which can respond to the change of natural
environment as well as itself. As characterized by J Holland, the CAS is a dynamic
network of a number of agents which holds similar ideas to MAS, every one of which is
doing a parallel decision making corresponding to other agents’ behavior and status. As a
result, the overall behavior of such system is emerged from a large amount of decision
making and parallel information processing. The normal property of such systems is the
high degree of the adaptive capacity when facing the perturbation caused by the
environment or itself. CAS also has the properties of adaptation, communication,
cooperation, specialization, and reproduction.
The compatibility of CAS and Biological system is studied, and it is claimed that it is
possible for a CAS system to perform global structure based functions by plugging in
design DNA as a global information source for the local agents to follow. This provides
the basic idea of how we can mimic the behavior of biological systems not only at the
10
high level of global behaviors such as jumping and running mechanisms, but a much
lower level of how the agents (cells) of the biological system can be used in design. In
this research, we introduce a new concept as Cellular Self-Organizing (CSO) system, and
lead the CAS system design into a much more basic level, instead of structure based
mechanism, we look into the very intuition of DNA, and Figure 1.5 shows the basic
starting point of this research.
Figure 1.5: Comparison of the design of Biological system and CAS system
For a normal biological system, when decomposing to the lowest level of functional unit,
cells use different protein productions to respond to the chemical, signal environment,
and those proteins will result in a global emergence of global function. For a CSO system,
it is possible to perform global structures by plugging in series of DNA-like instructions.
But if we need it to perform global functions in an unknown environment, the system
Chemicals and Signals
(Environment, System)
Protein Production
DNA
Biology
Process
+
Environment, System
Actions
dDNA ?
+
11
may need other control mechanisms besides the current design instruction. The following
questions will be addressed throughout this research for a mechanical complex system to
be as adaptable as a biological system, which is known as a natural-designed adaptive
complex system.
1. What is the difference in design for CSO systems and design of biological
systems?
Most complex biological systems have evolved for billions of years to ultimately
utilize the construction of protein to survive in a complicated environment like
earth. The design representation of the biological system is stored in serial coded
chemical nucleic acid well-known as DNA. Each of the cells holds a complete
DNA and performs the functions which are indicated by DNA. Also the DNA can
be mutated, mated as well as selected during the natural selection process. And
the function of DNA is to guide the production of protein. On the other hand, the
CSO system, which is composed of multiple agents which are known as
mechanical cells (mCells), the responses are not protein but the self-action. To
mimic the biological system, the current design DNA needs to be modified and
models are required.
12
2. How does CSO system achieve similar adaptability as biological systems?
First, we need similar functional entities to perform as DNA of biological cells,
which is stated in question 1. However, the complexity of DNA is still a current
problem for scientists yet to discover. We know that the DNA forms a very
complicate network to guide the production of protein under very restricted
situations. The interaction and interconnection within DNA itself are as complex
as the overall system. Also, the dilution and diffusion of the chemicals provide a
way of feedback from the natural environment to cells. Instead of only mimicking
the DNA design in biological systems and programming the limited intelligence
into mechanical agents, we can provide self-organizing rules for local agents to
provide similar functions as DNA network based on the solution of previous
question .
1.6.2 Introduction of Behavior and Regulation into CSO system design
There is a possible solution for the CSO study and design based on knowledge from the
understanding of the biological system. Behaviors are known as local decision making
for agents, and regulation here refers to the control mechanism for local agents to ensure
a global result. Regarding mimicking the operation of the biological system, research has
been extensively done on this design methodology.
13
Current research has been focused mostly on pattern formation for a multi-agent system.
Some studies have been done in lattice systems to determine if there is a solution locally
for each cell to follow and the global pattern is ensured. Nagpal provides the
computational origami language (OSL) to solve this problem by determining the
instructions for some of the lattice cells. Zouein in his research of DNA based design for
CAS system also provides an answer to global pattern formation by using the DNA idea
of assigning location information for each cell.
Computational Embryogeny was introduced by O.Yorgev in solving an automatic design
problem; they demonstrate a growth from a single cell to support a horizontal force which
meets certain criteria. Different from Zouein’s work, they use an indirect coding for the
growth. This growth instruction is for the single cell at the beginning like a zygote. In this
way, the system is more like a natural system in the way of developing from a single unit
to a complete system.
Other than simulation, several groups have already developed their cellular based robots.
Robots based on those cells can have a variety of performance and operation capabilities
based on different designs of those robots. Overall, the research of those real made
cellular robots focused on how to make the modular (cell) of the system, and what is the
interface design when two or more modular cells need to connect to each other. Superbot
(ISI, USC) also presented several formations or structures in simulation for different
purpose. Digital Hormone idea was in development during the process of designing the
14
control mechanism for Superbot. It presents an incomplete solution to how to control this
cellular based robot (W. Shen, 2002). The digital hormone is information density which
is generated by the information source; the cellular system will decide based on the
density of the information what next step is. Several particularly defined problems were
solved and simulated in the demonstration, and the self-decision making is addressed to
be the key for those systems’ success.
To sum up, although research has deeply studied cellular based system design, the lack of
understanding of the control, behavior, and emergence leaves much to be discovered.
Next, the article will present our research approach to look into the problems.
1.7 Academic Objectives
Any CSO system is difficult to understand, and there seems to be no clear connection
from local individual agent to global overall system effect. The biological system
provides an example of how nature designs complex systems. The basic elements of CSO
system are the parts (i.e., mCell in this research) of the system which are independent and
interacting with other parts or systems. And since the system is performing as the parts
are doing independent parallel actions and decision making, the key to resolve the
complexity problem is to understand the regulations of those actions and decisions, and
the biological systems provide a natural solution by cells and DNA. To answer those
problems we discussed in the previous section, the following approaches are taken for
this research work:
15
1. Develop models for the functional requirement, system state, environment and
local behaviors for a CSO system;
2. Introduce models mCell and the behavior oriented DNA (B-dDNA) to CSO
system design. The newly defined mCells and B-dDNA will provide a better
understanding, representation and support development of the CSO systems;
3. Link the model of mCell and B-dDNA with the concepts of functional
requirements, environment, behaviors and design;
4. Introduce a new Field driven Behavior Regulation (FBR) method to as self-
organizing strategies for designing CSO systems;
5. Develop a simulation platform for studying and testing different design problems
and different control parameters in FBR;
1.8 Thesis Organization
Chapter 2: Related Work
In this chapter, we present a literature review of the relevant work which has been done
in three specific fields: engineering design theory and methodology, self-organizing
systems, and biology-inspired engineer design. With the goal of introducing a new design
of the complex adaptive system, we discuss several significant results from the three
fields and suggest generating key concepts from those achievements. A conclusion of
these fields and a suggestion of how to utilize those key concepts are presented at the end
of this chapter.
16
Chapter 3: Basic Idea and Approach
This chapter starts with the motivation and a new design process which is based on the
new design requirement and the mimicking of natural design. Then it follows with
discussion of the overall design framework with three cornerstones. After the discussion
of design framework, several key concepts are introduced and some of the research
problem details are discussed for each of the key ideas.
Chapter 4: Behaviors, Fields and FBR modeling
This chapter discusses two key concepts in the design framework, behaviors and Field
driven behavior regulation (FBR). Based on the model of behavior, a new design DNA
(B-dDNA) is presented. After the Behavior based dDNA introduction, the chapter talks
about the control mechanism – FBR of designing a CSO system, and FBR will guide the
local decision making process and the emergence of the overall system performance.
Chapter 5: Case study and CSO system simulation
Two examples of CSO system design are discussed in this chapter. It starts from the
problem statement, and use the behavior oriented DNA design approach to decompose
the functional requirements and develop behaviors correspondingly. We will present the
FBR of those two case studies, followed by the simulation of how the system utilizes this
design framework to achieve adaptability.
17
Chapter 6: Contributions and Future Directions
This chapter discusses the major contributions of our research work and several proposed
directions are given for further study.
18
Chapter 2: Related Work
2.1 Introduction
Our work is built upon three engineering research areas; these three areas have been
studied extensively to get a clear and concrete result of solving the current engineering
problem. First, as a designer of the any mechanical system, our work is strongly related to
the design theories and methodologies. The field of complex adaptive system design also
plays a crucial part in our research. Last but not least, to solve the current engineering
problem, the field of biology and biology inspired engineering design also strongly
connected to our research. A graphical representation of our research field relating to
those three fields and the emergence may be seen in the Figure 2.1.
1. Engineering Design
Theory and
Methodology
2. Self-Organizing
Systems
3. Biology Inspired
Engineering
CSO
1
2
3
Figure 2.1: The CSO research and related fields
19
Engineering Design Theory and Methodology: This area has the purpose of helping the
designer in the design problems with the development of design theories and
methodologies; it can be applied to both system design and process design. Usually, a
design problem is an open-ended problem where there is no clear or best solution. The
complexity will result in unpredictable system behavior and a higher potential of system
failure, and it may be generated in the design or during the design process.
Self-Organizing Systems: This field is related to the development and investigation of a
system which can respond to the environment change or the system’s self change. The
process involved in such a response is often known as reconfiguration. Reconfiguration
in shape formation is a process of the composing elements of the system changing the
overall shape by rearranging the connections (edges) of their parts based on the changing
of the environment or the system itself. For a system, reconfiguration could be changing
the roles of different parts, changing the process, and changing anything that may
influence the overall performance of the system.
System Biology and Biology Inspired Engineering Design: System biology is the field
which studies the control of living creatures within the DNA network. Moreover, this
area is connected with the design and development of artificial systems which could
mimic the living beings’ behavior and performance based upon the principles and
methods from the study of nature and natural systems. The focus of this area in our work
is the principles which are developed from nature. The algorithms such as neural
20
networks, genetic programming and genetic algorithm will be presented as they pertain to
system design.
In the following sections, we present the results of all the three fields which are related to
our research in developing the behavior base of the multi-agent self-organizing system.
First, the related engineering design methodology is discussed. Next, we discuss some
methodologies developed for self-organizing systems and some self-organized complex
cellular based adaptive system design. Finally, we discuss some system biology research
and some biology inspired design principles.
2.2 Engineering Design Theory and Methodology
Engineering design theory and methodology is one of the crucial processes that all the
engineers should follow in order to get a better chance to have a good design. A design
theory or methodology may be used for the engineer to have a better performance or it
can be used as the guideline for a good enough product design. In industry, the more
proper the method they use to develop, the quicker and the better their product will be
designed. There are multiple design methods as discussed in the following section.
2.2.1 Systematic Design
Systematic Design (Pahl and Beitz 1996) was developed based on engineering practice
experience. Systematic Design provides an insight study from the early stage, where the
problem of the design needs to be defined clearly, to the final stage, where all the details
21
of the design are confirmed and the product is ready for prototype. The design process is
divided into four different components, which are the planning and clarifying the task
phase, the conceptual design phase, the embodiment design phase and finally detailed
design phase.
Phase 1: Planning and clarifying the task
At the very beginning of the design, to fulfill the market need and the company purpose,
the production planning must do search or survey for product ideas. The clarification of
the task requires information of the requirements and the existing constraints.
Phase 2: Conceptual Design
This step is the key step of how successful the design could be. And from the systematic
design, this step involves the problem abstraction, functional structure initialization,
working principle determination and other necessary steps.
Phase 3: Embodiment Design
This step is using the economic and the technical criteria to develop the layout
specifications.
22
Phase 4: Detailed Design
This is the step with all of the details of the design, such as dimensions, arrangements and
forms for the individual components.
2.2.2 Axiomatic Design
Axiomatic design is a system design methodology which is developed by Nam P. Suh
(1990, 2001).There are two key components. The first component is the four domains for
a design which consists of customer domain, functional domain, physical domain and the
process domain, and the design is the interplay of those four domains. Each of the
domains holds a hierarchical structure for decomposition from the top level and a
horizontal mapping from the other domain. And the other key of the whole design
process is a zig-zag mapping from top left in the customer domain to the bottom right of
the process variable domain. Moreover, Suh also developed two axioms which can aid
the zig-zag mapping and the two axioms can also be used to indicate the relative quality
of the design.
The independence axiom declares that design should “maintain the independence of the
functional requirements.” Because of the independence of the requirements in function
domain, the resulting design solutions in the physical domain are not related with each
other, so that later if the requirement changes or the design solution changes, it will not
influence other parameters in physical domain. This independence lowers the complexity
23
of the system. The information axiom declares that design should “minimize the
information content”, which suggests that the simpler it is, the better the design will be.
2.2.3 General Design
General design (Yoshikawa 1981; Tomiyama 1995) states that the design process is a
solution search out from design specifications. The three key elements of the general
design are entity and entity sets, attribute and attribute values and functions. Entity sets
are sets which contain entities which are real objects which existed in the past, exist now
or will exist in future. Attributes are properties which can be sensed by scientific means.
A function is a particular behavior which will happen under certain specific
circumstances, and the overall function is known as a latent function. Moreover, this
theory provides a description of the ideal world relation of the three key components, and
also the way to solve the real world problems under some real situations when the ideal
knowledge is not existent.
2.2.4 TRIZ
TRIZ or Teoriya Resheniya Izobretatelskikh Zadatch is another methodology, which has
been developed by Genrich Altshuller since 1946 (Altshuller and Shulyak 1998). The
translation of TRIZ is “the theory of solving inventor’s problem”. The idea of TRIZ is to
provide an objective framework for the designer to access the solutions from other fields
of science and technology, and the origin of TRIZ is based on reviewing tens of
thousands of patent applications during his work. Four key components of TRIZ are the
24
contradictions, ideality, functionality and the use of resources. Different from the
axiomatic design, the better design in TRIZ is the design that utilizes the maximum
potentials and uses most resources and less material. And the methods that TRIZ provides
are not exact solutions of certain problems but it provides ways of “innovative” thinking
of the design of the system when contradictory happens in design.
2.2.5 Design for Adaptability
Another special topic about design is how to design in early stage to determine the
functionality and performance features to make the system adaptable. The requirement of
a system with adaptability has been proposed by Gu (2007), the adaptive design or
product development is defined as the design that can easily be changed to serve different
requirements. Gu used three different measures, namely, extendibility of functions,
upgradeability of modules, and customizability of components; also, for all the design
candidates, in addition to the adaptability measures, there are other life-cycle aspects like
production costs (including part cost, assembly cost) and user operation-ability which can
also be taken into consideration. Some important components of how the adaptive system
should behave have been discussed in Gu’s work, and the measurement or the unit for
those components mostly is value of money, and this measurement serves as the criteria
when several design candidates are accessible and understood by the designer. The
design for adaptability method serves as an evaluation method for the design by using
cost as the major criteria.
25
2.3 Self-Organizing Systems
Another fields of research that is closely to this research is self-organizing systems. A
self-organizing system has its internal components organized through processes of
attraction or repulsion, and these processes are majorly internally generated to change the
complexity of the overall system. These systems are very common in biology, human
society, chemistry, etc. They can be categorized into self-repairing systems, self-
assembling systems and self-reproducing systems in a general way. Self-repairing system
research is majorly inspired by organic systems. As addressed by this research field,
adaptability is dependent on the self-reparability of a system. Most organic systems hold
such ability, e.g. self-healing in biological systems such as healing in small cuts or self-
healing in losing tails of certain lizards. The major research in self-repairing systems is
the reconfigurable systems. The research in self-reconfigurable system will be discussed
in the following section as the physical realization of biology inspired product
development. Self-assembling systems are majorly researched in organic chemistry
(Whitesides and Grybowski, 2002), and the research’s focus is to understand the structure
in biology. The examples found in bio-chemicals are numerous (Desiraju, 1989; Jones
and Chapman 1995; Kuman et al., 1995; Bongrand, 1999; Evans and Wennerstrom, 1999;
Neidle, 1999; Thomas, 1999; Grantcharova et al., 2001), and micro and nano-scale in
other engineering fields (Tien et al., 1998; Oliver et al., 2001; Vauthey et al., 2002). The
self-reproducing systems are systems which are “capable of constructing a detached,
functional copy of themselves” (Zykov et al. 2005), with abilities of making and/or
catalyzing copies of itself. One major issue in self-reproducing systems is scalability,
26
although research shows high potential in simulations (Chirikjian 2002; Bulter et al.,
2002b; Mytilinaios et al., 2004; Rubenstein et al., 2004) , normally, to scale such system
up or down using the same or different components often requires a redesign of the
overall system or system components. The self-organizing system provide a deep inside
research potential to future system design through a self-repairing, self-assembling and
self-reproducing way, and the inspiration from natural systems provides this research a
new perspective of looking into the biologically inspired product development.
2.4 Fields of biology-inspired product development
Much research has been done in this biology inspired product development field in order
to create systems, materials, and other useful artifacts. And most of such research can be
generally classified into three groups: bio-mimicking engineering, bio-utilizing
engineering and bio-information inspired engineering as shown in Figure 2.2.
27
Bio-mimicking
engineering
Bio-information
Inspired
engineering
Bio-utilizing
engineering
Biology Inspired
Engineering
Figure 2.2. The three major fields of Biology Inspired Engineering
2.4.1 Field 1: Bio-mimicking engineering
This is the major field which “studies models in nature and imitates or takes inspirations
from these designs and processes to solve human purpose” (Benyus, 1997). In this
research, we categorize all the study of the phenotype of natural system in this pile.
Robotic, fluid mechanics, and dynamics in the engineering field extensively studied the
principles and optimization from nature, such as the swimming mechanism of fish
(Triantafyllou, et al, 1995, 1998, Barrett, 1996, Michael, 2001, Anderson, et al, 1997,
Mason, et al, 1999, 2000, Yu et al, 2003, 2004, Hirata, et al 2000), the flying mechanism
28
of birds and insects (Deng et al, 2001, Sun et al, 2002, 2005, Dickinson et al. 1999,
Fearing, 2000, Spedding, 2003, Rosen et al, 2004), the mobility of cockroach (Watson et
al. 1997a 1997b; Bachman et al. 1997; Nelson et al. 1998; Full et al. 1991; Cham et al.
2000. 2002. 2004; Bailey et al. 2000; Clark et al. 2001), the biped walking mechanism
(Praff et al. 1997; Hirai 1997. 1998) and the swam snake movement (Chirikjian et al.
1991, 1992, 1994; Paap et al. 1996, 1999; Tsakiris et al. 2004, 2005). Other than the
mimicking of the mechanical movement, the utilization of the mechanism or natural
phenomenon provides us micro-physical level of examples for product development. Sto
AG mimic the self-cleaning characteristic of a lotus leaf with similar microstructure
(StoLotusan) coated on the surface so that the drop of water will roll off the surface and
carry dirt with it. These researches and engineering provide a deep phenotype level of
knowledge when dealing with similar effects as natural systems achieved. With more and
more discoveries in natural systems, this field will provide more and more natural
examples for engineers to consider and use in designing artifacts.
2.4.2 Field 2: Bio-utilizing engineering
This field is one of the major research trends in biomedical and chemical engineering, the
process control within biology cells may be utilized in protein product manufacturing and
other chemical products. Cellular level research provides another page of how natural
systems can help to create artifacts. As early as when humans harvested pearls from
living shelled mollusks, the study of mechanical or chemical product development from
natural systems became one of the most important product development fields of study.
29
After the genetic research had been established and studied, researchers started to modify
the genetic serial codes in simple cells that lead to different chemical production at the
cellular level, and this research opened a new phase of the bio-utilizing engineering from
finding the best conditions from natural systems to create better natural systems for
human purposes. To address those problems in understanding the system approach of
biological systems, system biology focuses on the complex interaction as an inter-
disciplinary study (Snoep, Westerhoff, 2005). The related disciplines, e.g. phenomics,
epigenomics, genomics, metabolomics, interactomics, etc. have been used to obtain,
integrate, analyze and store to study the complex biological system toward both the
understanding and engineering.
2.4.3 Field 3: Bio-information inspired research and product development
Bio-Information inspired research is a branch which is closer to cybernetics in science
and engineering. The cybernetics is a study of the nature at the system development level.
This research provides an inside level of how the information for the biological systems
is stored, exchanged and used, and focuses more on a process level for the emergence,
which provides opportunities for more adaptive design.
The Foundations:
The bio-information inspired research is a major field of Cellular Self-Organizing
systems, which includes the following specific features:
30
1) Cellular/Modular Multi-agent system (emerged from individual)
The organizational group of agents can be used to simulate the individual behaviors
in a group, and mimic the natural system on the structure level.
2) Information is distributed to all/some agents
a. Rules;
b. Procedures;
c. Interactions;
d. (Self-interest)
3) Two hypotheses:
The two major hypotheses of this research area are presented as follows:
Link Existence: There is a link between local agents and global effects through
mimicking the natural process, and there is design potential for local agents which
will leads to expected global performance;
Biology Information: This biology information inspired path helps for adaptability
and understanding of the natural process;
The Approaches:
The research in this field started as early as cybernetics, introduced in middle of 20th
century. Cybernetics is the study of regulatory systems with “closed loop” properties,
which means systems process and interact with themselves, and grow from themselves.
31
And any research in this context may be a sub-study of a cybernetics system and in a
particular orientation of bio-information inspired research as Figure 2.3 indicates.
Bio-Information Inspired
Product Development
Bio-Information Inspired
Product Development
Representation
Representation
Process/Control
Process/Control
Problem Solving
Problem Solving
Physical
Realization
Physical
Realization
Figure 2.3: Categories of the Bio-Information Inspired Product Development (BIPD)
Bio-information inspired product development has four major issues, including
representation, processing, problem solving/calculation and physical realization. Major
problems and approaches are addressed, such as the minimal representing information,
how the emergence can be guided, and what mathematics can be used to solve problems,
etc.
Approach 1: Representation
Representation is mostly focused on how to represent a cybernetic system and what is
the necessary/complete information for building a cybernetic system. Most of the studies
are done within 2D/3D simulation and the major job for those is to form a stochastic
located system with certain predefined shape requirements. The related fields are listed in
Figure 2.4.
32
Representation
Representation
Rule based
Rule based
Behavior based
Behavior based
Grow Sequence
Grow Sequence
Figure 2.4: Three major approaches for representing a CSO system
Cellular Automata (Ulam, Neumann, 1940s, Wolfram, et al) is a study field of
understanding the emergence process of a system by representing the system with simple
rules/initial conditions, and this research is focused on the dynamic “growth”. With
binary defined states of single agent, and a distributed rule of switch status through a
limited neighborhood condition law, a system emerges from a single cell or small group
of cells to a global pattern with different properties. Wolfram (1983) published his work
in studying some essence field of cellular automata, which he termed as elementary
cellular automata. These are one dimensional, and each cell can only observe the two
cells next to itself. The results of different rules vary dramatically; Wolfram categorized
the rules into 4 different sub-categories, homogenous, chaotic, loop and complex. In the
1970s, the game of life extended the study in a 2D field growth problem. This research
opened a new research field of the minimal information needed to present a complete
global pattern, and the convergence of a local to global problem.
33
One of the early stage studies from local to global linkage is the Turing’s
Morphogenesis model. The diffusion-reaction equation from natural systems provides
one way of representing a dynamic system with gradient information. The usage of the
gradients can be: leader selection, selective propagation, region selection, coordinate
system, patterning of activities, communication, and drive motion. Although this model is
not a bio-information inspired multi-agent study, it provides several key factors for the
studying of both natural systems and bio-information inspired system, such as a non-local
process like thermal diffusion (known as Brownian motion), other mechanical stresses
and motion. Those factors link the local information with the global pattern, and the
environment with the structure of the system. This model influenced the following
research in this field.
Origami Shape Language (OSL) was proposed by Nagpal (2002). It is a localized
sequence operation to form a global shape. OSL is a constructively describing language
instead of an assembling process. It uses the 7 Huzita-Hatori axioms and two types of
folding processes (known as mountain and valley). The axioms create a “globalized”
folding sequence. Each of the folds is based on previous ones with new points. Those
points are determined by a localized gradient map. For information at the local level,
origami exploits the biologically inspired computing methods such as gradients, tropisms,
seepthru and induction. This scale-free language can transform 2D shapes and also some
3D shapes automatically, and it provides a deep research of how the local andglobal are
linked in pattern formation. Daniel Yamins, in his PhD dissertation thesis, provided a
34
simple condition that is both necessary and sufficient for a global to local pattern
formation problem. As a start point, Yamins’s condition is that if the global pattern has a
locally checkable and stoppable state, the global pattern can be solved locally.
Instead of Design a structure, Antonsson (2010) proposed a building sequence
language/approach to represent structure to a serial growth code. Antonsson, et al,
developed a system to mimic the growth of natural beings and apply the growth
procedure to the real world engineering problem by autonomous solving simulation.
Embryogenesis is a biological concept that stands for the process by which the embryo is
formed and develops. In Computational Embryogeny, it is another representation of the
final design result, or predefined shape by using the growing rules or instruction sets.
Because of the difficulty in telling the design result from growing rules or sets, the
computational embryogeny provides the possibility that the cellular system will result in
some designs which are not expected. As defined by Antonsson, there are three different
types of embryogenies, which are external, explicit and implicit embryogeny. External
embryogeny provides the cells with global patterns or sub-patterns, and the cells, based
on those patterns, grow the shape; in explicit embryogeny, direct growth rules cause the
cell to follow in the formation process; in implicit embryogeny, the cells follow certain
rules in growth, similar to cellular automata in that the cell can sense the nearby situation
and decide what is the next step. Or Yogev achieved the use of explicit computational
embryogeny to grow a shape in 3D which can carry a wind load at the top of the shape.
35
Approach 2: Control/Operation
The second approach is focusing on how to understand and control the existing self-
organizing system, and these are the approaches which try to identify how to operate a
specifically designed structure or system.
For H Lipson’s (2000, 2005, 2007) self-reconfigurable robot, a partitioning, analyzing
and optimizing agent based self-modeling process is proposed for multi-agent based
structure. The structures or systems are more or less similar to neural network with
neurons and links. Lipson has addressed the design problem of the adaptive robots or the
evolutionary robots as a synthesis problem. He used neural network to solve some control
problems of different pre-designed robots, and he experimented with a control problem of
forcing a center of mass to be at a predefined height with one leg malfunctioning of a
four leg robot. He proposed an algorithm to automatically design moving robots from
lower level building blocks. The design space consisted of bars and linear actuators for
the morphology and neurons for the control. Because of the design space, he used a tree
representation for the design. Based on the tree representation of the design, which is
good for structure, he used a robotically-reconfigurable structure for further research of
how the system should respond to self-change, how to optimize the design from an
algorithm view.
36
Another research approach is the information communication based approach. These
researches focus on how and what information is needed and how the information is
distributed in the system or in the environment. There are two researches in this field we
would like to mention, because they are closely related to CSO design through a swan
intelligence way.
Ant Colony Optimization Algorithm (ACO) is a self-organizing emergence process for
finding a routine solution in an “unknown” field (Grassé, 1959; Deneubourg et al. 1983;
Dorigo et al. 1991, 1997, 2004; Bianchi et al. 2002); the information is left with explorers
and strengthened by the quantity,with the shortest routines being strengthened and longer
routines “evaporating”, the solution will gradually emerge, instead of being designed.
ACO also uses the concept from Turing’s Morphogenesis model in the simulation, and it
provides a preliminary emergence solution for the design problem in fields with limited
knowledge.
Digital Hormones Model (DHM) is the model generated from the biology hormone
distribution for a gradient field as morphogenesis. W. Shen et al. (2002, 2004) developed
a method called Digital Hormone Model, which is inspired from the biological discovery
of Hormone, a key factor that leads the cell from local distribution to a global pattern.
Each cell is performing due to several crucial factors, the sensor information, the local
variables, the connection information and the information it received (Hormone
information), and each of the action/behavior will be calculated by a probability function
37
using the predefined function. The DHM performs as the information distribution which
is generated by each cell, or it can be known as the propagation rate of the information
flow throughout the cell colony. Shen also applied this DHM method to a swan robot
called CONRO to simulate the movement and developed several local rules for the cell to
follow when different situation happened, and later development of Superbot also used
the idea of how the information is distributed in DHM. Superbot, similar to other cellular
systems, achieved reconfigurability due to its identically designed local single cell bot.
Approach 3: Calculation/Problem Solving (Evolutionary Computation)
This approach is focusing on the management of complexity in nature and applies some
key processes to engineering system problem solving.
The evolution design arises because of the introduction of the way that natural evolution
adapts gene sequences for computer algorithm to adapt and search for an
optimal/acceptable solution (Lee et al. 2001). There are two main types of the
evolutionary coding system, which are genetic algorithms (GA) and genetic
programming (GP), the key difference is that GA uses a serial binary code to represent
the solution while GP uses a hierarchical computer programming to search for the
solution.
Genetic Algorithms (GA) were created by John. H. Holland (1975). The purpose of GA
is to optimize or improve the solution on a global system scale without a full
38
understanding of the system. There are three major components in a genetic algorithm,
serial coding/ representation of the problem, fitness function and genetic operators. The
genetic operators can be selection, mutation and crossover. Selection is used to “keep”
the proposed solution with better “potential” or more “close” to the optimized result, and
“kill” those with low potential; Mutation is used to create some arbitrary changes in the
overall population of proposed solutions with relatively low possibility. Crossover is use
for the proposed solutions to reproduce with traits of each parent scheme and transfer
information to the offspring solutions.
Genetic Programming (GP) utilizes the similar idea as GA but the operant is different
than a real string sets but a tree structure which can be of any length. GP was initially
created by Cramer in 1985. There are three other important components which are not in
the GA. The first two components are terminal and function sets. Terminals are the nodes
in a genetic programming tree. And the third component is Automatically Defined
Function (ADF). The ADF is the region which great alternation happens during the
execution of GP.It evolves along with the calling program.
Both genetic algorithm and genetic programming share similar properties because
theymimic DNA and evolutionary processes. The normal mode of this computation is by
creating a random population at the very beginning and using a set of the evolution
techniques (selection, mutation and crossover) to grow the population with potential
39
correct solutions, and repress the population with wrong solutions. After each generation,
the solution population will get closer to the optimal result.
Neural network computing is a model of nonlinear complex relations between input and
output (statistical). The artificial neural network (ANN) is a mathematical model using
the connectionist to computation, most ANN are adaptive systems that can change their
inner interconnection based on the external or internal information flows. The inspiration
of ANN is to study the nervous system with neurons, axons, dendrites and synapses. The
ANN mimic the interconnection of biological neural networks by using simplified
artificial neurons and create a network with different weights on different connections.
Throughout a continuous learning phase, the system will reach an adaptive result in
connection for certain information flow. This ANN provides a way of solving non-linear
calculation problems, mostly in statistics and finding patterns/relations within data.
Approach 4: Physical realization
Due to the limitation of the conventional robot design, the modular based self-
reconfigurable robotic systems have the promise of high versatility and high robustness,
but they also address several new problems, such as motion planning, design, and control
of a system with various morphologies (M. Yim, W.Shen, al 2007). The motivation of the
modular based robotic system is to be potentially more adaptive than the conventional
systems, and the replacement of the modular parts is much easier since the modules are
40
identical. Moreover, the cost of building a modular system is much less than conventional
specific system design, because making many copies of one or few modules is much less
in mass production. Several real modular robotic systems have been designed and
operated in real fields. Superbot is designed and developed by Shen et al., and its
modules have 3 DOF, and can be connected with each other sharing power by one of six
identical dock connectors. The operation of Superbot is done by a real-time operation
system and the digital hormone inspired control (Shen et al. 2006). Molecubes are
developed by Zykov et al from Cornell University. The Molecubes have one DOF and
physical reproduction of four module robot was demonstrated (2005). The Miche system
was developed by Rus et al. from MIT. The Miche is a lattice system which can form
arbitrary shapes by connecting and communicating with its immediate neighbors.
2.4.4 Summary
Biology inspired product development rises in order to deal with design complexity.
Because of the similarity of the functional requirements and environment, to learn from
the natural solutions is a shortcut for some existing design problems. Moreover, the field
of biology information inspired research is developed to mimic the evolutionary or
development process of natural systems to get the advantage of building/construction
through the process instead of an assembling process from predefined engineering parts.
These approaches have the following new aspects:
41
1. The design complexity is transferred from overall system performance to the
system constructive process;
2. Natural process of evolution becomes possible for simulated system design;
3. Design information can be processed in local agent level;
Because of these differences, these approaches provide the possibility of changing the
system through the constructive process and it can lead to the adaptive system design. We
proposed the following table to compare different approaches in different design aspects.
dD CA GA ACO CE OR HL
Serve Human Purpose Y N Y L Y L L
Human Designer Involved Y(L) L Y L L L Y
Pure distributed intelligence/behaviors Y Y N Y N N N
Constructive Y Y N N Y Y L
Environmental Influence Y N N Y N N N
Failed Redesign N ? Y N Y Y N
Optimization N N Y L Y N Y
Knowledge Level(Global/Local) L L G L G G G
Cellular Differential Planner (G/L/N) L N -- N N G&L G
Table 2.1: The comparison of different approaches in CSO system design
42
Table 2.1 shows the difference between different approaches. Note that OSL, ACO can
only serve limited human purpose, such as pattern formation or optimization for
particular problem. Another important difference is that if the existing system fails or
undergoes requirement changes, almost all the approaches except ACO to need a
redesign or reconstruct, and this is one of the arguments against multi-agent approach in
design because the redundancy of agents did not provide enough system adaptability but
increased the complexity and built difficulty. Another important aspect is the cellular
level of differentiation. A pure homogeneous system with identical agents may lead to a
useless system, but an involved global controller for behavioral differentiation in OSL
and other adaptive systems may be critical to define the adaptation of a system.
43
Human
Society
High
Local Intelligence Complexity
(Reasoning/Rationality)
Global Regulation Complexity
Low
Low
High
Biological
System
(mamals/
insects)
Simple
Mechanical
System
Complex
Mechanical
(Spacecraft,
etc)
CSO System
DHM
Computationa
l Origami
Computationa
l Embryogene
Ant Colony
Optimization
Swam
Intelligence
Human
Society
High
Functional Complexity
Environmental Complexity
Low
Biological
System
(mamals/
insects)
Simple
Mechanical
System
Complex
Mechanical
(Spacecraft,
etc)
CSO System
DHM
Computationa
l Origami
Computationa
l Embryogene
Ant Colony
Optimization
Swam
Intelligence
High
Low
CSO System Potential
Figure 2.5: Comparison of complexity in different approaches towards CSO system design
44
Figure 2.5a and Figure 2.5b compare the existing research approaches towards BIPD.
Please note that the complexity of the system is the number of different system functional
states, and these two figures show only the abstract/related position, not the exact
position because of the difficulty in determining different design results through different
approaches. According to Ashby’s law of requisite variety, the system needs to have
complexity similar to the working environment in order to adapt. On the other hand, if
the system needs to adapt to new tasks or new functional requirements, the corresponding
complexity would require a new way of representation and embedment in the system
design. CSO systems in this research provide some advantages in the system states. Our
previous work on CSO generated a structure and function based design DNA concept and
developed system formation mechanisms (Zouein, 2008, Jin et al, 2010). This research
extended the previous research to a new design information as behavior with a dynamic
and probabilistic localized control mechanism which increases the system robustness and
resilience.
2.5 Conclusion
In this chapter, we went through the three major fields, engineering design theory and
methodology, self-organizing system design, and biology inspired engineering design.
The biology inspired engineering design and the study of self-organizing systems provide
the research base and the adapting of the idea from those two fields to the mechanical
design is the key point of this research. In the next chapter, we will discuss the new
approach to solve the problem of how to accomplish high adaptive capacity.
45
Chapter 3: Basic Idea and Approach
3.1 Introduction
Engineering Design is often understood as a process of designing systems to meet the
needs or requirements which are preset by customers. Normally a design is judged by the
functions, performance, cost, reliability and other properties which relate to specific tasks
or goals. Over the past centuries, the conventional design process has normally
progressed through requirements, functions, parts, assembling, and system behavior
process. This has been proven to be a very efficient process, and in this way, every part’s
functions and behaviors are determined by the designer and the physical structure is also
predefined by the designer. However, these designs with highly constrains from the parts
function and physical assembling structure can not prevent the system complexity from
growing and may restrain the system from flexibility and adaptability. While natural
systems with different design blue print and decentralized grow, control and adapt
method provide solutions to the earth environment. Some of the fundamental differences
in those two different processes will be introduced and studied in this chapter, and a new
framework for the approach of complex system design will be presented.
46
3.2 Motivation
The conventional design process often includes the steps which are shown in Figure 3.1.
Requirement Function
Parts/
Components
Assembling System
Complex
Mapping
Mapping
Simplify
Simple
Complex
Figure 3.1: Conventional Design Process
Figure 3.1 shows that conventional design expects simple and easy problems and can be
achieved by parts or components with limited understandable complexity. As a result,
when dealing with functions or physical parts/components, it often involves a hierarchical
structure. Designers achieve the simplified and decomposed functions with different
individual parts; then they create the overall design by placing the parts together
following physical rules or system requirements. The expectation that each part will work
as predicted in order to achieve certain global system behavior patterns often fails when
the system grows complex enough, because a simplified individual may not perform as
expected when interactions, interrelations, and interdependencies are overlooked during
the process. A redesign usually is the only remedy because other physical rules and
requirements need to be taken into consideration. The requirement of understanding all
the aspects of future design may go beyond the capacity of any recent design method or
theories since the complexity in information is out of bounds of any knowledge base.
47
However, nature solves the design problem by another approach, which is a complex-
complex-complex approach. The Figure 3.2 demonstrates the natural process in a general
way.
Survive
(Requirement)
Function
Cells’ Behaviors
System
Evolution
Reproduction
Selection
Complex Complex
Complex
Emergence
Figure 3.2: Natural design process demonstration
One of the differences from conventional mechanical design is that nature does not try to
understand all the complexity at the system level. The requirement for natural systems is
known as surviving in the environment, and the functions and behaviors of processing of
materials, signals and energies are evolved simultaneously. Systems with greater ability
to survive are reproduced, and the survivors are selected to be able to reproduce
themselves in the design circuit again. In the center of the picture, the behavior of cells
(often known as the production of different proteins, or other signals) is the only
“controllable” design parameter; even the growing of the individual cell is achieved and
controlled by the production of proteins. Nature selected this process to build the world
of millions of creatures without trying to understand the inside complexity of all their
components, systems, or processes, and the system’s global behavior is achieved by
48
emergence from individuals, rather than carefully planned process. The boundary of
“Design” by nature and “Self-Organize” by natural system or individual cells is blurred
and the understanding of the survival requirement is embedded into the process. The
capability of such processes leads us to introduce and study how this process of leading
individual behavior to global functionality can be mimicked in mechanical design, and if
it is possible, new machines with great adaptability/flexibility may be developed, and the
ultimate goal of “self-adapt” or “self-redesign” may also be accomplished.
Another success of the natural design is that the control mechanism is already embedded
in the design itself. Contrast this to a human design process. Normally we design the
physical components/parts for a limited purpose, and we design the control method to
guarantee that all of the components work in a certain range of their capability after
assembling to a system. Sometimes, the interrelations between different parts’ behaviors
and capabilities are very difficult to determine. Some unnecessary control may limit
system performance, and some lack of control may result in a design exception. For
example, the compatibility problem of computer components in the early stage of the
design of PC resulted in standardization in computer hardware design, and the same
situation happened to the software design for modulation of the components. New
interfaces are designed with different speeds and different functions, and, every time, the
previous design had to be modified or redesigned and rebuilt.
49
System
Behavior
Performance
Control
Mechanism
Observers Actuator
Normal Mechanical System
Control
Natural System
Behavior
Performance
Chemicals & Signals
Natural System
Figure 3.3: The "Feedback Loop" of Mechanical and Natural System
As you can see from Figure 3.3, the natural system has a different mechanism. Instead of
a separate control design from physical components/structure design, the normal natural
system has an “embedded” control mechanism (studied by system biologists) which
exists in a DNA transcription network, chemotaxis signal transduction network, and other
natural “circuits”. Some of those controls are now known as network motifs in natural
gene circuit study. Besides the selection of different chemicals in natural systems, the
major difference is that the evolution and upgrading of those mechanisms are
simultaneous with the evolution of the system itself. The growth, differentiation,
migration, functioning, and dying are governed by this mechanism and physics. This
natural solution to adaptation gives us the possibility that to find out how the natural
system works may provide a new way to treat the growth in complexity of current design.
Moreover, a further understanding of the relation between a single cell and the whole
natural system may provide a way to understand how the modeling and understanding
from the local may influence or determine the behavior of a global emergent system.
50
To summarize, it is the elegance of nature’s solution to complexity in both design and
control of the emergence from local to global that encourages us to study a cellular based
mechanical system and how this kind of system can achieve adaptability and flexibility.
3.3 Design problem for a CSO system
A thinking framework of designing a CSO system will be presented and discussed in this
section.
3.3.1 A Traditional Approach
As discussed in the related work from the previous section, a great effort has been done
to identify the properties and the thinking framework of designing cellular based systems.
From those related works, Figure 3.4 shows a thinking process of designing such system.
Global Functional
Req
Local Functional
Req
Local Behavior
Environment
Global Behavior
Global Effect
+
① ②
③
④
(Global Form)
Figure 3.4 The tradition thinking framework of CAS system design
A traditional way of designing a local-to-global system is to determine an optimized
solution for local behaviors. Under the hypothesis that the emergent global behavior is
guaranteed to converge, an optimal or at least an acceptable set of local behaviors can be
51
designed for certain purposes. Normally, because most design is based on the form or the
physical structure of the overall system, a design for a complex system is to determine the
connection/edge information of all components which will result in a global pattern
formation. In this section, a general discussion of those four steps is presented, and a
detailed picture of those steps is discussed in the following chapter.
1. From global functional requirements to local functional requirements/behaviors
Although whether there exists a link between the global functional requirements to local
functional requirements/behaviors is still a question, some research has already presented
several possible solutions for different specific problems. As discussed before, Nagpal
and Yamins present a thinking framework and also an algorithm for some 2D or 3D
shape formation. Each cell has only a local idea of the overall system, but a distributed
intelligence is calculated by a centralized intelligence and embedded to different cells in
this system. Different from Nagpal, Antonsson solved a growth problem from a single
cell instead of a distribution problem for multiple existing cells. Those two approaches
can be viewed as algorithms for a multi-agent constrained optimization; the global
functional requirements are translated into a cost function which needs to be optimized.
Moreover, the functional requirement for those systems is understood by the system’s
intelligence. The limitations arise when different problems or new functional
requirements are needed, as the system may result in another totally different design, and
a system rebuild is sometimes required for these cases.
52
Another way of studying this phenomenon is a bottom-up approach like the cellular
automata; test each rule to see whether a global behavior will be generated by simple
local rules.
And from the related research, an advantage of those designs for cellular systems is that
each of the cells has its own intelligence to deal with the current situation, and this
provides each cell a chance to choose its own behavior, but to what extent the overall
system will benefit from this change to traditional design is yet to be discovered.
2. The operating environment influence on an existing system
There are multiple cases for an environmental influence in design.
Case One, it is ignored. For example, a design in the summer possibly can work in a
much colder weather in winter even if the designer does not take this change of
environment into consideration. Or if the designer follows certain design methods to
choose the design, a minor change in the operation surroundings may havea small
influence on the system, but the system still can perform.
Case Two, it is beyond. Some extreme situations are very difficult to deal with or it takes
much more effort than it is worth. Although the environment is fully understood, the
design will not work in those environment. For example, a personal PC does not work in
water.
53
Case Three, it is considered. For most design, the working environment such as
temperature, humidity is considered as a constraint of the design. If the system is working
in the predefined environment, the performance is guaranteed.
Case Four, it is not considered, beyond, and cannot be ignored, but sometimes, the
system can still perform the designed functions with some modification of the system’s
form or behavior. Not all the environment combinations can be studied and understood
from conceptual design level, and sometimes, even it is possible, the information quantity
and the corresponding complexity is not controllable. This is the situation where the
flexibility of a system can be beneficial.
3. The emergence from local behavior to global effect
As a design problem, the emergence is always the result of a distributed cellular system,
but the problem is that the functionality is hardly predetermined or controlled from the
local level. The question of this research is to find out if there exists any mechanism
which could make the cellular system self-controlled in the local level, with this self-
control leading to a global functionality.
In this research, we look into the natural system which “solved” all the three problems in
biological system design; the next section will present the fundamental natural design
point we are looking into.
54
3.3.2 A Natural Approach
The normal natural system is a developmental process. A simplified model for the
development of a normal natural system is shown in Figure 3.5.
http://memsliu.pme.nthu.edu.tw/MSCL%20Projects/
Liver%20cell%20patterning.files/image003.jpg
Cell Tissue
Organ and
Organ System
Cell Growth Differentiation & “Morphogenesis”
Figure 3.5: The natural approach of "designing" a system
The biological cells are normally developed from a single cell (generally, zygote), and
developmental biology studies the process of cell, tissue, organ, and anatomy. Normally
this is known as cell growth, cell differentiation and “morphogenesis”. The growth and
control of the genetic process of cell development provide two idea of dealing with an
adaptive system.
1. DNA
DNA performs as a local to global link for any single cell in the biological system to
perceive/store the overall picture of the system, and it also provides us a chance to
reproduce the biological system with only piece of the system as long as it holds
55
complete design information within the DNA. Zouein and Jin(2010), did initial research
and proved that it is possible for a mechanical system to store the connection/assembling
information to reproduce the overall design by mechanical cells.
2. Cell Differentiation / Morphogenesis
In a normal biological system, the cells grow and divide into identical cells with the
potential to give rise to specialized cells; these cells are known as stem cells. Stem cells
themselves cannot perform any specialized functions but grow and divide in indefinite
periods. After a period of time, the cells start to differentiate and “restrain” their ability to
only express a part of their DNA series through genetic regulation. Although there are
still some arguments about the cause of cell differentiation, whether it is through a
determined process of gene expression or it is a stochastic gene expression through a
Darwinian selective process (Cell Darwinism), this fact that the production of certain
proteins can regulate other proteins’ production provides the the natural system a chance
to know the best way of dealing with the expression of certain actions. A model of
morphogenesis in biology provides us an inside look at the relations between several
gene expressions and the structural consistency of protein. The drosophila morphogenesis
is well studied, and several interacted proteins and corresponding pair rule genetic
regulatory network is present to explain why drosophila has its special pattern (Nüsslein-
Volhard, Wieschaus, 1980).
56
3.3.3 Our Approach
Combine the previous two approaches to the design of a cellular complex system; the
following are the keys to advance the current understanding of designing such a system.
1. The environment or functional requirements are impossible to fully determine and
understand for the design; a redesign sometimes costs too much, and new control
mechanisms may not be easily adapted to the original system;
2. A set of identical cells (they perform the same in similar situation) can perform very
limited function, the differentiation in performed different behaviors is required for
more robust system design;
3. The specialization of identical cells requires a distribution of particular actions based
on the physical structure of the system; it can be achieved by information distribution.
In the biological system, tissues are composed from one set of similar-functioning
cells, and the distribution of protein serves as the information.
To solve the design for robustness and resilience of a CSO system in an unknown
environment, besides a well-designed structure, more information is needed, and it needs
to perform as morphogenesis for the cell specialization. This research proposes a new
approach for designing such a system.
57
Global Functional
Req
Local Functional
Req
Local Behavior
Environment
Field based Regulation
(FBR)
Global Behavior
+
① ②
③
④
+
+
①
Local Behavior
Capacity
Local Actions
Global Effect
Figure 3.6: Our approach of designing a CSO system
In Figure 3.6, one more approach is required and acts as the regulatory genetic network
for the local cell behavior. We believe that through this process, a designer does not have
to specify all the processes for all the situations and that the field driven behavior
regulation (FBR) for behavior may provide the system some alternative choices for
dealing with exceptions on a cellular level.
As shown in Figure 3.6, the thinking framework of this research is composed of 4 steps:
Step 1, The translation process from global functional requirement to local functional
requirement (or detailed functional requirement in cellular level if possible);
Step 2, The mapping process from local functional requirement to the behavior of the
mechanical cells;
58
Step 3, The Field driven behavior regulation process (FBR) determines the Behavioral
Distribution (Behavioral Morphogenesis) from three factors: global functional
requirement, current environment, and local behaviors of cells, and this research focuses
on this process;
Step 4, The emergence process is based on the differentiated local behavior and leads to
a global behavior of the system;
As shown in the previous chapter, some research already focused on the design of local
behaviors for a cellular system, and the result of that research mostly tried to prove if
there is an optimal solution existing for a particularly defined problem. This research is
focusing on the FBR for behavior differentiation which links the requirement and
environment information with the local decision-making process, and it may undercover
some insight for the mathematicsl model which can be used universally, for any purposes
in any environment, for the understanding of the “situation” from a local cell point of
view.
3.4 Field driven behavior regulation (FBR) for CSO Framework
Our proposed work is not a full picture or an overall design principle for designing a
particular system, but it is an experiment to understand how mimicking the crucial factor
of morphogenesis in natural process of cell-to-organ development can provide another
route to solve the current design problem for flexibility or adaptability. This section
59
mainly discusses one of the complex adaptive systems: CSO system. This section starts
with the hypothesis, and then it discusses details of the framework of the BDA and FBR
in CSO.
3.4.1 Hypothesis
The following hypotheses are introduced as the foundation of our research.
Hypothesis 1: In order to achieve high adaptive capacity, the system should be complex
enough.
Although it still requires more deep investigation of this assumption, as Ashby (1956)
and Bar-Yam (2003) discussed, in order for any system which needs to survive in a
complex environment, the complexity of this system should at least be increased no less
than the complexity of the environment. Therefore, increasing the system level
complexity (more system states) seems to be one of the reasonable ways to achieve the
overall adaptive capacity of the system. By introducing the cellular system, it actually
increases the information processing capacity of the system, and may lead to better
adaptability in the system.
Hypothesis 2: For any cellular system which has high adaptive capacity, it can be
achieved through the self-organizing approach.
60
This approach is mainly from biological system ideas, because of the success of the
current biological system, it is reasonable for us to believe that the self-organizing
approach is a possible method for systems to achieve high adaptability. In this research
work, the self-organizing approach is called FBR.
Hypothesis 3: Purposeful self-organizing and adaptability can be achieved by proper
behavioral distribution of the components of a complex adaptive system.
The proposed approach is achieved by FBR, which allows the system to control itself
through an individual component’s behavioral selection, and nature already shows plenty
of examples. It is reasonable to believe that the FBR can be achieved by proper
assignment/determination of the individual behaviors, and the emergence of those
individual behaviors will result in a system complex enough with high adaptability
capacity.
3.4.2 The Framework
To achieve a way of designing a complex adaptive system, having a better understanding
of the relationship of such system design and the current design methodologies, and
solving the current system problem of high flexibility/complexity, we propose our
framework of Cellular Self-Organizing System (CSO). It serves two different purposes:
understanding the complex system as well as a metaphor of assistance forfuture complex
system design. You can see the framework in Figure 3.7.
61
Design
Representation
(Behaviors)
Task Field
(ENV+FR)
Behavior Field
(System
response)
CSO (BDA&FBR)
Framework
Figure 3.7: Field driven behavior regulation for Cellular based Complex Self-organizing
System Design Framework
This Behavior based approach (BDA) and Field driven behavior regulation (FBR)
framework is based mainly on biology ideas, complex system study and other related
fields. The biological systems have achieved much more complex and adaptive systems
than any of the current design. How do the local elements such as cells understand the
current state, and how can they decide what actions or reactions to be performed? The
research in this topic has led us to develop this frame work.
This framework is composed of three different components, Design Representation, Task
field Representation and Model, and Behavior field. Other than the previous work of
62
three corner stones, this framework is more detailed on the specific study of the behaviors
and decision-making of the local agents, and less dependent on the structure of the
system. This framework is a new view of multi-agent system behavior design and study
with the concepts from biology, which will lead the design of CSO system being similar
to biological systems. Moreover, using the approach of this framework will increase the
possibility of creating a CSO system with greater adaptive capability.
The purpose of this research is to get a better understanding of the relation between the
behavior of local cells and the global system functionality in a natural way of view.
Previous research has already specified a great deal of the representation and formation,
and this work will change the foundation of the biological-mimic representation of the
system, establish the link between the requirements and the CSO system, and discover
some of the key properties of the behaviors of the system from local to global. The focus
of this research is the FBR of behavior and developing a new design representation and
control method which are based on the new requirements.
3.5 Field driven behavior regulation (FBR) and CSO
As in the framework, the center is BDA and FBR, which can be considered as the rules
for the local agents to follow when forming the system, and the study of these two may
provide heuristics, guidance and concept limitations of the designer in designing such
system. There are several complex systems already established, such as human society, or
the global economic system. Each individual in those systems has very limited
63
knowledge of the system, but no matter how much information they know, they organize
and make judgments and act simultaneously, and a global effect forms from those limited
self-interested local actions. We believe that it is possible to locate some of those hidden
principles (in social psychology, it is called a global self-unconsciousness) in the higher
level. These principles may not be limited to certain specified requirements or specified
systems, but can be generalized to a certain category of design problems; in this way, we
can have a better understanding of the current complex system, and it may provide us a
new way of designing a complex system. One of those principles we introduce is called
FBR in this research.
Moreover, there are multiple ways to control a system. Normally in engineering,
feedback control is used, and FBR in CSO system can also be recognized as a feedback
control for a CSO system design. Because CSO can also refer to a multi-agent system
(MAS), and if all the agents in a MAS can do whatever they would like to do, the system
will produce the maximal entropy (Shannon) and the system can be meaningless and
chaotic. From the global to local, it is required to introduce certain control or regulations
for the agents to choose not all, but some, of the actions they can do in certain situations,
and we believe this will produce at least certain performance and certain required global
behavior because of the limitation of the local actions which are regulated by FBR in this
proposed research. A further detailed discussion of the FBR will be presented in the later
chapters.
64
The FBR is the link between the global and the local; based on the different requirements,
the FBR will be used to generate the decisions from the local functions or behaviors.
When the behaviors are determined, every mCell will have a copy of this behavior base
which is coded in Behavior based design DNA, and a global effect will be emergence
from the local behaviors with the requirements fulfilled. A complete set of FBR will be
the ultimate goal of CSO system design since it links from requirements and environment
to the real design of behaviors. Although it is still a question whether it is possible to
define the FBR base for all CSO system design, this research addresses the properties of
the FBR and certain ways of defining those FBR.
3.6 Behavior based approach to design DNA representation
One of the corner stones is the representation of the design; we believe that it is possible
to store all the information which is needed to guide the formation and operation of a
CSO system in a Behavior based design DNA.
Although it seems to be very straightforward to conclude that it is possible for a system to
possess certain global performance from the individuals’ decisions and actions, it is still a
question whether there exists a “global-to-local” solution for any kind of the engineering
problems, which means, if there is a global effect of a certain open or closed system,
there exists at least one solution for all of the individuals in order to achieve the global
effect. Since in this research, we are not dealing with the proof of such “global-to-local”
solution existence problem, we suppose the required global effect or global functional
65
requirement can be achieved by local cells with limited local actions. The research
question is how to achieve those global functional requirements and the system can still
possess enough adaptability capacity. The Figure 3.8 shows the demonstration of CSO
local to global and global to local relations.
Local Individual
Behaviors
Global System Effect
(Complex Behavior)
Task Field
(Current State) Emergence
Changing
Enviroment
Infor In
Figure 3.8 The CSO behavior emergence in changing environment
The BDA and FBR play very important parts of the CSO. Because it is required to design
the behaviors of the mCells in a CSO system, the key point is how to find or generate
those behaviors to become a complete set, so that each individual mCell follows this set
to become a system whose requirements are fulfilled. In this research, we only focus on
how the system chooses which of those behaviors need to be performed under certain
conditions, and the set of the behaviors has already been defined. This research is based
66
on those existing behaviors of local agents in order to develop a system which can
perform acceptably with enough adaptability in the real field operation. A mathematical
representation of behaviors of the local agents is shown in Equation (3.1).
𝐵 = { 𝑏 1
, 𝑏 2
, 𝑏 3
, … , 𝑏 𝑛 } (3.1)
The behavior set of a complex adaptive system is composed from a collection of different
behaviors of the pre-determined individual agents as shown in the equation. It left us a
question, how to code this information in a DNA-like serial so that the upper corner stone
can be achieved. The later chapter will present our model of the behavior based design
DNA (B-dDNA) in detail, and it keeps all the necessary design information
3.7 Modeling and representing the current state (task field)
Like any other control system, the CSO system with FBR still requires modeling of the
current state, and it is very important that enough information of the current situation is
fully represented by the local agents. The limitation for such a modeling process includes:
only local agents perform as observers, and their understanding of the overall system is
bounded; the decision making process is only limited to local cells which means there
may not exist a “super” brain agent which tells other agents what needs to be performed.
67
Env 1
Env 2
Req 1
Figure 3.9: The limitation of local self-organizing process
As you can see from Figure 3.9, Because of the system is fully self-organizing, it is
possible that the cells can perceive only a limited current status due to the limitation of
localized view. For example, the upper left cells only know they need to perform the
actions which needs to fulfill requirement 1, only part of left cells know there is the
environment situation 2, and Env 1 is understood by right black cells. The problem is to
determine how the system can perform and survive by a pure, localized, agent-based,
parallel decision making process. Before giving the decisions which the local agent needs
to make, the first problem is how to present the information such as localized functional
requirements or environment status. A better representation may result in a better design
and better understanding of the overall process. A later chapter will discuss a possible
representation and its related properties.
68
3.8 FBR (behavior field) Model
The last cornerstone of the framework is the link between the understanding of current
status and the actions/behaviors the local agents need to perform. It is still a new area
where not much research has been done in the design engineering field, but in biology,
the field of system biology and developmental biology really study the expression of
certain parts of DNA and its related transcription factor (the protein that controls other
proteins’ production). Several network motifs have been found and the mathematical
model of those network connections have been developed (Uri Alon, 2006). The result of
some protein production becomes the morphogenesis for the cell development, as it
prevents cells from expressing other parts of DNA to differentiate the cells from other
cells, forming the overall system. Since the mechanical cells have similar capabilities and
properties to biological cells, if it is a possible way to mimic the behavior of the
biological cells in a way of decision making – choose certain action / express certain
DNA – it will be very helpful to understand how the biological system can achieve
adaptability and how the mathematical models in the biological system can be used in an
engineering design for a complex cellular system.
69
Sx
Y
Z
Network Motifs: theory and experimental approach Uri Alon Nature Review June 2007 Volume 8
Figure 3.10 The Network Motifs: Inner interactive control in DNA
As in the Figure 3.10, a particular network motif in the biological system can result in a
specific design pattern of the signal-response in the production circuit of a biological cell.
As the FBR apply to the decision making of mechanical cells, it is possible that the
mechanical cellular system may hold similar patterns in deciding the actions. This
research is focusing on how those signal patterns in biological systems can be used in the
mechanical cellular system and what kind of mathematical model can be defined in a
mechanical cellular system. A further discussion of the Network Motif in biological
systems and the FBR in mechanical systems will be presented in the later chapter.
3.9 Conclusion
In this chapter, we discussed the global framework for CSO systems which has been
developed in order to address two issues: CSO design and the control of both
development and operation of CSO system. In order to achieve the design of CSO,
Behavior based approach (BDA) and Field driven behavior regulation (FBR) are
introduced as the link from global requirements to local behaviors. A further discussion
70
of behaviors is discussed and modeled in order to mimic the natural design of the
biological system. In the next chapter, a further study of CSO system will be presented to
show the difficulties, properties, and possible approaches of defining those FBR.
71
Chapter 4: Behaviors, Fields and FBR modeling
4.1 Introduction
The previous chapter introduced a framework for designing the control of a CSO system
by way of limiting local agents’ actions. Two major components are B-dDNA (BDA) to
represent the system design information, and FBR to determine what needs to be
performed. In this chapter, we will provide a way of modeling the behavior design
problem for this research, and a more detailed research in FBR is proposed. This chapter
will follow the thinking framework in previous chapters to discuss details in the
framework.
4.2 BDA: from requirement to behaviors
How to define the behaviors of the agents of a complex adaptive system is one of the two
major problems. If the agents’ behaviors are well defined, the system will possibly have
the desired global emerged function. The problem is not easy to solve due to the open
ended questions for any general design problem. This section starts with a general
approach of such design problems, and multiple research ideas will be presented. To
solve the design problem from general functional requirement to local agents’ behavior,
the designer normally takes these three steps:
72
Three Steps/Concepts:
Translation: Translation is defined as the transfer from the global requirements to the
local requirements if it is possible. Because there are different kinds of requirements, the
translation may be different. For example, if there is a global function requirement of
holding an object, it can be decomposed to locate the object and apply forces around the
object. It could be translated directly as mCells in a certain period need to provide the
function of sensing the targeted object, and mCells should be able to output forces when
they are around the targeting object. There is some argument about the local requirement
of the existence issue, because the decomposition of the global requirement may lead to
the local requirements eventually.
Mapping: Mapping is defined as the design process from the local requirements to the
behaviors of mCell. Different from any other mechanical system design, mapping is not
to design a physical entity with the required function performance, but a behavior which
can fulfill the requirement fully or partly.
Emergence: The emergence can be defined as the arising global effect (behavior) during
the process of the self-organizing of local agent behaviors in CSO. Similar to the idea of
"the arising of novel and coherent structures, patterns and properties during the process of
self-organization in complex systems" (Corning 2002) but the emergence here is defined
particularly for the CSO system.
73
The Figure 4.1 shows the overall demonstration of the three concepts and the relation
with FBR:
Figure 4.1 Overall demonstration of Translation, Mapping, and Emergence
Because of the difficulties in controlling the emergence of the behavior, by looking at a
local agent’s action, it is very difficult to understand or predict the overall performance.
And the emergence is one of the necessary properties for CSO system, because the
complexity has promised a certain level of adaptation ability in increasing system states.
In order to ensure the system’s overall performance, the translation and the mapping play
a very important part in the design process. We will present the challenge, the definition,
and our model of solving those problems in this chapter.
4.3 Behavior Design Challenges
The problem of designing a complex system is not trivial,, and multiple approaches have
been extensively studied and algorithms such as GA and GP have existed for decades,
but a general method for global-to-local design has never been proposed.
74
1. Modeling
In modeling of a complex system, the first question is how much information is required
to be recorded as the definition or representation of a complex system. Several
approaches, such as Nagpal’s Origami representation of pattern formation, the CORNE
for the Superbot, and the computational embryogeny for the development of a structure
have all achieved the purpose of recording the emergence of a novel complex system by
different approaches, e.g. local rule based representation, communication information
density based control mechanism and development rule based system formation
representation. The limitation of those models when applying to a generalized system
such as CSO is not easy to conquer because those solutions are for different requirements,
and based on those requirements, the solution is good enough to meet all the demands. As
a result, for the modeling of a more universal purposeful CSO, a new model is required,
and the necessary information needs to be recorded in the model.
2. Link of Local to Global and Global to Local
This problem exists in multiple fields of study; for example, in philosophy and
organization studies. One of the traditional divisions is to distinguish the study of
organization in two sub-objects, one is the “micro” study which studies the individual or
small group and another is “macro”, which focuses on the organization as a whole system.
From those studies, we understand that the global behavior of an organization is
emerging from lower level individuals or groups, and the emergence means that first, the
75
system lightly constrains the agents’ behavior, and second, the sequence of the actions
which an individual takes does not matter greatly to the overall system. Emergence
occurs due to the interconnectivity, which stands for the intricate relations across
different scales and feedback, and it is not strictly controlled by a centralized intelligence,
but rather by a distributed network with different levels of organization. Another related
study is called integrative level, or organizational level, it studies a set of phenomena
emerging on other pre-existing phenomena of lower levels; for example, life is emerging
on non-living entities. The theory of integrative levels claims that the natural world is an
organization with increasing complexity in a series of levels (Blitz 1992, Grolier 1974).
To sum up, the complexity (to increase system states so that they better fit different
situations than a rigid system) is one key to understanding the link between local and
global, and the knowledge is still lacking due to the maintenance and control of the
possible local/system state involved. How to generate the local requirement from the
global requirements is in the “counter” direction of emerging. Whether it is possible in a
general case of all requirements is still a challenging question.
3. Requirements to Behaviors mapping
Moving from the requirements to a physical entity or process is the activity of design. In
CSO system design, the problem is a different approach: it is the design of a system’s
behavior when all the physical entities are built, and the capabilities of those entities are
known to the designer. First, the capability of each individual is more than enough to
76
fulfill the requirement, but it is unclear that whether the system may have a good enough
performance because some parts may perform unnecessary actions or even “wrong”
actions. Second, some requirements of a CSO system need to be fulfilled at different
times. However for the behavior design of local agents, it is not clear how to manage the
time scale of each individual since they may be on a different timeline. In other situations,
the control of time may limit the adaptation capability of the system by limiting the
“freedom” of behavior. Third, some requirements may introduce one to multiple
behaviors to fulfill. As discussed later in this chapter, some requirements are very general,
and there is no way any of the behavior can fulfill it individually, but the requirement can
be embedded into other requirements by FBR. For example, one adaptation requirement
could be that the system needs different complexity during the operation; it is
meaningless if this requirement is not embedded into other requirements which can
increase the local choices.
To sum up, the challenge of designing a CSO system’s local behaviors exists in multiple
perspectives which need further studying, clarifying and verification. The next section
will discuss the model of the component mCell for a CSO system.
4.4 Mechanical Cell (mCell) and Individual behaviors
Mechanical Cell (mCell) is the smallest structural and functional unit of a CSO system,
and it is similar to the idea of the cells in biological system. There are plenty of examples
which make novel use of mechanical cells such as Superbot from USC, Miche from MIT,
77
etc (source). For this CSO system design, the Mechanical Cell (mCell) is not any of those
real, designed cells, but it is a general name of the smallest components of a CSO system;
mCell can be either homogeneously designed or heterogeneously designed, and the
appearance or the structure of the mCell is not unique. From the definition of the Living
System, the mCell should possess the ability to process material, energy and information.
There are several assumptions of the mCell, and based on these assumptions, it is
possible for an mCell to mimic the behavior of a biological cell:
Assumption 1 (Cellular Capability): An mCell has the ability to perform preloaded
programs, sense a limited world, process sensory information and incoming
communication, decide on its action, and interact with others.
Assumption 2 (Cellular Limitation): mCells have limited sensors, limited range for each
sensor, limited communication range with others, and a limited number of possible
actions.
78
Mechanical Cell
Local Behaviors
Sensor
Information
Communication
Action
FBR
Biological Cell
DNA
Environment
Info
Protein
Production
Survive?
Biological Info
Figure 4.2: Mechanical Cell Model vs. Biological Cell
The Figure 4.2 shows the model of a normal mCell with the comparison to a real
Biological Cell. As you can see from the figure, it has a very similar model to biological
cells. The mCell is also controlled by design representation or so called B-dDNA. The
inputs are the sensible worldand the communication information from other mCells, and
the outputs are the behaviors and communication. The feedback is not the living or dead
judgment but FBR as a self-evaluation-behavior regulation. The output is different
because, as a mechanical system, the response can be either mechanical action in energy,
material, or signaling and/or establishing communication to other cells.
79
Definition 1 (Mechanical Cell): mCell = {Cu, S, A, B};
where Cu: control unit; S = {s1, s2, ...}: Sensors/Sensor information; A = {a1, a2, ...}:
Actuators/Actions; B: designed behavior, or design information.
Behavior in CSO is defined as the mCells’ actions under certain situations since the
mCells are limited in capability, and the world which an mCell can perceive is from the
sensors which are manufactured on the mCell and the communication information from
other mCells which are also limited. It is possible to use a limited number of labels to
label all the possible actions, sensor information, and communication information. Part of
design of CSO is to design the “link” from the current state to the next action, and it is
defined as Behavior in a CSO system. For a better understanding, you can refer to the
Figure 4.3.
80
…
b
1
b
2
b
3
b
4
Behavior Base Perception Base
Action Base
Sensor Information/
Communication
Current Action
Figure 4.3: The model of Individual Behaviors
The behaviors of mCells are the links from the left side, where all the perspectives of the
world that an mCell can “see” are stored, to the right side, where all the possible actions
that an mCell can perform are stored. As it is a network from a limited number of
possibilities to the other limited number of choices, the links in between should also be
limited.
Because the goal of introducing CSO is to solve the current design complexity problem,
the behaviors cannot form a random network; at the same time, the result of such a
network may not produce any required performance, and from the study of system
biology of the network of DNA, nature only chooses some of the network connections.
What decides the links in the network is the key question of such design. We propose a
81
possible solution to decide/choose one behavior over another or decide what is necessary
in the network, through FBR. By meeting the FBR, the system may become robust and
resilient, or exhibit better performance in certain jobs. A more detailed discussion of the
BDA and FBR will be present in the next section, and this research is mainly working in
this area of CSO.
4.5 Definitions and Models
We defined the “condition” of the combination of sensor information and current action
as State in CSO; this State can refer to a real experiencing local agent state or a situation
state for behavior design.
Definition 2.1 (State): State = {S, A}
State is used to represent the situation which has the combination of the sensor
information S and current Action A.
Definition 2.2 (Current State): State = {S
C
, A
C
}
where 𝑆 𝐶 ⊂ 𝑆 and 𝐴 𝐶 ⊂ 𝐴 are currently sensor information and actions, respectively.
Current State is used to represent the situation which the current mCell is in. From the
state, we can define the behaviors for CSO.
82
Definition 3 (Behavior): b = {S
E
, A
E
} A
N
where 𝑆 𝐸 ⊂ 𝑆 and 𝐴 𝐸 ⊂ 𝐴 are the state with sensor information and/or actions,
respectively; and 𝐴 𝑁 ⊂ 𝐴 are next step actions.
A behavior b is the designed next action for some states when these states are known and
generated by the designer. The Cu of the mCell can calculate or determine the situation
and choose on next actions as b indicates. The design information of a CSO system in this
research is the fully developed behaviors for each mCell.
Definition 4 (Behaviors of System): BoS = {B
1
, B
2
, ..., B
n
};
where B
1
, B
2
, ..., B
n
are the behavior sets of all mCells in the system.
The design information for CSO is the set of all the behavior sets, and the BoS is designed
by an engineer or engineers. This design approach is different from the previous approach,
whose design information is the structure and function of the system (Zouein, Jin, 2010).
The design information is the set of all the behaviors for different mCells in different
“locations” in a CSO system; also this BoS is supposed to be designed by a designer or
designers. If all mCells share the same behavior set B, the CSO system is said to be a
homogeneous system. Otherwise, the CSO system is treated as heterogeneous.
83
From the current four definitions, it is seen that these concepts provide a solution to the
requirements which are stated in the previous section. We introduce a general way of
presenting a CSO system via designing the individual behaviors. Similar to biological
cells, if all behaviors in BoS are stored identically in every mCell in a CSO, this CSO
system has the same potential as a biological system for reproduction or adaptation.
Another similarity is that biological cells, even if they hold all the information in DNA,
can only present some of the activities due to different situations/locations/signals
through DNA transcription from certain receptors.
There are two questions still remaining for design:
1) How do we generate or define behavior sets for each mCell?
2) How do we generate “laws” in order that mCells can self-organize and emerge to
system level functionalities?
In order to solve problem 1, we propose a new representation for this CSO design for the
functional requirements. When the functional requirements are clearly defined, it is then
possible for a designer to design, and the mCells to check the performance. In this
research, we define the functional requirement with similar representation like the state as
following:
84
Definition 5 ( Functional Requirement ): FR
i
= {S
i
, A
i
}
where 𝑆 𝑖 ⊂ 𝑆 and 𝐴 𝑖 ⊂ 𝐴 form a specific state.
The reason why the functional requirement has the similar representation as State is that
first functional requirement normally serves as a goal situation, or desired situation for a
system to be working at. This goal is set by the designer or designers of the system, and
in that state, the functions are achieved; second, the functional requirement needs to be
perceivable for local mCells to recognize if the stated function requirements’ sensor
information and actions are currently achieved. This functional requirement definition in
cellular systems is more general than most of the current function definitions because
most of the current research is more focused on actions than sensor information as
functional requirement or function, and with the addition of sensor information, this
definition may provide a more detailed and more accurate definition. The second problem
is going to be mentioned and approached in later chapters with the FBR definition.
4.6 Behavior design in CSO
The design process of a CSO system in this research becomes a design process from the
functional requirement as goal state to local behaviors in Equation (4.1):
85
FR
1
= {S
1
, A
1
}
FR
2
= {S
2
, A
2
}
FR
3
= {S
3
, A
3
} BoS = {B
1
, B
2
, ...,B
m
}
(4.1)
...
FR
n
= {S
n
, A
n
}
For this design process, in a conventional approach, these functional requirements need to
be decomposed from higher and more general levels to a very detailed level in order to
simplify the task and make the detailed design possible. This process is called function
decomposition in multiple design approaches, e.g., systematic design (Pahl and Beitz
1976) or axiomatic design (Suh, 1990). Assuming that for CSO system behavior design, a
full decomposition is achieved through axiomatic design, the design result can be shown
as following in Equation (4.2):
FR
1
= {S
1
, A
1
} b
1
FR
2
= {S
2
, A
2
} b
2
FR
3
= {S
3
, A
3
} b
3
BoS = {{b
1
}, {b
2
}, ..., {b
n
}} (4.2)
...
FR
n
= {S
n
, A
n
} b
n
Equation 4.2 shows that all mCells in a CSO system may have a specified behavior set
which is different than the behavior set of other mCells. It is the result of the design
approach that all design parts are assigned with different functions in different physical
locations in order to achieve the overall requirements, and in this way, each mCell
86
becomes a component with a limited behavior set. Following the traditional approach in
CSO system design, it has the advantages of highly distributed behaviors, which may lead
to higher the efficiency, performance and lower the cost of the system without some
unnecessary behaviors. We call this approach top-down function based approach.
However, if the design goal is to achieve high system flexibility, these approaches may
lead a CSO system down the same path as a traditional system to failure. First, this top-
down function approach can provide limited system robustness; when the designer has
limited predictive capability regarding the working environment, the system will be
unpredictable in unforeseen situations as discussed in previous chapters. Second, when
the some of the functioning mCells fail, the system which depends highly on those
components will be unpredictable; in this way, the resilience of such a system is limited.
In this research, we propose a behavior based approach (BDA) to the design of CSO
system which does not require all functions or situations to be foreseen by the designer or
designers, nor do all the behaviors have to be perfectly predefined. Moreover, BDA
allows the mCells to have partial or full behavior sets, to allow for redundancy so that it
is possible for other mCells to replace the malfunctioning mCells to provide resilience in
the overall system. This BDA is not against any of the current top-down approaches
because the traditional approaches still can be applied in CSO system design in order to
define the edge functions and behaviors. However, in design for adaptability, the
complete set of functions, behaviors and working environments are very difficult to
foresee, and sometimes, they are impossible to foresee. The emergence of self-organizing
87
from mCells’ redundant behaviors will change the system functionality in different
situations or system states. As a design approach, we suppose we have met the sufficient
requirement of the behaviors from the behavior design process in CSO BDA approach,
but there exist two problems The first one is how we can code those behaviors in an agent
based system, and the second one is how the emergence of the system is guaranteed or
guided to determine the desired system level functionality. Both of those problem
approaches are going to be addressed in the following sections.
4.7 Model of behavior based design DNA (B-dDNA)
This part will introduce the design information of CSO by using a newly designed DNA,
behavior based design DNA (B-dDNA).All the necessary design information is stored in
this DNA serial. We shrink our agents of CSO from any agents to mCells. The B-dDNA
will introduce all the information which needs recording, and it is stored in mCells to
define the possible behaviors at the local level.
4.7.1 Design DNA
Behavior based design DNA (B-dDNA) is used to store all the necessary design
information of a CSO system. Because the system is emerging from individual agents, the
design information of CSO is composed mostly of the local information of the local
individuals. We suppose the storage of local behavior is sufficient for a CSO system.
88
4.7.2 Behavior based design DNA (B-dDNA) Modeling
DNA is a serial coded with only 4 different “symbols”, and for different species the
number of the series are different, humans has 24 coupes of the serials among which
there is a couple of X and (or) Y defining the gender of human beings. Although it is
difficult to tell which of those 24 are only for certain purposes, it can be concluded that
different series have different purposes. Because the B-dDNA is similar to DNA in nature,
which represents the design information of the system as well as the behaviors of
different mCells, it is important that it holds all the necessary information for the system
to be developed and operate in the field. As explained in the previous section, the
behaviors should be encoded, and it is enough to use the behavior to represent the design
since nature does it this way, the later section will present the DNA we used in this
research work.
Four components of the DNA
There are four major components of the DNA representation, which are header, behavior
set, actions and sensor information, which can be found in Figure 4.4.
89
Header Behavior Set Actions Sensor
Information
Figure 4.4 Four components of DNA representation
1. Header
Header is for explanation of the objectives or the goal of this system or subsystem and, it
is more for the designer to understand why there are such systemss, and its purpose.
Other information such as the date on which it was created, or who is the creator may
also be stored in the header section.
2. Actions
The Action set phase is the database of all the possible actions of the mCells. For
example, there is an mCell which can perform a movement action from the original
position to another position, like a leg or an arm; there should be a single line of this
action in the Actions section, starting with an identification number.
90
Each of the components in the Actions can be modeled as following:
a (Id, Action)
As shown above, the components in Actions are simple. Each has two different variables,
identification number and actions. The actions are predefined in each mCell, and as a
result, can be a simple word or identification number. Moreover, there is an action which
exists in all the DNA layers and sub-layers, which is dead, coded with an id of 0.
For example, the Figure 4.5 shows a possible example of operation behavior/action:
(Actions)
a(A0, Dead);
a(A1, Communicate Location to Left);
a(A2, Connect);
...
Figure 4.5 Example of Actions
3. Sensor Information
Sensor information or Perception is a section which contains all the self-operation
judgments, which can also be regarded as an evaluation phase. The evaluation
(perception) is coded as a series of calculation formulas which includes the local sensor
91
variables, connection information and the communication information, and every
evaluation has a predefined value for comparison of the result of the formula, and each
evaluation criteria will only return a binary true or false value. If the value is true, the
self-check will take into consideration the next proposed action or behavior. However, in
the Sensor Information phase, there are only the criteria without any suggested movement
or action.
The following is the normal or standard form of the evaluation criteria:
s(Id, formula, comparing operator ,comparison value)
For example, one of the rules of any cellular based system could be the following:
All the mCells should connect to at least one other mCell in order to keep itself in a living
situation of the whole system.
Translated into the coding, this says that the four edge sensors should have at least one
with a true value. Using the ES1, ES2, ES3…ESn to refer to the edge sensors of an mCell,
the sensor information for the rule is as follows:
92
s(2, ES1 V ES2 V ES3 V…V ESn, =, TRUE);
All the rules of self-organizing can be coded into two parts. One is the action which the
mCells or subsystem can do, and the other one is the sensor information that all the
mCells can use as the judgment of the current state of the system to decide what to do
next.
4. Behavior Representation (Gene Series of mCells)
The behavior representation is the key component of the dDNA. It has the behavior
information stored for each possible assembled arrangement cell. As a result, the
representation should have the following model:
B(mCell Id, behaviors)
Because of the need for heterogonous CSO systems, the mCell id is a requirement for the
identification of different mCells. A normal Behavior has the following three components:
Id, sensor information(SI), and suggested action(A). Normally, the self-organized mCells
keep monitoring or calculating all the behavior representation’s sensor information and
sort the possible actions in a list, and they then choose all the suitable behaviors to
perform. The example of an mCell’s representation in the DNA is shown as following:
93
B(12, b(b1, (SI)s1, (A)a1), b(b2, (SI)s1 V (SI)s2, (A)a2 & (A)a3))
The example is for a 2D cellular based mechanical system. The mCell has an id of 12.
The number of behaviors of this mCell is two; the first one is using sensor information of
id 1, if the result of the related sensor information is TRUE, it then performs the action in
the A list with id a1. It is also possible to do the logic calculation between sensor
information, and then propose certain actions.
DNA Encoding
The encoding part of the DNA is transferring the behavior design of the CSO system to a
serial coded structure which can be manipulated by genetic algorithm to mimic natural
mutation and crossover in the evolution process. For the real DNA, the behavior of a
natural cell is to produce protein in certain situations. The process is that the stimulus
protein acts as an activator which can bind to certain pattern of the DNA, which is called
the promoter, and activate the transcription of the mRNA which will lead to the
production of the indicated protein. For the design DNA of the CSO, the encoding of the
DNA can mimic the pattern of natural DNA. First, the situation analysis can be coded as
the promoter and after the promoter, and it will be the actions which are proposed to the
mechanical cell. The best pattern for a system is depends on how complex the system is
or will be. For example, if the maximum number of the actions and the sensor
information which mechanical cells can have is n, and the pattern for indicating this is an
94
action or sensor information that should at least use m bits, then m needs meet the
requirement in Equation (4.6) to fit the n actions/sensor information.
𝑛 + 2 ≤ 2
𝑚 < 2 × (𝑛 + 2) (4.3)
where 𝑚 ∈ [1,2,3, … ]
We propose to use 2 special patterns as the indicators, which can be either sensor
information or action. Normally, the pattern can be any possible pattern, but in this
research, we propose that using a serial bit of all ones as sensor information indicators,
and all zeros as the action indicator. We use numbers for all the actions and sensor
information in the serial code as a number between 0 to 2
m
.
For example, the behavior can be coded in the following pattern as shown in Table 4.1.
Behavior Design Serial encoded dDNA
Sensor infor (Total 13) s
11
[1111]
Indicator
[1100]
Sensor information
Action (Total 12) a
5
[0000]
Indicator
[0101]
Action
Final Serial Code …[1111][1100][0000][0101]…
Table 4.1: Example of Translation from Behavior Design to Serial Code
In this way, all the necessary design information of CSO design in this research can be
coded and used to a mCell based CSO system. The later section is going to discuss the
95
second problem, how we can guide the emergence process in order to achieve system
level functionality.
4.8 Field driven Behavior Regulation (FBR)
4.8.1 Initial Goal
All man-made aftifacts need to serve some human purpose, and how to achieve those
purposes is the main engineering problem. There is no exception for CSO system. By
looking into natural ways of solving this problem, the cells are differentiated from the
stem cell into different cells with different specifications. For a normal biological system,
the development from zygote can be characterized into a process of self-copy and
differentiations. The stem cells, existing in most organisms, can renew themselves
through a process called mitotic cell division and specialize into a variety of different
cells for different purposes. Other than the stem cells, normal or “simple” cells are
specialized for single or multiple purposes. The ability of growing and transferring stem
cell provides the natural way for a system response at the component level. However, the
mimicking of this process is not easy since there can never be unlimited sources to grow
and build the components of a real mechanical system, and the only way is by changing
the behaviors individually using local intelligence of the components of the CSO system,
and this reorganizing or reconfiguration of the current situation may be able to hold the
original requirements and provide adaptability. In a normal natural system, a stem cell
96
has no specific functions beyond growing and differentiating. If there is no differentiating
of the mCells (no preference for some behaviors over other behaviors under similar
situations), the mCells can be treated as stem cells with no or limited functions that can
emergent to a global effect.
For a CSO system design, if the environment is fully understood and all the possible
states are determined, the designer can, based on all the cases, design the local behaviors
of a system, and at the same time, he does not need to embed the whole behavior of the
system in B-dDNA. This research is focused on how to solve a design problem for
flexibility or adaptability with an assumption that designers also have limited
understanding of the environment and the system, and a distributed intelligence and a
copy of overall design inside every local component (mCells) will provide the potential
for more adaptability, and the following exception may happen to such designs:
1. There are more behaviors can be performed in the situation but the designed
capability of the mCells will only allow some not all to be performed;
2. None of the behaviors meet the current state (environment and system state) perfectly
(Environment Exception);
3. There are conflicting behaviors proposed under certain environment situations;
To summarize, the perfect Behavior design for a CSO system needs to fulfill the
following requirements:
97
1. The environment is purely understandable;
2. The relations between every state are clearly defined;
3. There is no intersection among all conditions of the behaviors, or it will not harm the
system (no conflicting actions proposed, etc);
Since there is a limitation on the designer, how the system can choose the right behavior
to perform is a question. In this section, a detailed discussion of the FBR will be
presented to demonstrate what the FBR is and how the FBR works when the exceptions
happen; several research issues will also be addressed.
4.8.2 Emergence of CSO systems Behaviors
It can be inferred from the above definitions and discussions that BDA with redundant
behavioral sets for CSO endows the system with a plethora of system states. Consider a
CSO system, CSO
1
, composed of two mCells, mCell
1
and mCell
2
, which have identical
behaviors embedded. We have:
𝐵𝑒 ℎ( 𝐶𝑆 𝑂 1
) = 𝐵𝑒 ℎ( 𝑚 𝐶𝑒 𝑙 𝑙 1
) × 𝐵𝑒 ℎ( 𝑚 𝐶𝑒 𝑙 𝑙 2
)
= � 𝑏 1
, 𝑏 2
, … , 𝑏 𝑝 �
𝑇 × � 𝑏 1
, 𝑏 2
, … , 𝑏 𝑝 �
=
⎣
⎢
⎢
⎡
𝑏 1
𝑏 1
𝑏 1
𝑏 2
𝑏 2
𝑏 1
𝑏 2
𝑏 2
⋯
𝑏 1
𝑏 𝑝 𝑏 2
𝑏 𝑝 ⋮ ⋱ ⋮
𝑏 𝑝 𝑏 1
𝑏 𝑝 𝑏 2
⋯ 𝑏 𝑝 𝑏 𝑝 ⎦
⎥
⎥
⎤
(4.4)
98
The mCells in a CSO system can have a large number of behaviors, which means the p in
the Equation (4.4) can be a big number. In this way, consider a CSO system with n
mCells and each mCell holds m behaviors in the behavior set. The total possible system
state for a CSO system is n
m
. When some behaviors of a behavior set have multiple
possible actions or multiple possible parameters, the potential system states can be much
bigger. In this way, the system has a lot more potentials in order to deal with exceptions
for the emergence process.
Following the biological example of stem cells, we explore CSO systems with
homogeneous setup in this research. Similar to the differentiation process for biological
cells, the homogeneously designed mCells with at least two behaviors will differentiate
and perform “special” sub-sets of the behaviors in different periods of the task during the
process of emergence. The overall process of the system is expected to perform as a
biological system’s tissue or organ formation in the way that the self-organizing process
from mCells will create functional blocks (areas) with some or all of mCells for certain
tasks; after the task period ends, the function blocks will “deform,” and mCells will, or
have the potential to, form other blocks for different tasks through similar self-
differentiation processes. The BDA provides a very large exploration of system states as
following:
99
CSO increases the system robustness through the cellular self-organizing formation,
because the system emerged behaviors (potential states) may provide functional basis for
new functional requirements, and
CSO increases the system resilience with large numbers of identically designed mCells,
and this design approach decreases the importance of each individual in the way of
system emergence. When one mCells malfunctions, other similar mCells will replace it or
provide similar behaviors.
Designing the CSO system through BDA is a difficult task. Because the biological
development process is still not well-known, the designer or designers for CSO or other
complex adaptive systems have to face the big challenge to design and develop the
system emergence. In this research, we use a “guided emergence,” rather than
uncontrolled emergence, by applying rules for local mCells to follow, and in this way, the
self-organizing and emergence will be “guided” to the desired system. The following
section will discuss and present the rules we used to solve the second problem from the
previous section.
4.8.3 FBR Definition
To solve the second problem, we looked into different design approaches, and we use a
new biological information inspired approach. In developmental biology, Morphogenesis
is defined as the control process to cause the cells to initiate shapes. As a fundamental
100
aspect, it plays as a major factor in cell growth and cell differentiation. Morphogenesis in
biological systems can be regarded as a global distribution of certain proteins which can
limit the expression of certain DNA, and this limitation is used to link a higher level
functional distribution to lower levels of the fundamental components. Since it is almost
impossible to define a complete design of different behaviors for different mCells in a
different structural location, and it makes the design of CSO system similar to other
complex system design. So the control of CSO system is how we can limit the actions of
each mCell in a similar environment with insufficient design information. In this research,
we suppose that the system will work perfectly if all the environment and system states
are known and considered by the designer during the design of behaviors, and this
research focuses what happens if the environment has exceptions.
The Input & Output:
As discussed in the previous chapter, the FBR is used as a control method to limit the
behaviors from performing. The Figure 4.6 shows the related entities of FBR.
FBR
Global
Functional Req
Environment Local Behavior
Figure 4.6 The related entities of FBR
101
The FBR takes the necessary information, functional requirement, environment (current
state), and local behavior as input, and the output of FBR will influence the choice of
local behaviors. A direct way of the output is the probability or priority of different
behaviors, and this research also uses this idea of a numeric probability/priority list of
numbers as a main output.
4.8.4 Current Approaches
1. Mathematical Approaches
Multivariable Optimization and FBR
FBR in this research has some similarities with the multivariable optimization problem in
the multi-agent area of computer science. FBR has a similar hypothesis that there exists a
better solution among finite or infinite possibilities of solutions. The Table 4.2 shows a
simple compare between FBR and multivariable optimization.
102
FBR Multivariable Optimization
Problem Domain Multi-agent CSO system Various systems
Algorithm
Determined within a finite
domain
Multiple possibilities, depends on
problem
Goal Find an acceptable result Find an optimized result
Cost Formula Weakly defined Firmly defined in most cases
Realization
Dependent
CSO Depends on problem
Variables
Behaviors Depends on problem, multiple
numerical parameters in most
cases
Table 4.2: The comparison of FBR and Multivariable Optimization
As shown in the Table 4.2, the FBR is a different problem from multivariable
optimization; this research is to prove this FBR exists in a finite domain, which means
that the algorithm from the functional requirement, environment and behaviors to the
FBR can be achieved by predetermined algorithms, and there exists a way to design the
intelligence of local mCells so that those mCells can perform acceptably. In other words,
if the FBR is determined, we can plug in CSO system, and the system can work without
changing much of calculations in the current decision. It is totally different from a
multivariable optimization problem that uses different algorithms to approach the
optimized set of variables to a given cost function or a set of functions. The following
section talks about the difference between this approach and GA/GP.
103
Genetic algorithms and FBR
There may be some misunderstanding of this method with current GA/GP algorithms. In
GA/GP, each of the agents (individuals) holds a set of the optimal result, which is a
complete set to a certain optimization problem, by using the natural processes such as
mating, mutation and killing populations; good ones survive with higher probability in
order to approach the optimized solution. But FBR in CSO system is a different approach,
each agent can only understand a limited part of the goal, by a properly assigned FBR,
and the agents can choose their own actions through a distributed intelligence In this way,
an overall behavior will be achieved without anyone within the system can understanding
the global behavior. It is more like natural systems when exceptions happens, e.g.,
platelets will gather together to the wounded area to stop bleeding, but the rest of the
system retains the original function until the wound influences other system’s
performance. The population of mCells in CSO systems cannot be killed because they are
the function performers, not one representative of an optimization solution.
Fuzzy Logic/Probabilistic Logic and FBR
Fuzzy Logic and Probabilistic Logic are two math tools to deal with reasoning that is
approximate rather than precise. Both of those two math tools have some similarities with
FBR in the form of the results. The output of FBR can be treated as a “truth” value range
from 0 to 1, and the process of FBR holds similar processes of determining the right
behavior by determining the fuzzy impact of different environmental factors. FBR will
104
adapt some of the fuzzy logic ideas of dealing with non-deterministic factors and logic.
The research of FBR is not trying to find a case-by-case fuzzy logic for certain logic
design problems, but trying to design a collective intelligence or a guideline for the agent
to emerge.
2. Biological Approaches
For a biological system, one of the ways that bacteria think is through bacterial
chemotaxis, which is a simple logic of choosing actions of a bacteria in a certain
concentration of gradient chemicals. FBR is more like natural solutions.
Figure 4.7: Signaling cascade for photophobic response for Microbial rhodopsins
105
...
64
34
80
75
CheA
CheB
CheX
CheY
Signal (Env) Protein Circuit Protein Concertration Sense Response
Figure 4.8: Demonstration of a Protein Network in Biological Chemotaxis
The Figure 4.7 provides a real protein network fora Microbial rhodosins in a response of
photophobic, and Figure 4.8 provides a demonstration of a normal bacterial chemotaxis
network in a response to some signal inputs. Those signals can be treated as
environmental information. From those two examples, we can clearly see that a
biological system’s cells can sense some chemical or physical signal by its construction
and use a biological network to control the production and the delusion travelling of
certain proteins inside, and those proteins can directly impact the responses, such as
motor frequency or protein production of cells. For FBR, we propose similar responses
for mCells to “think” as a natural cell. Since we can manipulate the intelligence of the
mCells, we do not have to generate a complicated network for the calculation of the
protein concentration, and we can use mathematical multivariable calculations to mimic
this process We believe that the signal from the environment and system itself can be
synthesized into a mathematical multivariable calculation process based on the current
physics, chemistry, mathematics and biology research. In this way, a CSO system can
106
hold one of the important factors which leads to the adaptation and flexibility of the
natural system. Nature’s success provides us hope.
4.8.5 Field driven Behavior Regulation (FBR)
From the previous section, in biology systems, a “chemical field” with multiple contents,
e.g., proteins, acids, other signals is created by the distribution of morphogenesis, and this
“field” triggers cellular functions and differentiation in the biological world. A biological
cell will produce the proteins according to the physical “location” with the respect to this
“chemical field”. In this research, we extend the idea of “chemical field” to a more
general case of “field” as the information distribution within the CSO system, and we
introduce FBR to regulate the cellular level of behavior and build the CSO system. The
following sections will present the details of FBR.
1. Two Fields: Task Field and Behavior Field
The CSO system in this research has a large potential state space to explore but is still
limited by the local design of mCells. Because the mCells have limited sensory
information (at least bounded in sensor quantity) and possible actions (bounded in
actuators), any task (from FR) and any operation environment (can be both fully known
and limited known) can be represented or “sensed” by local mCells using the presentation
of FR, Env and sensory information for mCells. We defined such representation as the
task field (tField) for CSO system. We have
107
Definition 6 (Task Field): tField := {FR, Env, S }
where, FR: function requirements; Env: environmental information; S:one mCell’s
sensory information;
obs
m
d
Figure 4.9: Example of Task Field
Figure 4.9 shows a simple example of task field in a two-dimensional space. It is a task
field representation for one mCell travelling in a field to the final destination d with
obstacles obs on the side of the path. There are two major parts of this field; one is the
destination d which creates an attraction field in order to “drag” mCells to fulfill the
functional requirement of reaching it, and the other is the obstacle obs which creates a
repelling field in order to prevent the mCell from reaching it in a similar way. For a
simple task as shown in Figure 4.9, these two fields are considered as a full information
representation of FR and Env for the mCell which is going to decide the next moving
direction.
108
When mCells are operating in a CSO system, the purpose of FBR is to differentiate the
mCell’s next actions so that a general function can be emerged, so within the CSO system,
there will be a behavior field (bField) which is determined by mCells based on the
current task field.
obs
m
d
Figure 4.10 Example of Behavior Field
Figure 4.10 shows a simple example of a behavior field in a similar task field as in Figure
4.10, the dashed line surrounding mCell m is the potential next moving direction (next
action) if there is a trivial task field without any obstacles or destination. The black curve
surrounding m is the behavior field in which some directions have higher priority or
possibility than other directions because of the combination of repelling fields and
attracting fields in the task field.
109
The bField can have different scientific meanings. One possible closest meaning is the
“reward” from a general cost function, as it shows the potential good result from
choosingto do this as next action, or on the opposite side, it can be considered as “risks”,
the potential risk to fail the task. The behavior field shows a priority for the behaviors in
either way of presentation.
2. Two Steps: FBR
FD
and FBR
DM
There is a process from tField to bField in the previous section, we define the calculation
process as FBR
FD
, and the following definition is to define both bField and FBR
FD
in this
research.
Definition 7 (Behavior Field): bField = FBR
FD
(tField)
where, FBR
FD
: FBR operator for field transformation; bField: behavior field; tField:
task field.
The bField is a localized representation only for one mCell in a CSO system, when in
different “locations” of the system, the bField is calculated similarly but with different
results since the task field is different corresponding to the “location” of different mCells.
Please be note that the “location” here means a related location in the CSO system, which
can be in 2 dimensional or 3 dimensional assemblies. If we consider the mCells’
behaviors, form the next highest “reward” or least “risk” behavior or action to the lowest
110
“reward” or most “risk,” we can form a behavior profile with different values to represent
the potential for the next action. In this way, we define the behavior profile:
Definition 8 (Behavior Profile):
bProfile = FBR
FD
(tField, B)
where: bProfile := {(b
1
, p
1
), ..., (b
n
, p
n
)}; & [b
i
ϵB, 0 ≤ p
i
≤ 1, 1≤ i ≤ n] indicates
(behavior, probability) pairs for a mCell to choose its next actions, and n is the
number of possible behaviors in mCells’ behavior set; tField: task field; B:
mCell's behavior set
After the behavior profile is calculated, an mCell can choose one or several behaviors
from the list of behaviors as its next action. In our research, we introduce FBR
DM
for
behavior selection:
Definition 9 (Behavior Selection): b = FBR
DM
(bProfile)
where: bProfile := {(b
1
, p
1
), ..., (b
n
, p
n
)}; & [b
i
ϵB, 0 ≤ p
i
≤ 1, 1≤ i ≤ n] ; b: selected
behavior b ϵ B.
111
Figure 4.11: An Illustration of Field driven Behavior Regulation (FBR) in CSO Systems
Figure 4.11 shows how FBR works in a CSO system. Using the previous discussion, the
functional requirement and working environment can be presented by actions and sensor
information. If the mCells have identical sensor information and current action as
functional requirement, the mCells just keep what they are doing currently. If the current
state of an mCell is different from the functional requirement, the FBR will be applied to
calculate the behavior field and profile from tField, and a decision will be made to reduce
the difference.
To sum up the FBR, we use the “field driven” to mimic the natural process of the
differentiation of the stem cells to functional tissues. In this FBR approach, we introduce
a new presentation as “information distribution” of the requirement, tasks, and
environment information in a local level, rather than a “physical field” such as P.H.
gradient in chemistry, or magnitude, electrical, gravity, and attraction in physics. We
112
assume that all the necessary information (FR, ENV, S) is sensible and interpreted to a
tField, and the mCell uses the two steps to transform the tField to bField and bProfile.
And in this way, the CSO system has the hope of self-organizing and emerging to a
system through a uniform representation as fields of the world and next actions. In Figure
4.12, it shows that we use the two fields to represent both the “outside” of the system and
“inside” of the system in the way of fields.
External
(Outside of the System)
Internal
(mCells)
Functional Req
+
Env (Imagine)
Behavior Capacity
Functional Req
+
Env (Real)
Behavior Perform
(Next Action)
Field driven
Behavior Regulation
Emerge
tField
bField
Figure 4.12: The summary of two fields in CSO system
In our CSO framework, the goals for different mCells are similar to keep the current
information the same or very close to the functional requirement in simple tasks, or part
of the functional requirements in complex design. The mCells need to keep themselves in
the right states, and the goals are presented as “attractors” in tField. Normally, for a
simple task design, the functional requirement is constant, but the working environment
is changing. Iin this way, the tField is changing correspondingly by changing “attractors”.
The mCells, when working in such a tField, will change their behaviors in order to
pursue the different “attractors”, and the emergence of different local behaviors will
113
emerge to global effects to change the state of the system. In this way, the system has the
dynamics to self-change and self-organize in order to counter the exceptions in the
environments.
4.9 Conclusion
This chapter mainly discusses the main part of the CSO system design, behaviors, and
FBR. Because the global system is achieved by local components with limited knowledge
and capability, the behaviors of local components play an important part in the overall
system performance. Based on the requirement of the behavior design, a detailed design
of the behavior based DNA (B-dDNA) is presented to meet the requirements of the
representation of the CSO design. All the three key components of the DNA are
introduced, Behaviors, Actions and Sensor Information, and with these three components,
the DNA can fully represent the sufficient design information, and the system can be
built upon this information. Moreover, this chapter proposed a new approach to
determine the local behaviors through a process of FBR, and the result will form a bField
from tField for different behaviors. The next chapter will present a computer simulation
based case study for CSO system design, and in the simulations, the behaviors and field
driven behavior regulation are used and applied.
114
Chapter 5: Case Study and CSO Simulation
5.1 Introduction
5.1.1 Objective
The objective of the case study is to demonstrate two concepts of BDA and FBR that can
be utilized in CSO systems in the simulation level of study. When the behavior design
and suitable FBR are used, the system can behave high flexibility when operating in a
changing environment. The demonstrated system is different from traditional
manufacturing processes, and the system is varied in different arrangements to achieve
different tasks in the process of the design problem. Although the traditional
manufacturing process has been and is being optimized, the system through this process
is normally rigid and cannot easily change itself, because even if the system can perceive
the exception the system cannot change correspondingly. Another problem for CSO
systems in this case study is that normally it requires a centralized processing unit to
handle all the exceptions and assign new function distributions for different agents. In
this case study, we would like to research pure localized intelligence from mCells for
them to cause the emergence of different performances during tasks.
The goals of the simulation are to demonstrate the self-emergence process to a system
level of functionality of this CSO system without any other information than only
behaviors as the design information, and this process is similar to the natural process of
translating the DNA to real protein production. In order to show the functionality, we
115
strictly limited the communication ability between mCells and embedded only self-
interest behaviors. Furthermore, we would like to demonstrate the system level of
flexibility (adaptability) in the way of system dynamical change in different situations. In
previous work from Zouein, a direct assembling from hard coded edge information for
CSO system is studied and demonstrated. The system in Zouein’s work can self-assemble
to different arrangements, and based on a previously designed function distribution,
become a system; the limitation of the previous simulation is that it did not demonstrate
when and how the system needs to or can change itself in a real-time operation
environment.This is the key point of this research: to demonstrate this real-time
emergence from B-dDNA and FBR based CSO system.
5.1.2 Multi-Agent system simulation and MASON
The simulation environment we use for CSO is a multi-agent system simulation
environment. A multi-agent system is essentially a system composed of agents which
have intelligence and can interact with other agents and the environment. The agent in a
multi-agent system can make its own decisions and do its actions in the environment.
Because the mCells can be considered as intelligent agents with their ability of perceiving
the environment, making decisions and performing actions in different situation, the CSO
system in this research is also a multi-agent system.
The normal research in multi-agent systems is that the designer or user creates a group of
agents, defines the interaction rules or behaviors for those agents, and later simulates how
116
those rules or behaviors emerge to a system level of some pattern, shape or behavior.
Some general theories can be generated in order to understand much more
complicated/complex systems than those simplified simulations. These researches, such
as bird flocking simulations, weather patterns prediction, etc. are proved to be very
efficient. We use multi-agent system simulation in this research to simulate the self-
assembling process and self-reconfiguration of the CSO system through its self-
organizing capacity..
This approach of validating B-dDNA and FBR using multi-agent system simulation is
suitable first because CSO system is a specialized multi-agent system with a different
setup; second, the multi-agent system has advantages in researching self-organizing
emergence processes, and last, some of the natural systems such as ant colonies or
genetics, have already been studied using this tool, and the process is proven to be
successful. In the simulation, we will embed homogeneous design information of
behaviors in every mCell by describing the behavior using direct rules instead of indirect
coding as we discussed in previous chapters because the behaviors of the simulation
mCells are simple enough (less than four behaviors), and using indirect coding may lower
the efficiency without other obvious advantages. As decentralized self-organizing rules,
we will use FBR as guidance for mCells to choose acceptable actions when exceptions
happen. In this way, both ideas will be presented and the simulation will show the
advantage and disadvantage of this process in designing CSO system.
117
This simulation uses the multi-agent system simulator, MASON, which is an open source
multi-agent simulator, written by Sean Luke, Gabriel Catalin Balan, and Liviu Panait
from George Mason University. It simulates the simultaneous and real-time decision and
actions for self-organizing using a random scheduling system builder.
5.1.3 Expected result
We hope it is able to demonstrate CSO system’s self-organizing behaviors through BDA
and FBR through modeling the CSO system as a multi-agent system and using MASON
as the simulation tool. We need some algorithm to abstract the real system’s sensor and
actuator to computer language and produce the source code in JAVA. Furthermore, we
need abstraction and coding of the “field” as working environment representation. Then
we must link the behavior with this “field” and show how the “field” will lead the
emergence from local to system functionality. MASON, which serves as both intelligence
embodiment and GUI, is used to show first that the behavior design in the local level is
possible for CSO system emergence, and second, the “field” in FBR can have its physical
appearance change dynamically when the operation situation varies in real-time
simulations. Through the demonstration of a system level of self-assembly and self-
reconfiguration, we have in fact achieved our goal of developing a CSO system (although
in the computer simulation level) through BDA whose complete design information is
behaviors, and FBR ensures the system flexibility (adaptability) when dealing with
varying environments. It is because the CSO system’s system representation,
environment perception and exception handling are based upon the principles that of
118
ideas from biological systems, through BDA, and FBR, CSO systems can have a high
system level of flexibility with limited but complete information stored in each of the
mCells. The aspects of developing new systems through this approach are to design
different behaviors and discover and classify the information distribution by different
mathematical models, which are treated as subsequent future work.
5.2 Simulation
The goal of this simulation as discussed above is to demonstrate self-organizing in the
operation of the system. As discussed before, the behavior of each cell is governed by the
B-dDNA of the CSO system. What we will set out to do is to demonstrate the emergence
of the CSO in a simulated environment. Each of the agents in the MASON environment
represents an mCell which has local intelligence with limited understanding of the overall
system and the environment. Therefore, we set out to develop a set of protocols to govern
the interactions of those agents in both development and the field operation which can
utilize the B-dDNA information in much the same way that the real biological cells
interact and collaborate.
The accomplishment of this simulation is to picture out a real-time self-assembling and
self-reconfigurable multi-agent system which can respond to environmental change and
self-change, and these changes are not achieved by a centrally controlled intelligence but
a distributed intelligence within each cell. Each cell will have the ability to sense
information about the environment and decide the action it needs to perform; other
119
possible behaviors such as sending and receiving information from other mCells are very
limited in this simulation because they are not the focus of this simulation.
To demonstrate we can using BDA to represent the system, and the FBR is suitable for
system emergence. Two computer simulation based case studies were performed and
presented. The first one is to show how a single mCell system is designed through BDA,
and what the two fields are, and the second one is to show the emergence of fields and
behaviors for CSO system in a simple example.
To summarize the design problem of the following case studies, we hypothesize the
following:
1. CSO systems can self-organize through the BDA and FBR approach for both
single cellular systems and multiple cellular systems;
2. CSO systems can mimic dynamic change like the stem cell’s ability to
dynamically change its own behavior;
3. FBR plays a very critical component in the CSO system real-time performance;
4. Global functionalities as dynamic tasks in operation can be achieved by self-
organizing from local mCells’ behavior and behavior decision making;
5. CSO systems can have both robustness and resilience through BDA and FBR
approaches.
120
5.3 Case Study 1: Single Exploration Cell
5.3.1 Problem statement
The goal of the first case study is to demonstrate the design process from the behavior
and how the field is investigated and used. A system level of multi-agent emergence will
be presented in the second case study. The design problem of this case study is for a
single mCell to travel through an ill-known fields to a given destination. Suppose it
knows it will encounter obstacles but the location of those obstacles are unknown. The
functional requirements are as following:
FR
1
= “travel to the destination point”
FR
2
= “avoid obstacles while moving”.
We assume the mCells in this simulation have the ability to move freely in the 2D plain,
and it has the action of choosing the next moving direction with limited quantity, i.e.,
only 10 to 20 equally distributed directions are possible, and it can sense the obstacles
within a certain range. We also assume that the destination is always known to the mCell.
As a designer we can design the corresponding behaviors:
121
b
1
= “move to the direction toward destination”, and
b
2
= “move away from the direction to obstacle”.
We also assume that the obstacles between the start point and the destination can be
everywhere, and the amount of obstacles is changeable. The purpose of this case study is
to show how the mCells choose their actions (behavior), i.e., next direction to move,
when given these two functional requirements and the sensor information.
5.3.2 Task Field
To analyze this problem, we first look into the task field of this problem. As designers,
we may not know where the obstacles are exactly located but we can assume the obstacle
is within the sensor range, and that it is sensed by the mCell. tField of this problem is
composed of two major fields: the attraction field of the destination and the repelling
fields of the obstacles. There are various strategies for the mCell to understand the
situation which may include the distances of the obstacles, or how close it is to the
destination direction. In this simulation, we only consider the angles of the obstacles in
the direction without considering the distance of the obstacles for two reasons. First, it
will simplify the field; second, it shows that even when lacking some information, the
FBR will provide enough information for the mCell to choose the “right” action to
perform. In Figure 5.1, we use a parameter θ to represent the field from the destination
122
point and β from the obstacles. Please note that the β could be multiple because it
depends on how many obstacles the mCell senses at a certain time.
obs
m
d
θ
β
Figure 5.1: Task field for mCell m in single exploration case study
In this way, we have the overall task field for mCell (m in Figure 5.1):
tFiled
m
= {θ; β
1
, β
2
, ..., β
n
}; where, n = no. of obstacles
All the angles in the presentation are relative to some absolute coordinate system, and we
use the default coordinate system from MASON in this case.
123
5.3.3 Behavior regulation
As discussed in previous chapter, the FBR for CSO system has two steps:
Step1: Transform tField into bField through FBR
FD
Step2: Select a specific or several behaviors/actions through FBR
DM.
FBR
FD
: In this case study, bField is as simple as one set of actions which are composed
of different directions the mCell can choose, and so is bProfile. We assume that the
mCell is taking the next possible moving direction of α, and suppose that both the
attraction field θ and the repelling field β have the same effects on the decision (which
could be optimized, since during different stages of the task, the destination may be far,
and the effect of obstacles may be greater than the destination, but without considering
one crucial factor of the distance, this simulation still shows that it is possible for the
single mCell to travel with less trouble). The FBR
FD
will return the likelihood of whether
the direction is going to be taken. The bField or bProfile is constituted from the
emergence of these two different likelihoods with a limited quantity (equal to how many
directions which mCell can take) around the mCell. So the FBR
FD
we introduce for this
case is as follows:
124
𝑏 𝐹𝑖 𝑒𝑙𝑑
𝑚 ( 𝛼 ) = 𝐹𝐵𝑅
𝐹𝐷
( 𝑡𝐹𝑖 𝑒𝑙𝑑
𝑚 , 𝐵 ) = { 𝛼 , 𝑝 𝛼 , 𝑞 𝛼 }
= { 𝛼 ,
1
√2 𝜋 𝑒 −
( 𝛼 − 𝜃 )
2
2
,
1
√2 𝜋 � 1 − 𝑒 −
( 𝛼 − 𝛽 𝑖 )
2
2
� , … ,
1
√2 𝜋 � 1 − 𝑒 −
( 𝛼 − 𝛽 𝑚 )
2
2
� }
(5.1)
where, α: possible direction for the next
p
α
: probability that direction α should be taken
q
α
: probability that direction α should be avoided, and there could be multiple
obstacles being sensed by mCell, i.e., Obs
i
, Obs
j
, … Obs
m
FBR
DM
: The next step is for the mCell to choose one action, namely one next direction
with current bField or bProfile. There are several mechanisms we can use in this case, we
define two different type of behavior selection: “select the highest possibility” and “select
any one good enough”, and the second one has different subtypes, i.e., “select among the
highest few (limited in quantity)” and “select among the several highest with certain
criteria (limited in quality)”. We introduce the following two FBR
DM
as following:
FBR
DM-B
= [Select the behavior with the highest value in the bField]
FBR
DM-G
= [Select any action from the behaviors that has a bigger than threshold
value in the bField]
125
In the following simulation, we investigate both of the strategies for a single cell in a very
complicated environment, and show the comparison of the two cases.
5.3.4 Simulation Result
To generate the simulation, there is some preset attribute of the mCell explorer: the size
of the mCell is measured as 16 in diameter in the MASON GUI environment; obstacles
are measured 20 in diameter. The size of the simulating environment is 900 (height) x
675 (width). Relative to this value, the speed of the mCell is 2 /s; the sensible distance for
obstacles is 12.
In this case study, we will show how the above mentioned behavior field can be useful
and the effectiveness of applying different behavior selection strategies.
126
Step: 17
Step: 165
Step: 268
Step: 319
Step: 411
Step: 558
Step: 650
Step: 692
Step: 730
Figure 5.2: Simulation Results of a Single mCell exploring in a Random Obstacle Field
Simulation Results
Figure 5.2 shows the screen captures with the time steps showing below the picture, and
it is from one of our simulations. As shown in Figure 5.2, a single explorer mCell (in
green color) can travel from a randomly assigned position on the left edge to a randomly
generated given destination (in blue color) close to the right edge. All the positions of the
obstacles (in brown color) are randomly generated for each simulation run. The setup of
the size, speed, and sensing distance of the system can effectively change the system
mCell explorer
Destination
Obstacles
127
performance, but since in this case study, it is not an optimization process, and the focus
is to discover if the FBR can effectively be applied to a single cell CSO system.
Particularly in this simulation, from step 411 to step 558, the mCell explorer spent 147
steps “wandering” the narrow “gate” built by the obstacles and found a way to go through
the “gate”. Another example is from step 650 to 730, where the mCell spent much less
time to travel past the three obstacles around the destination. The mCell explorer in this
case study makes its decision based on the functional requirements (tasks) and its sensory
information (environment) without any memory of the past experience, and planning of
different situations as predefined rules or guidelines. The FBR
FD
keeps transforming the
tField into a localized bField, and in this way, the mCell can “know” and calculate the
possible valid behaviors to perform at a particular moment of the task. The next step of
FBR
DM
will let the mCell choose the applicable actions to perform as the next step. The
designer can make various different combinations of the two steps of FBR and discover
that good ones depend on the task complexity and the CSO system complexity.
To test the FBR performance against the different complexity of the environment, the
simulation generates more obstacles, the density of obstacle increases, and the mCell may
have a greater chance to get trapped on the way or collide with the obstacles and fail the
task. First, our simulation can verify this assumption. Second, we investigate the different
FBR strategies in FBR
DM
from the previous section. The two different strategies are the
“behavior selection”, i.e., FBR
DM-B
(select the best) and FBR
DM-G
(select from good
128
enough, i.e. top 40%, randomly). In this case study, we ran 500 times for every obstacle
density from 40 to 120 (increasing by 10) randomly assigned obstacles to test several
results, i.e., “success rate”, two different categories for failure, and the average steps to
reach the destination.
Figure 5.3: Comparison of “Success Rate” of "Select the Best" (FBR
DM-B
) and "Select from
Top 40% randomly" (FBR
DM-G
)
Figure 5.3 indicates that in this case, FBR
DM-B
(select the best) works worse than FBR
DM-
G
with the FBR limits of the highest 40%, and when the number of obstacles increases,
FBR
DM-G
advantage increases as well. The following result can show some more
interesting perspectives on those two strategies.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
40 50 60 70 80 90 100 110 120
Success Rate
Number of Obstacles
FBRDM-G FBRDM-B
129
Figure 5.4: Comparison of “Failure Rate” of "Select the Best" (FBR
DM-B
) and "Select from
Top 40% randomly" (FBR
DM-G
) due to collision to obstacle
Figure 5.5: Comparison of “Failure Rate” of "Select the Best" (FBR
DM-B
) and "Select from
Top 40% randomly" (FBR
DM-G
) due to reaching the limitation of steps
0
0.02
0.04
0.06
0.08
0.1
0.12
40 50 60 70 80 90 100 110 120
Failure rate from colision to obstacles
Number of Obstacles
FBRDM-G FBRDM-B
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
40 50 60 70 80 90 100 110 120
Failure rate from limitation of steps
Number of Obstacles
FBRDM-G FBRDM-B
130
The Figure 5.4 and Figure 5.5 used the same simulation runs from Figure 5.3, and these
two figures show the detailed research for the reasons of the failure; one reason is the
collision, when an mCell explorer chose the wrong way to go; the other reason is due to
the limitation of steps (we set this step limit to be 2000, and the reason will be discussed
later). As those two figures show, the failure reasons of the two strategies are different;
FBR
DM-B
(select the best) will fail more due to the limitation of steps, while FBR
DM-G
will
fail more due to the collision. The reason for this situation is that when the mCell chooses
the best in bProfile, it is almost impossible for mCells to choose the “wrong” way to
collide, because it can always go backward to avoid the collision, but it is more likely that
it will run into a “loop” of going forward and backward several times and fail the task.
The next figure compares the performance of the steps of these two strategies when
successful in a task.
131
Figure 5.6: Comparison of “Number of Steps” of "Select the Best" (FBR
DM-B
) and "Select
from Top 40% randomly" (FBR
DM-G
) when success in task
Figure 5.6 shows that the different performance of these two strategies, with both the
mean value and standard deviation as the bars. The average number of steps of FBR
DM-B
is about 520 for all cases for different obstacle densities. The reason is that the width of
the environment is 900 and the speed of the mCell is 2/step, and this indicates that
FBR
DM-B
always performs the best because it always chooses the “best direction,” and the
result is that energy is saved, while FBR
DM-G
has lower performance in number of steps
but an increased the “success rate” by adding “randomness” to the selection process of
the behavior, and “randomness” is reflected in the number of steps.
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
40 50 60 70 80 90 100 110 120
Number of Steps
Number of Obstacles
FBRDM-G FBRDM-B
132
The next step for this case study is that we investigated the different FBR
DM-G
with
different setups of the “rate” of choices, from 80% to 10% with FBR
DM-B
which can be
considered as a top 1 choice.
Figure 5.7: Comparison of “Success Rate” and “Number of steps in successful runs” of
different FBR
DM-G
in the environment with 90 obstacles
Figure 5.7 is the result for the FBR
DM_G
in the environmental setup of 90 obstacles. Each
of the test results runs 500 times. As the figure shows, with an increasing the acceptance
percentage from bField, the system become more flexible in the way that the “success
rate” increases from 0.3 to 0.74. When the accept rate starts from “the best” to top 40%,
and decreases dramatically from 70% to almost zero because the FBR
DM_G
with 100% is
a pure random process. This indicates that when the system has some randomness, and
the randomness of the system is within the range of the required randomness of the
300
400
500
600
700
800
900
1000
1100
1200
1300
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 20 40 60 80 100
Averge Steps
Success Rate
FBRDM_G Limits (in percent)
SuccessRate Average Step
133
environment, the flexibility of the system from the internal variability can handle the
exceptions very well, while when there is more randomness than needed, the system will
lower the performance by wasting energy, i.e., more steps to finish same task. There is
further research in how the “flat area” of Figure 5.7 is determined and how to optimize it,
and it is beyond the goal of this research, which will be listed as a further direction. The
next figure is to show when the complexity of the environment increases from 90 to 120
in the obstacle number, and the consequences for different FBR
DM_G
.
Figure 5.8: Comparison of “Success Rate” and “Number of steps in successful runs” of
different FBR
DM-G
in the environment with 120 obstacles
Figure 5.8 shows that when the environment become more complex in the way that more
obstacles make the task more difficult to accomplish, the FBR
DM_G
performs much better
than FBR
DM-B
(15% compared with approximate 60%) with an increase of only about
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0 20 40 60 80 100
Averge Steps
Success Rate
FBRDM_G Limits (in percent)
SuccessRate Average Step
134
100 steps on average. It indicates that the randomness helps in increasing the system
complexity to deal with working environment complexity, which makes the system more
flexible.
To summarize the previous several experimental results, they indicate that the FBR
strategies have profound influence on the individual mCell’s performance in certain tasks,
and if adding “randomness” to FBR, the system has more “intelligence” in the way of
dealing with complexity. If the FBR strategy is “select the best”, the mCell targets on a
single direction for its next move. If the complexity of the environment is low (number of
obstacles is small), this strategy is good in the way that it handles both tasks and uses less
energy (steps) to complete, but when the complexity of the environment increases, it is
more likely for the “select the best” strategy to get trapped by its own “best” direction in
a loop and make the task impossible to accomplish. For “selecting among the better ones”
strategy, even when the environment complexity increases, because of the randomness
embedded, the system itself becomes complex in the way that it is not a determined
process, and this complexity makes the mCell able to sustain the environment variance
and maintain some performance. This proves the first hypothesis from Ashby’s Law of
Requisite Variety.
135
5.4 Case Study 2: CSO Mover System
5.4.1 Problem statement
In the previous case study, we demonstrated the tField formation from decomposition
from functional requirements and environment, the process from tField to bField, and the
selection process of FBR
DM
. It showed that the system can have the flexibility of
handling the environment without any learning or memory. To further investigate the
BDA and FBR for multi-agent systems, we conducted the second case study. We want to
discover how FBR can influence the emergence of a multi-agent system and how mCells
can “collaborate” to accomplish a single task. In this case, the task for multiple identical
mCells is to move an object from the randomly generated start point to a predefined
destination point (randomly generated). Similar to the previous case, the two locations of
object and destination point are on each side of the simulation environment. The
obstacles are also randomly located in the 2D simulation environment. To simplify this
problem and investigate the emergence process with less design influence, all mCells can
only push the object from its center to the object’s center. At any moment of the
simulation, each mCell can decide where its relative location is and push the object with
its own effort without knowing where other mCells are. The overall movement of the
system is determined through the emergence all pushing efforts from each mCell.
In the case study, it can be seen that the success of the overall task is determined by the
relative location of each mCell within the system, and the challenge is that each mCell
does not know and cannot calculate the location because it has no calculation ability
136
regarding other mCells in the system. The behavior or action for the mCell is to choose
where the relative location is.
To summarize the functional requirements of this case study, the following functional
requirements need to be considered.
FR
1
= “stay close to the object”
FR
2
= “push object to destination direction”
FR
3
= “avoid obstacles”
An mCell can determine a relative location too, so the designed three corresponding
behaviors are:
b
1
= “move to locations as close as possible to the object”
b
2
= “push the object towards destination”
b
3
= “push the object away from obstacles”
The mCells’ capability in this case study is similar to the previous case, which means
they can sense the destination anywhere but obstacles within only a certain range; in
137
addition to that, they can sense where the object is. We are going to present the tField of
the second case study.
5.4.2 Task Field
The task field of this case study is similar to the previous case study; we have a similar
attraction field which is created by the destination, and a repelling field which is created
by the obstacles. In addition to those two fields, we have another attracting field from the
object. In this research we use θ to represent the destination field, β to the obstacle field,
and d as the relative distance from mCell to Object. Figure 5.9 shows the related task
field for mCell m, and other mCells, i.e., mCell n, p, and q are shown in a dashed line.
The tField for mCell m is shown as following:
tFiled
m
= {d,θ; β
1
, β
2
, ..., β
n
}; where, n = no. of obstacles
m
Destin-
ation
Obstacle
n
p
q
mCells
Pushing Direction
α
β
θ
Object
Figure 5.9: Tasks Field for mCell m in CSO Mover case
138
5.4.3 Behavior Regulation
The two steps of FBR which are described in the previous case study is similar, instead of
having the mCell decide the direction of next move, we have the mCells decide which
relative location from object to perform the push action. To further address the new issue,
we decompose the problem’s FBR
FD
into the following three behaviors:
FBR
FD
: The following is that detailed decomposition of the bField we want in case study
2.
b1:
Sensor Information: Distance to Object (d);
Figure 5.10: Demonstration for the bField of behavior 1 in ideal situation
Figure 5.10 shows the demonstration of bField ideally with the green zone being
preferred, the yellow zone is acceptable red zone to be avoided.
139
Similarly for behavior 2 and 3:
b2:
Sensor Information: relative angle (α) between mCells with Object and mCells with
Destination (θ);.
Figure 5.11: Demonstration for the bField of behavior 2 in ideal situation
The Figure 5.11 shows the ideal demonstration of the bField for behavior 2.
b3:
Sensor Information: relative angle (α) between mCells with Object and mCells with
Obstacles (β).
140
Figure 5.12: Demonstration for the bField of behavior 3 in ideal situation
The Figure 5.12 shows the demonstration of FBR for behavior 3. And the real field for
this problem is the dynamic regulation overlapping of the three different fields, and this
field will determine the locations for different mCells.
For simplification, we use the normal distribution for all three behaviors, and the overall
bProfile is designed to combine all the bFields together (because each of the mCells has a
different bField) by choosing the minimal profile of behavior rate (the relative locations).
Similar to the previous case study, we introduce the following FBR
FD
, when there is only
one destination and one obstacle which can be sensed:
141
𝑏 𝐹𝑖 𝑒𝑙𝑑
𝑚 ( 𝛼 , 𝑑 ) = 𝐹𝐵𝑅
𝐹𝐷
( 𝑡𝐹𝑖 𝑒𝑙𝑑
𝑚 , 𝐵 )
= { 𝛼 , 𝑑 , 𝑝 𝑑 , 𝑝 𝛼 , 𝑞 𝛼 }
= { 𝛼 ,
1
√2 𝜋 𝑒 −
𝑑 2
2
,
1
√2 𝜋 𝑒 −
( 𝛼 − 𝜃 )
2
2
,
1
√2 𝜋 � 1 − 𝑒 −
( 𝛼 − 𝛽 )
2
2
� }
(5.2)
where α: the angle corresponding to an arbitrary predefined coordinate
d: the related distance.
p
d
: probability that distance d should be taken
p
α
: probability that pushing direction α should be taken
q
α
: probability that pushing direction α should be avoided
FBR
DM
: After the bField is established, the next step is that the mCell needs a selection
mechanism. The mCells have two actions to perform in this case study: the first is to
maintain the current relative location, and the second is to change to another location. In
this case study, we design the mCells to change their locations when the value of the
current FBR
FD
is below a threshold and get to the location with a higher value:
FBR
DM
= [Select any action, randomly from the actions that have a bigger than
threshold probability in the bField]
142
5.4.4 Simulation result
In this section, we will present the results from the previously assigned behaviors and
FBR to demonstrate how those approaches can be useful for a CSO system which is
emerged from local mCells’ individual behaviors.
Step: 36
Step: 97
Step: 180
Step: 306
Step: 369
Step: 483
Step: 556
Step: 766
Step: 879
Step: 923
Figure 5.13: Simulation for design case 2, CSO Mover simulation results
mCell Mover
Object
Obstacles
Destination
143
Figure 5.13 shows the screen captures with the time steps showing below the picture, and
it is from one of our simulations. The blue square is the object which needs to be pushed
from the left to the destination on the right, the blue circles are the mCells, and the brown
circles are the obstacles. The simulation shows a CSO emerged system that has the
overall task done from the local mCells’ relative locations to the object. The mCells in
the system attempt to get to a “highly” recommended location from the bProfile when the
current location’s bField value is below a threshold. The simulation is to study the BDA
and FBR for CSO, so the communication is limited between mCells, and they do not
know where others are, but only try not to overlap with each other. In this way, the design
effort is reduced and the system has fewer memory and learning abilities. With every
mCell pushing the square object, the location of the object is the emergence of the overall
CSO system. In this simulation setup, the CSO mover almost always achieved the task of
pushing the object to the final destination.
The advantage of the CSO system in this case study is that the overall system structure is
not predefined, and the behaviors of each mCell are not predesigned respectively for
different situations, but the CSO system can purely emerge from localized mCells’
behaviors. BDA and FBR are used to provide guidance for the emergence process to
ensure that the task is achieved. The key functions or behaviors within a system
(normally embedded in some key parts/subsystems) are distributed, and the system is
flexible with the increasing of complexity based on the Kolmogorov complexity measure
(Li and Vitanyi 2008). The advantage of system variance from the dynamics of different
144
mCells which takes a long description to present the whole system with processes is the
goal of this research, and the challenge of how we can “control” the emergence seems to
be tractable if suitable FBR is assigned. We used a very simple representation from
statistics to describe FBR
FD
, and it is possible we can generate a set from both biological
system morphogenesis and system engineering examples, which may lead to better
performance and less design effort for CSO.
One advantage of this behavior based design is that the shape of the Object and therefore
the shape of the overall system are not predefined and limited in any way. The mCells
observe the world and decide on their behaviors locally, as the global behavior and
results emerge. Based on the Kolmogorov complexity measure (Li and Vitanyi 2008), our
CSO system of multiple mCells can be considered highly complex since the states of each
mCell change dynamically without certainty, and it takes a rather long description to
capture the whole system. However, using FBR makes it possible to regulate mCells'
behaviors and to lead the emergence process to a productive direction.
145
Step: 46
Step: 110
Step: 206
Step: 298
Step: 419
Step: 563
Step: 755
Step: 905
Figure 5.14: Illustration of the dynamic bField of the CSO Mover in the simulated field of
obstacles
Figure 5.14 illustrates the dynamically changing bField throughout the overall task, and
the influence of the choices of different mCells. Figure 5.14 shows the time steps for the
case study from start to end. The bField changes due to the relative locations between the
destination, object and obstacles. To demonstrate, we use different colors for different
calculated values. In a unified value of bField, the green zone signifies that the value for
146
current position is 0.75 to 1, the yellow zone represents 0.5 to 0.75, the orange zone
represents 0.25 to 0.5, and the red zone represents values from 0 to 0.25. The mCells
attempt to choose the green zone or yellow zone to occupy, and the zone will change in
value and color corresponding to different locations. This CSO system adapts to the new
situation even though the system is composed of simply designed mCells (the capability
of one mCell can only push the object in one direction) when the FBR is used to guide the
emergence of the overall behavior of the system. The CSO system in this case study can
move the object in an ill-known environment with unknown location and numbers of
obstacles from only localized sensor information and decision making processes. It
indicates that the “field” and FBR can be used in situations where some natural field
exists (i.e., physical and chemical field).
Another simulation for case study 2 we ran is to examine what the system performance
would be with some of the mCells inactive while the mission is being performed.
147
Step: 126
Step: 270
Step: 377
Step: 445
Step: 647
Step: 881
Figure 5.15: Resilience Test by Deactivating 4 of 12 mCells at Step 400
Figure 5.15 shows the resilience test of the overall system by “deactivating” some of the
mCells in the simulation. The result above shows that the FBR is a decentralized control
method, which serves as the guidance for each mCell, so the number of mCells can be
expanded to any amount which is reasonable; second, there are 4 mCells “dead” at step
400 of the simulation, because of the control decentralization, the system still keeps
enough performance and delivered the object. The CSO system in this case study is
resilient in the way that it can adapt more mCells than needed, and also deal with the loss
Deactivated
mCells
Deactivated
mCells
148
of some mCells without much influence on the success of the task (it may influence the
system performance in other aspects, such as efficiency, time, or cost due to the number
of mCells).
The CSO system is much more flexible than traditional systems in the way that it
provides redundancies in the components of the overall design. And the CSO system in
this study is more resilient than traditional systems with specified functions or behaviors
for different parts and components. The CSO system with BDA and FBR can maintain its
performance by dynamically changing the mCells’ functionality and behavior distribution
due to a different situation’s field information when the environment and system changes.
5.5 Conclusion
To answer the hypotheses from section 2 of this chapter, we can draw the following
conclusions:
Hypothesis 1& 2: The simulation for two case studies in this chapter verifies that a CSO
system which is composed from either a single mCell or multiple mCells can self-
organize and dynamically change its behavior with the guidance from FBR.
Hypothesis 3: The detailed strategy research in case study 1 verifies that FBR is the key
factor in this approach because FBR is the “source” of randomness, and this randomness
can increase the system complexity in order to provide enough system variety to deal
with environmental change;
149
Hypothesis 4: The case study 2 verifies that a global functionality (moving the object)
can be emerged from the local behaviors (relative locations) from mCells in Figure 5.13,
Figure 5.14 and Figure 5.15;
Hypothesis 5: Both of the case study 1 and 2 verifies that the CSO system can have
robustness so that when the environment is ill-known and complex, the system still
performs acceptably, with regard to completing the task, but may lose performance with
regard to energy/time consumption (number of steps) in Figure 5.3, Figure 5.6, Figure 5.7
and Figure 5.8, and case study 2 shows that the CSO system has resilience because of the
redundancies and pure distributed FBR guidance.
In this chapter, we present two design problems; the first problem is a single explorer to
test the validation of BDA and FBR and investigate the FBR influence on performance of
single cell CSO system; the second is a multi-mCell CSO system to move one object
from its initial location to a final destination. Based on these requirements, functions
were decomposed, and the corresponding behaviors were designed, and both tField and
bField are examined. And the result of the simulation shows that it is possible for a CSO
system to have adaptable performance when the “unpredictable” situation happens in the
environment and system itself. To summarize, we have accomplished the development of
a CSO system through BDA and FBR and identified the key factor of FBR. This new
approach may open a new phase of research and development in CSO systems in order to
create a more intelligent and adaptable system.
150
5.6 Limitations
The first and most prevalent limitation with all the demonstrations and simulations in this
research is that all simulations are achieved in two dimensions. We strongly believe that
the FBR idea can be adapted to three-dimension problems with very few problems since
the tField is generated locally and the mathematical link from tField to bField is easily
modifiable to adapt to three-dimensions. If we have the FBR adapt to three-dimensions,
the system is more similar to a real-world problem or example. Second, the mCells in
both case studies can freely choose the direction or locations in the two dimensions, but
in real world, the free moving capability may not be possible, and the local
“transformation” of a design may be bounded in physical constraints. This will be a
future research point for later research. Finally, the design problems for these two case
studies are simple cases, and the influence of different sensor information, behaviors, and
corresponding FBR may be complex and not easy to understand, and the overall
performance may not be easily guaranteed. The research for CSO systems through BDA
and FBR is from the inspiration from natural systems, and natural systems have more
constraints as the information may only be physical entities such as certain chemicals,
p.H. gradients, and magnetic or electrical fields, so it is easier for a designer to use
numerical entities to create a high dimensional interactive FBR. As a result, we still have
a better chance to apply FBR in CSO systems for complex functions.
151
Chapter 6: Contributions and Future Direction
Concludeing the dissertation, we provide a detailed list of the contributions and future
directions of this research.
6.1 Contributions
The contributions of this research are as following:
1. Introduced a new CSO design approach through a non-deterministic behavior and
field approach to design a system with high flexibility/adaptability;
This new approach is a dynamic design approach focusing on the emergence of
the agents in local level for some global level of functionality. In this way, the
number of overall system achievable states is enormously increased, and more
functionalities or functionality/behavior distributions can lead to high system
flexibility/adaptability.
2. Developed a new framework of CSO design by focusing on the behaviors, which
leads to the emergence of both structures and functions;
This new framework is more focused on the individual behaviors and how to
guide that behavior/behavior selection instead of a traditional approach
framework for engineering design through a function to structure to behavior
approach. This new framework allows designers to do a “bottom-up” design by
focusing on how to design the approach/process (i.e. behaviors in CSO) locally,
and create a dynamically varying “field” for the self-organizing process globally.
152
3. Introduced a new design concept known as “field”, (i.e., tField and bField in
CSO), to approach the problem of linking global attributes with local behavior;
“Field” in this research is from the extraction of the key principle in biology,
morphogenesis. “Field” in CSO systems represents a new concept of how the
information of the overall environmental influence, behaviors, and functions is
presented and distributed within the system, and the system benefits from this
dynamically changing information distribution to both reflect the environmental
changes and system responses.
4. Introduced a new concept of B-dDNA focusing on using the behavior information
as the key information for the design of a CSO system from the inspiration of the
biological system’s DNA;
5. Developed a new method which allows mCells to self-calculate and self-decide
the next behavior to mimic the stem cell’s function distribution in simple
mechanical design problems using the environmental sensor information as
signals and B-dDNA as the DNA in biological systems;
6. Developed the simulations for CSO systems to utilize the B-dDNA and the “Field”
to self-organize in both single mCell systems and multi-mCell systems, and
visualized the CSO self-reconfigure adaptive process when the environmental
factors vary in operation.
7. Further researched the randomness in a single mCell, and its influence on the
overall performance of the task in simulation, finding that a guided complexity in
the system can help deal with the complexity in working environment.
153
6.2 Future directions
Some future directions are proposed in Figure 6.1:
Figure 6.1: Proposed future direction of this research
This research allowed the CSO system to have the abilities to change corresponding to
different situation of both the environment and system itself by the emergence behaviors
from local mCells’ decision making. These decisions are guided by FBR as a calculation
method not due to the initial design of different system states; this is similar to what
happens in natural world. Later research may raise the problems of how to deal with
“tissues” in mechanical design (the mCells with similar behavior choices). Also, further
Stage 1
(Previous)
•Cellular Self-Organizing System (reconfigurable and replaceable)
•dDNA: represent the CSO system with Structure, Behavior and Function information
•Structure reconfigurable system
Stage 2 (This
work)
•Behavior based approach (focusing at individual behaviors and selection)
•B-dDNA and FBR: creates link between the local agent with the system and
environment variance
•Emerged system from localized behavioral differentiation (cellular level)
Next Stages
•Tissue Multi-functional Sub-system System
•FBR categorization (better understanding for natural process)
•More engineering and science methods to be tested in CSO system design
154
study of FBR and its influence is also required to provide more knowledge for both
mechanical design and biology research.
As Figure 6.1 shows, the future directions can be summarized as following:
1. Extend this method to different system domains and increase the design difficulty
with more functions and more complex environments, and expand the case study
to more sophisticated problem domains;
2. Examine the trade-offs of having various combinations of mCells including
heterogeneous mCell designs and homogenous mCell designs, and the trade-off in
other problem domains such as some physical connections and transformation
rules between different substructures with the system we presented in this
research;
3. Further research into the FBR, because the FBR idea is generated from biological
morphogenesis, it is possible to categorize FBR
FD
into several categories due to
different tasks and sensor information as natural systems without the limitation of
physics in nature. Extend the research in FBR
DM
and further investigate the
influence of randomness on the overall system of multiple mCells;
155
4. Expand and further develop the computer model for CSO:
a. Increase behaviors in order to allow the system to emerge to a complex
system with multiple purposes and examine the case studies to guide the
behavior design for designer;
b. The future system should be easy to adapt to the change of “field”
(FBR
FD
), and adapt new sensor information as the difficulty of the task
increases. In this way, other mathematical representations can be used to
check the performance for the CSO system emergence;
c. Communication between mCells can be allowed, and the interaction
model should be able to adapt to the system and the information
representation can be investigated. Questions, i.e., the representation is
better as another “field” or a new interaction model, can be further
researched;
5. Develop the laboratory environment to demonstrate the field and CSO system for
some applications other than computer simulation;
156
Bibliography
1. Alon, U. (2007) “Network motifs: theory and experimental approaches”, Nature
Reviews Genetics Volume 8, 450-461.
2. Altshuller, G. and Shulyak, L., (1998). “40 principles—TRIZ keys to technical
innovation,” Worcester, MA: Technical Innovation Center, Inc.
3. Altshuller, G., (1999). “The innovation algorithm, TRIZ, systematic innovation
and technical creativity.” Worcester, MA: Technical Innovation Center, Inc.
4. Ashby W.R. (1958). “Requisite variety and its implications for the control of
complex systems.” Cybernetica 1:2, p. 83-99.
5. Attridge, T.H., (1990). “Light and Plant Responses: A Study of Plant
Photophysiology and the Natural Environment.” Routledge, Chapman, and Hall,
Inc.
6. Audesirk, G., Audesirk, T., Byers, B.E., (2007). “Biology: Life on Earth with
Physiology (8th Ed.).” Benjamin Cummings, San Francisco.
7. Autumn, K., Sitti, M., Liang, Y.A., Peattie, A.M., Hansen, W.R, Sponberg, S.,
Kenny, T.W., Fearing, R., Sraelachvili, J.N. and Full, R.J., (2001). “Evidence for
van der Waals adhesion in gecko setae.” PNAS 99, pp. 12252-12256.
8. Bailey, S.A., Cham, J.G., Cutkosky, M.R. and Full, R.J., (2000). “Comparing the
locomotion dynamics of a cockroach and a shape deposition manufactured
biomimetic hexapod.” International Symposium on Experimental Robotics
(ISER2000).
9. Ball, P., (2001). “Life's lessons in design.” Nature, 409. pp. 413-416.
10. Barrett, D., Grosenbaugh, M. and Triantafyllou, M., (1996). “The optimal control
of a flexible hull robotic undersea propelled by an oscillating foil.” Proceedings
of the IEEE AUV Symposium, pp. 1-9.
11. Bechert, D.W., Bruse, M., Hage, W. and Meyer, R., (2000). “Fluid mechanics of
biological surfaces and their technological application.” Naturwissenschaften, 87,
pp. 157-171.
157
12. Beer, R.D., Chiel, H.J., Quinn, R.D. and Ritzmann, R.E., (1998). “Biorobotic
approaches to the study of motor systems.” Current Opinion in Neurobiology, 8,
pp. 777–782.
13. Beni, G., (1988). “The concept of cellular robotic system.” Proceedings of the
IEEE Symposium on Intelligent Control, pp. 57-62.
14. Bentley, P. J., (1999). “Evolutionary design by computers.” Morgan Kaufmann
Publishers Inc.
15. Bongard, J., Zykov, V. and Lipson H. (2006), “Resilient Machines through
Continuous Self-Modeling", Science, 314. no. 5802, pp. 1118-1121.
16. Bongard, J., Lipson, H. (2007), “Automated reverse engineering of nonlinear
dynamical systems", Proceedings of the National Academy of Science, vol. 104,
no. 24, pp. 9943–9948.
17. Bongard, J., Zykov, V., Lipson, H., (2006) “Resilient Machines Through
Continuous Self-Modeling” Science 17 Vol. 314. no. 5802, pp. 1118 – 1121.
18. . Bongrand, P. (1999). “Ligand-receptor interactions.” Rep. Prog. Phys. 62, 921-
968.
19. Bowyer, A., Vincent, J.F.V., Bogatyreva, O. and Pahl, A.K., (2003). “Data
gathering for putting biology in TRIZ.” Proceedings of TRIZCON.
20. Butler Z., Murata S., Rus D., (2002b). “Distributed Replication Algorithms for
Self-Reconfiguring Modular Robots,” Distributed Autonomous Robotic Systems,
5, pp. 37-48.
21. Butler, Z., Kotay, K., Rus, D. and Tomita, K. (2002a). “Generic decentralized
control for a class of self-reconfigurable robots.” Proceedings of the ICRA ’02
IEEE International Conference on Robotics and Automation, 1, pp. 809-816.
22. Casal, A. and Yim, M. (1999). “Self-Reconfiguration Planning for a Class of
Modular Robots,” Proceedings of the SPIE Intl. Symposium on Intelligent Sys.
and Advanced Manufacturing, 3839, pp. 246-257.
23. .Castano, A. and Will, P., (2000). “Mechanical design of a module for
reconfigurable robots.” Proceedings of the 2000 IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp. 2203-2210.
158
24. Castano, A., Shen, W. M. and Will, P., (2000). “CONRO: Towards deployable
robots with inter-robot metamorphic capabilities.” Kluwer Academic Publishers,
Autonomous Robots 8, pp. 309-324.
25. Cham, J.G., Bailey, S.A. and Cutkosky, M.R., (2000). “Robust dynamic
locomotion through feedforward-preflex interaction.” Proceedings of ASME
IMECE.
26. Cham, J.G., Bailey, S.A., Clark, J.E., Full, R.J. and Cutkosky, M.R., (2002). “Fast
and robust: hexapedal robots via shape deposition manufacturing.” The
International Journal of Robotics Research, 21 (10).
27. Cham, J.G., Karpick, J.K. and Cutkosky, M.R., (2004). “Stride period adaptation
for a biomimetic running hexapod.” International Journal of Robotics Research,
23 (2), pp. 141-153.
28. Chaplin, R.C., Gordon, J.E. and Jeronimidis, G., “Development of a novel fibrous
composite material.” US patent no. 4409274.
29. .Chirikjian, G. S., Zhou, Y. and Suthakorn, J., (2002). “Self-replicating Robots for
Lunar Development.” IEEE/ASME Trans. on Mechatronics, 7, No. 4, pp. 462-472.
30. Chirikjian, G.S. and Burdick, J.W., (1991). “Kinematics of hyper-redundant robot
locomotion with applications to grasping.” Proceedings of the IEEE International
Conference on Robotics and Automation, pp. 720–725.
31. Chiu, I. and Shu, L.H., (2004). “Natural language analysis for biomimetic design.”
Proceedings of DETC ’04 ASME Design Engineering Technical Conferences and
Computers and Information in Engineering Conferences.
32. Crutchfield, J. P., Feldman, D.P., (2003) “Regularities Unseen, Randomness
Observed: Levels of Entropy Convergence,” Chaos, 2003. 15: 25-54.
33. Desiraju, G. R. (1989). “Crystal Engineering: The Design of Organic Solids.”
Elsevier, New York.
34. Dickinson, M.H., (1999). “Bionics: Biological insight into mechanical design.”
Proceedings of the National Academy of Science of the United States of America,
96 (25).
35. Dudley, R., (2002). “The biomechanics of insect flight: form, function, evolution.”
Princeton: University Press.
159
36. Evans, D. F. and Wennerstrom, H. (1999). “The Colloidal Domain: Where
Physics, Chemistry, Biology, and Technology Meet.” Wiley, New York.
37. Everist, J., Hou, F.; Shen, W. M.., (2006) “Transformation of control in congruent
self-reconfigurable robot topologies” IEEE International Conference on
Intelligent Robots and Systems, p 612-618.
38. Feldman, D. P., McTague, C. S., Crutchfield, J.P., (2008) “The organization of
intrinsic computation: Complexity-entropy diagrams and the diversity of natural
information processing” Chaos, Volume 18, 04106.
39. Fitch, R., Butler, Z., and Rus, D., (2003). “Reconfiguration planning for
heterogeneous self-reconfiguration robots.” Proceedings of the IEEE
International Conference on Intelligent Robots and Systems, pp. 2460-2467.
40. .Fogel, L.J., Owens, A.J. and Walsh, M.J., (1996). “Artificial Intelligence through
Simulated Evolution.” Wiley, New York.
41. French, M., (1994). “Invention and Evolution: Design in Nature and Engineering.”
2nd. ed Cambridge University Press.
42. Fukuda, T. and Nakagawa, S., (1987). “A dynamically reconfigurable robotic
system (Concept of a system and optimal configurations).” International
Conference on Industrial Electronics, Control, and Instrumentation, pp. 588-595.
43. Fukuda, T. and Nakagawa, S., (1988a). “Dynamically reconfigurable robotic
system.” Proceedings of the IEEE International Conference on Robotics and
Automation, pp. 1581-1586.
44. Fukuda, T. and Nakagawa, S., (1988b). “Approach to the dynamically
reconfigurable robotic system.” Journal of Intelligent and Robotic Systems,
Volume 1, No. 1, pp. 55-72.
45. George Zouein, Chang Chen, and Yan Jin (2010): Create Adaptive Systems
through “DNA” Guided Cellular Formation, Proceedings of First International
Conference on Design Creativity (ICDC2010), 2010, Kobe, Japan
46. Gilpin K, Kotay K, Rus D and Vasilescu I (2008) Miche: modular shape
formation by self-disassembly. International Journal of Robotics Research 27:
345–372.
47. Goldberg, D.E., (1989). “Genetic Algorithms in Search, Optimization, and
Machine Learning.” Addison Wesley Longman, Inc.
160
48. Goldberg, D.E., (1991). “Genetic Algorithms as a Computational Theory of
Conceptual Design.” Proc. of Applications of Artificial Intelligence in
Engineering, 6, pp. 3-16.
49. Grantcharova, V., Alm, E. J., Baker, D. & Horwich, A.L. (2001). “Mechanisms of
protein folding.” Curr. Opin. Struct. Biol. 11, 70-82.
50. Hackwood, S. and Wang, J., (1988). “The Engineering of Cellular Robotic
Systems.” IEEE International Symposium on Intelligent Control.
51. Hansen, U.N., (1999). “Modeling of bone microcracking.” NAFEMS World
Congress, pp. 1169-1179.
52. Heitler, W. J. and Burrows, M., (1977). “The locust jump. I. The motor
programme.” J. Exp. Biol., 66, pp. 203-219.
53. Hirai, K., (1997). “Current and future perspective of Honda humanoid robot.”
Proceedings of the IEEE International Conference on Intelligent Robots and
Systems, 2, pp. 500-508.
54. Hirai, K., Hirose, M., Haikawa, Y. and Takenaka, T., (1998). “The development
of Honda humanoid robot.” Proceedings of the IEEE International Conference on
Robotics and Automation, pp. 1321-1326.
55. Holl, S.M., Hansen, D., Waite, J.H. and Schaefer, J., (1993). “Solid-state NMR
analysis of cross-linking in mussel protein glue.” Arch. Biochem. Biophys., 302,
pp. 255-258.
56. Holland, J., (1975). “Adaptation in Natural & Artificial Systems: An Introductory
Analysis with Applications to Biology.” Control & Artificial Intelligence,
University of Michigan Press.
57. Hornby, G., Lipson, H. and Pollack, J., (2001). “Evolution of Generative Design
Systems for Modular Physical Robots.” Proceedings of the 2001 IEEE
International Conference on Robotics and Automation, pp. 50-56.
58. Hosokawa, K., Shimoyama, I., Miura, H. (1996). "Two-dimensional micro-self-
assembly using the surface tension of water." Sensors and Actuators a-Physical,
57, No. 2, pp. 117-125.
59. Iagnemma, K., Rzepniewski, A., Dubowsky, S., Pirjanian, P., Huntsberger, T. and
Schenker, P. (2000). “Mobile robot kinematic reconfigurability for rough –
terrain.” Proceedings of the SPIE Symposium on Sensor Fusion and
Decentralized Control in Robotic Systems III.
161
60. .Jackson, A.P., Vincent, J.F.V. and Turner, R.M., (1989). “A physical model of
nacre.” Composites Sci. Technol., 36, pp. 255-266.
61. Jin, Y., Li, W., Lu, C-Y.S., (2005). “A Hierarchical Co-Evolutionary Approach to
Conceptual Design,” Annals of the CIRP, 54/1:155-158.
62. Jin, Y. G. Zouein, S. Lu, “A Synthetic DNA based Approach to Design of
Adaptive Systems”, CIRP Annals – Manufac-turing Technology, Vol. 58/1,
pp.153-156, 2009
63. Jones, M. N. and Chapman, D. (1995). “Micelles, Monolayers, and
Biomembranes.” Wiley-Liss, New York.
64. Kitano, Hiroaki (2002), “Systems Biology: A Brief Overview”, Science 1 March
2002: V ol. 295 no. 5560 pp. 1662-1664
65. Knight, D.P. and Vollrath, F., (1999). “Liquid crystals and flow elongation in a
spider’s silk production line.” Proc. R. Soc. Lond., B266, pp. 519-523.
66. Kotay, K., Rus, D., Vona, M., and McGray, C., (1998). “The self-reconfiguring
robotic molecule.” Proceedings of the IEEE International Conference on Robotics
& Automation, pp. 424-431.
67. Koza, J.R., (1992). “Genetic Programming: On the Programming of Computers
by Means of Natural Selection.” The MIT Press.
68. Koza, J.R., (1994). “Genetic Programming II.” MIT Press.
69. Koza, J.R., Bennett, F.H., Andre, D. and Keane, M.A., (1999). “Genetic
Programming III.” Morgan Kaufmann Publishers, Inc.
70. Kumar, A., Abbott, N. A., Kim, E., Biebuyck, H. A. and Whitesides, G. M.
(1995). “Patterned Self-assembled monolayers and mesoscale phenomena.” Acc.
Chem. Res. 28, pp. 219-226.
71. Laksanacharoen, S., Pollack, A., Nelson, G., Quinn, R. and Ritzmann, R., (2000).
“Miomechanics and simulation of cricket for microrobot design.” Proceedings of
the 2000 IEEE International Conference on Robotics and Automation, pp. 1088-
1094.
72. Lee, C.Y., Ma, L., and Antonsson, E.K., (2001a). “Evolutionary and Adaptive
Synthesis Methods.” Formal engineering design synthesis, p270-320.
162
73. Lee, C-Y, Ma, L. and Antonsson, E.K., (2001b). “Evolutionary and Adaptive
Synthesis Methods.” Formal engineering design synthesis, pp. 270 – 320,
Cambridge University Press, USA.
74. Li, M. and Vitanyi, P. (2008) "An introduction to Kolmogorov complexity and its
applications", Springer
75. Li, Y., Xue, D., and Gu. P., (2007) “Design for Product Adaptability” ASME Conf.
Proc. 2007, 237.
76. Lipson, H. and Pollack, J., (2000). “Automatic design and Manufacture of
Robotic Lifeforms.” Nature, 406, pp. 974-979.
77. Lipson, H., (2005) “Evolutionary Robotics and Open-Ended Design Automation”
Cornell University, Ithaca, New York, USA.
78. Lipson, H ., (2007) “Evolutionary Robotics: Emergence of Communication,”
Current Biology, Volume 17, Issue 9, pp R330-R332.
79. Lobo, D.; Lipson, H.; Hjelle, D.A.; (2009), "Reconfiguration algorithms for
robotically manipulatable structures," Reconfigurable Mechanisms and Robots,
2009. ReMAR 2009. ASME/IFToMM International Conference on , vol., no.,
pp.13-22, 22-24.
80. Maher, M.L. and Poon, J., (1996). “Modeling Design Exploration as Co-
Evolution.” Microcomputers in Civil Engineering, 11 (3), pp. 195-210.
81. Maher, M.L., (1994). “Creative Design Using a Genetic Algorithm.” Computing
in Civil Engineering, ASCE, pp. 2014-2021.
82. Maher, M.L., (2001). “A Model of Co-Evolutionary Design.” Eng. with
Computers, 16, pp. 195-208.
83. Mann, D., (1999). “Creativity as an Exact (Biomimetic) Science.” 4th
Biomimetics Workshop at the University of Reading, UK. 24.
84. Mann, D., (2001). “An Introduction to TRIZ: The Theory of Inventive Problem
Solving,” Creativity and Innovation Management, 10, pp. 123-125.
85. Mason, R. and Burdick, J., (1999). “Construction and modeling of a carangiform
robotic fish.” International Symposium on Experimental Robotics.
86. Michalewicz, Z., (1994). “Genetic Algorithms + Data Structures = Evolutionary
Program.” Springer-Verlag, New York.
163
87. Modi, P. J., Shen, W. M., Tambe, M., Yokoo, M,. (2005) “Adopt: asynchronous
distributed constraint optimization with quality guarantees,” Artificial Intelligence,
Volume 161, Issues 1-2, Distributed Constraint Satisfaction, pp 149-180.
88. .Murata, S., Kurokawa, H., Yoshida, E., Tomita, K. and Kokaji, S., (1998). “A 3-
D self-reconfigurable structure.” Proceedings of the 1998 IEEE International
Conference on Robotics and Automation, pp. 432-439.
89. Murata, S., Yoshida, E., Kurokawa, H. and Tomita, K., (2001). “Self repairing
mechanical systems.” Springer Netherlands, Volume 10, Number 1. pp. 7-21.
90. Mytilinaios, E., Marcus, D., Desnoyer, D. and Lipson, H. (2004). “Designed and
Evolved Blueprints for Physical Self-Replicating Machines”, Proceedings of the
Ninth Int. Conference on Artificial Life (ALIFE IX), pp.15-20.
91. Nagpal. P., (2001) “Anthills Programmable Self-Assembly: Constructing Global
Shape Using Biologically-Inspired Local Interactions and Origami Mathematics,”
Massachusetts Institute of Technology, Cambridge, MA.
92. Neidle, S. (1999). “Oxford Handbook of Nucleic Acid Structure.” Oxford Univ.
Press, Oxford, U.K.
93. O’Neil, M., Ryan, C., (2000). “Grammar based Function Definition in
Grammatical Evolution.” Genetic Programming 2000: Proceedings of the 5th
Annual Conference, MIT Press, pp. 485-490.
94. Oliver, S. R., Clark, T. D., Bowden, B., Whitesides, G. M. (2001). “Three
dimensional Self Assembly of Complex, Millimeter-Scale Structures through
Capillary Bonding.” J.Am.Chem.Soc., pp. 8119-8120.
95. Pahl, G. and Beitz, W., (1996). “Engineering design: a systematic approach.” 2nd
ed. Sprinder, London.
96. Paresis, J.L., (1998). “Coevolutionary Algorithms.” In: Bck T, Fogel D,
Michalewics Z (eds). The handbook of evolutionary computation. Oxford
University Press, Oxford.
97. Parmee, I.C., (1997). “Evolutionary Computing for Conceptual and Detailed
Design.” Genetic Algorithms in Engineering and Computer Science. John Wiley
& Sons Ltd.
98. Penrose, L. S. (1959). “Self-reproducing machines.” Scientific American, 200, No.
6, pp. 105-114.
164
99. Potter, M.A. and Jong, K.A.D., (2000). “Cooperative Coevolution: An
Architecture for Evolving Coadapted Subcomponents.” Evolutionary
Computation, 8 (1), pp. 1-29.
100. Rediniotis, O.K. and Schaeffler, N.W., (1997). “Shape memory alloys in aquatic
biomimetics.” Proceedings of the 10th International Symposium on Unmanned
Untethered Submersible Technology, Special Session on Bio-Engineering
Research Related to Autonomous Underwater Vehicles, pp. 52-61.
101. Richard, N.L., (1995). “The Coevolution of Technology and Organization in the
Transition to the Factory System.” Report-no: 95-153, Department of Economics,
University of Connecticut.
102. Rodrigues, E. and Pozo, A., (2002). “Grammar-Guided Genetic Programming and
Automatically Defined Functions.” Proceedings of the 16th Brazilian Symposium
on Artificial Intelligence: Advances in Artificial Intelligence, pp. 324 - 333.
103. Rubenstein M., Krivokon M., Shen W-M, (2004) “Robotic Enzyme-Based
Autonomous Self-Replication”, IROS 04.
104. Rubenstein-Montano, B. and Malaga, R., (2000). “A Co-Evolutionary Approach
to Strategy Design for Decision Makers in Complex Negotiation Situations.”
Proceedings of the 33rd Hawaii International Conference on System Sciences.
105. Rus, D. and Vona, M., (1999). “Self-reconfiguration planning with compressible
unit modules.” Proceedings of the IEEE International Conference on Robotics
and Automation, 4, pp. 2513-2520.
106. Rus, D. and Vona, M., (2000). “A physical implementation of the self-
reconfiguring crystalline robot.” Proceedings of the IEEE International
Conference on Robotics and Automation, 2, pp. 1726-1733.
107. Rus, D. and Vona, M., (2001). “Crystalline robots: self-reconfiguration with
compressible unit modules.” Kluwer Autonomous Robots 10, pp. 107–124.
108. Saidani, S.; ,(2004) "Self-reconfigurable robots topodynamic," Robotics and
Automation, 2004. Proceedings. ICRA '04. 2004 IEEE International Conference
on, vol.3, no., pp. 2883- 2887.
109. Salamatov, Y., (1999). “TRIZ: the right solution at the right time: a guide to
innovative problem solving,” Hattem, The Netherlands: Insytec BV.
165
110. Salemi, B.; Will, P.; Shen, W.-M.; , (2003) "Distributed task negotiation in self-
reconfigurable robots," Intelligent Robots and System. (IROS 2003). Proceedings.
2003 IEEE/RSJ International Conference on , vol.3, no., pp. 2448- 2453.
111. Salemi, B., Shen, W. M. and Will, P., (2001). “Hormone controlled metamorphic
robots.” Proceedings of the IEEE International Conference on Robotics and
Automation, pp 4194-4200.
112. Sand, J. C., Gu, P., Watson, G., (2002) “HOME: House Of Modular
Enhancement---a Tool for Modular Product Redesign” Concurrent Engineering,
vol. 10 no. 2 153-164.
113. Sarikaya, M., (1994). “An Introduction to Biomimetics: A Structural Viewpoint.”
Microscopy Research and Technique, 27 (5), pp. 360-375.
114. Shen, W. M., Krivokon, M., Chiu, H., Everist, J., Rubenstein, M. and Venkatesh,
J., (2006). “Multimode locomotion for reconfigurable robots.” Autonomous
Robots, 20(2), pp.165–177.
115. Shen, W. M., Lu, Y. and Will, P., (2000a). “Hormone-based control for self-
reconfigurable robots.” Proceedings of Intelligent Autonomous Systems.
116. Shen, W. M., Salemi, B. and Will, P., (2000b). “Hormones for self-reconfigurable
robots.” Proceedings of Intelligent Autonomous Systems.
117. Shen, W. M., Salemi, B., and Will, P., (2002). “Hormone inspired adaptive
communication and distributed control for CONRO self-reconfigurable robots.”
IEEE Transactions on Robotics and Automation, Volume 18, Number 5, pp. 700-
713.
118. Shen W. M.; Salemi, B.; Will, P.;(2002), "Hormone-inspired adaptive
communication and distributed control for CONRO self-reconfigurable robots,"
Robotics and Automation, IEEE Transactions on , vol.18, no.5, pp. 700- 712.
119. Shen W. M.; Chuong, C. M.; Will, P., (2002) "Simulating self-organization for
multi-robot systems," Intelligent Robots and Systems, 2002. IEEE/RSJ
International Conference on , vol.3, no., pp. 2776- 2781.
120. Shen W. M.; Will, P., Galstyan, A., Chuong, C. M.; (2004) " Hormone-Inspired
Self-Organization and Distributed Control of Robotic Swarms," Autonomous
Robots, Vol. 17, No. 1., pp. 93-105.
166
121. .Stoy, K., Shen, W. M. and Will, P., (2002). “Using role-based control to produce
locomotion in chain-type self-reconfigurable robots.” Proceedings of the IEEE
Transactions on Mechatronics, pp. 410-417.
122. Suh, N. P., (1990). “The principles of design.” Oxford University Press, New
York, NY.
123. Suh, N. P., (2001). “Axiomatic design.” Oxford University Press, New York, NY.
124. Thomas, E. L. (1999). “The ABCs of Self-Assembly.” Science, 286, pp. 1307.
125. Tomita, K., Murata, S., Kurokawa, H., Yoshida, E. and Kokaji, S., (1999). “Self-
assembly and self-repair method for a distributed mechanical system.” IEEE
Transactions on Robotics and Automation. Vol. 15, No. 6, pp. 1035-1046.
126. Tomiyama, T., (1995). “A design process model that unifies general design theory
and empirical findings.” Proceedings of ASME’s DETC.
127. Tsakiris, D.P., Sfakiotakis, M., Menciassi, A., la Spina, G. and Dario, P., (2005).
“Polychaete-like undulatory Robotic Locomotion.” Proceedings of the
International Conference on Robotics and Automation, pp. 3018-3023.
128. Unsal, C. and Khosla, P., (2000). “Mechatronic design of a modular self-
reconfiguring robotic system.” Proceedings of the 2000 IEEE International
Conference on Robotics and Automation.
129. Unsal, C., Kiliççöte, H. and Khosla, P., (1999). “I(CES)-cubes: a modular self-
reconfigurable bipartite robotic system.” In Proceedings of SPIE, Volume 2829:
Sensor Fusion and Decentralized Control in Robotic Systems II, pp. 258-269.
130. Unsal, C., Kilic, H., and Khosla, P., (2001). “A modular self-reconfigurable
bipartite robotic system: implementation and motion planning.” Kluwer
Autonomous Robots 10, pp. 23-40.
131. Van Breugel, F.; Regan, W.; Lipson, H.; (2008) "From insects to machines,"
Robotics & Automation Magazine, IEEE , vol.15, no.4, pp.68-74.
132. Vauthey, S., Santoso, S., Gong, H., Watson, N. and Zhang, S. (2002). “Molecular
self-assembly of surfactant like peptides to form nanotubes and nanovesicles.”
PNAS 99, No. 5, pp.5355-5360.
133. Vincent, J.F.V. and King, M.J., (1996). “The mechanism of drilling by wood
wasp ovipositors.” Journal of Biomimetics, 3, pp. 187-201.
167
134. Vincent, J.F.V., (2000). “Smart by name, smart by nature.” Journal of Smart
Materials and Structures, 9, pp. 255-259.
135. Vincent, J.F.V., Bogatyreva, O., Bogatyrev, N., (2006). “Biology Doesn’t Waste
Energy: That’s Really Smart.” Proceedings of the SPIE, 6168:1-9.
136. Vincent, J.F.V., Bogatyreva, O., Pahl, A.K., Bogatyrev, N. and Bowyer, A.,
(2005). “Putting Biology into TRIZ: A database of biological effects.”
Proceedings of Creativity and Innovation Management, 14 (1)
137. Vincent, J.F.V., Jeronimidis, G., Topping, B.H.V. and Khan, A.I., (1992).
“Biomimetics of flexible composites: towards the development of new materials.”
Biomimetics, 1 (4), pp. 251-263.
138. Werfel. J., Nagpal. P., Seung. H. S., (2006) “Anthills built to order: automating
construction with artificial swarms,” Massachusetts Institute of Technology,
Cambridge, MA.
139. Whigham, P.A., (1995). “Grammatically based Genetic Programming.”
Proceedings of ML’95 Workshop on Genetic Programming - From Theory to
Real-Word Applications, pp. 33-41.
140. Whitesides, G. and Boncheva, M. (2002). “Beyond molecules: self-assembly of
mesoscopic and macroscopic components.” Proceedings of the National Academy
of Sciences of the United States of America, 99, No. 8, pp. 4769-4774.
141. Whitesides, G. and Grzybowski, B. (2002). “Self-assembly at all scales.” Science,
295, No. 5564, pp. 2418-2421.
142. Yamins. D., Nagpal. R., (2008), “A theory of local-to-global algorithms for one-
dimensional spatial multi-agent systems” Harvard University, Cambridge, MA.
US.
143. .Yim, M., (1993). “A reconfigurable modular robot with many modes of
locomotion.” Proceedings of the JSME International Conference on Advanced
Mechatronics, pp. 283-288.
144. Yim, M., (1994). “New locomotion gaits.” Proceedings of the IEEE International
Conference on Robotics and Automation, pp. 2508-2514.
145. Yim, M., Duff, D. G. and Roufas, K. D., (2000). “PolyBot: a modular
reconfigurable robot.” Proceedings of the ICRA IEEE International Conference
on Robotics and Automation, pp. 514-520.
168
146. Yim, M.; Shen, W. M.; Salemi, B.; Rus, D.; Moll, M.; Lipson, H.; Klavins, E.;
Chirikjian, G.S.; , (2007) "Modular Self-Reconfigurable Robot Systems [Grand
Challenges of Robotics]," Robotics & Automation Magazine, IEEE , vol.14, no.1,
pp.43-52.
147. Yim, M., Zhang, Y. and Duff, D., (2002). “Modular robots.” IEEE Spectrum
Volume 39, Issue 2, pp. 30-34.
148. Yim, M., Zhang, Y., Roufas, K., Duff, D. and Eldershaw, C., (2002). “Connecting
and disconnecting for chain self-reconfiguration with PolyBot.” Transactions on
Mechatronics, 7, pp. 442-451.
149. Yogev, O.; Shapiro, A.A.; Antonsson, E.K.; , (2010) "Computational
Evolutionary Embryogeny," Evolutionary Computation, IEEE Transactions on ,
vol.14, no.2, pp.301-325.
150. Yoshikawa, H., (1981). “General design theory and a CAD system, in Man-
machine communication in CAD/CAM.” North Holland, Amsterdam,
Netherlands.
151. Yu, C. H.; Nagpal, R.; (2009) “Engineering self-adaptive modular robotics: a bio-
inspired approach,” Proceedings of the 2009 IEEE/RSJ international conference
on Intelligent robots and systems, p.414-415, St. Louis, MO, USA
152. Yu, C. H.; Haller, K.; Ingber, D.; Nagpal, R.; , (2008) "Morpho: A self-
deformable modular robot inspired by cellular structure," Intelligent Robots and
Systems, 2008. IROS 2008. IEEE/RSJ International Conference , vol., no.,
pp.3571-3578, 22-26
153. Yu, C. H.; Nagpal, R., (2009) "Self-adapting modular robotics: A generalized
distributed consensus framework," Robotics and Automation, 2009. ICRA '09.
IEEE International Conference , vol., no., pp.1881-1888, 12-17
154. Yu, C. H., Nagpal, R., (2008) “Sensing-based shape formation on modular multi-
robot systems: a theoretical study,” Proceedings of the 7th international joint
conference on Autonomous agents and multiagent systems, Estoril, Portugal
155. Zouein, G., Jin, Y. (2008) “A biologically inspired DNA-Based Approach to
Developing Cellular Adaptive Systems”, USC, Los Angeles, CA, USA
156. Zykov, V., Mytilinaios, E., Adams, B. and Lipson, H. (2005). “Self Reproducing
Machines.” Nature, 435, pp. 163-164.
Abstract (if available)
Abstract
Complexity of a system grows as more functionality is required by customers and more unintended component interactions are added to the system by designers as they make design decisions. The increasing level of both intended and unintended complexity of systems has made it extremely difficult, if not impossible, for designers to ensure reliability of, and instill adaptability into, their designed systems. As demands for adaptive systems increase in areas such as space and ocean exploration and rescue and military missions, how to guarantee reliability and increase flexibility and/or adaptability of complex engineered systems is a major challenge. While research in biological systems has advanced our understanding of how these systems have been designed and developed and revealed their fundamental properties of adaptation through morphogenesis, there has been little research in exploiting the biological design process for the development of engineered flexible and/or adaptive systems. A new engineering approach is needed that can overcome the limitations of conventional design methods by applying the concepts and processes found in the development of biological systems. ❧ In this dissertation, we develop a framework for understanding the limitations of the conventional design process for designing complex adaptive systems. Based on the previous work on mechanical cell (mCell) based system formation, we propose a novel biology inspired system representation called Behavior-based design DNA (B-dDNA) for the development and operation of our Cellular Self-organizing Systems, or CSO systems for short. Based on the B-dDNA representation, a mechanism called Field driven Behavioral Regulation (FBR) is proposed that implements and synthesizes system Designing, Formation, Operation, and Adaptation processes. FBR of a CSO system is a mathematical and selection model that is shared by all mCells and specifies cellular behaviors corresponding to functional, system level, operational and adaptation requirements. Our research results have demonstrated the feasibility and advantages of the B-dDNA representation and FBR based mechanisms. Two case studies along with more detailed computer simulations are provided to demonstrate the power of B-dDNA and FBR for designing and developing complex engineered adaptive systems that possess inherent capabilities to cope with increasing level of system complexities and to exhibit high level flexibility/adaptability in response to both task and environmental changes that occur in mission situations.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Dynamic social structuring in cellular self-organizing systems
PDF
A meta-interaction model for designing cellular self-organizing systems
PDF
A biologically inspired DNA-based cellular approach to developing complex adaptive systems
PDF
Behavioral modeling and computational synthesis of self-organizing systems
PDF
Modeling enterprise operations and organizations for productivity improvement
PDF
Managing functional coupling sequences to reduce complexity and increase modularity in conceptual design
PDF
A synthesis approach to manage complexity in software systems design
PDF
A social-cognitive approach to modeling design thinking styles
PDF
Control of spacecraft with flexible structures using pulse-modulated thrusters
PDF
Reward shaping and social learning in self- organizing systems through multi-agent reinforcement learning
PDF
Dynamic analysis and control of one-dimensional distributed parameter systems
PDF
Nonlinear control of flexible rotating system with varying velocity
PDF
Active delay output feedback control for high performance flexible servo systems
PDF
On the synthesis of controls for general nonlinear constrained mechanical systems
PDF
Design, modeling and analysis of piezoelectric forceps actuator
PDF
An approach to dynamic modeling of percussive mechanisms
PDF
Modeling and vibration analysis of wheelchair-users
PDF
Collaborative stimulation in team design thinking
PDF
Large-scale path planning and maneuvering with local information for autonomous systems
PDF
Control of two-wheel mobile platform with application to power wheelchairs
Asset Metadata
Creator
Chen, Chang
(author)
Core Title
Building cellular self-organizing system (CSO): a behavior regulation based approach
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Mechanical Engineering
Publication Date
04/27/2012
Defense Date
01/27/2012
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
adaptive system,cellular system,complex system,Field,OAI-PMH Harvest,self-organizing
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Jin, Yan (
committee chair
), Flashner, Henryk (
committee member
), Shiflett, Geoffrey R. (
committee member
), Wang, Pin (
committee member
)
Creator Email
changche@usc.edu,evanu@hotmail.com
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-15083
Unique identifier
UC11290020
Identifier
usctheses-c3-15083 (legacy record id)
Legacy Identifier
etd-ChenChang-666.pdf
Dmrecord
15083
Document Type
Dissertation
Rights
Chen, Chang
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
adaptive system
cellular system
complex system
self-organizing