Navigation Control and Path Mapping of a Mobile Robot using Artificial Immune Systems

R. Challoo, P. Rao, S. Ozcelik, L. Challoo, & S. Li
International Journal of Robotics and Automation, Volume (1): Issue (1) 1
Navigation Control and Path Mapping of a Mobile Robot using
Artificial Immune Systems
Rajab Challoo & Prashant Rao
Dept. of Electrical Engineering & Computer Science
Texas A&M University-Kingsville
Kingsville, Texas 78363-8202 USA
Selahattin Ozcelik
Dept. of Mechanical & Industrial Engineering
Linda Challoo
Dept. of Educational Leadership & Counseling
Shuhui Li
Dept. of Electrical & Computer Engineering
University of Alabama
Tuscaloosa, AL 35487-0286 USA
Abstract
This study aims to apply Artificial Immune Systems (AIS) to a mobile robot making it
capable of traversing an unknown environment and mapping it while looking for the
target. We have implemented a mixture of Antibody-Antibody (Ab-Ab) interaction
algorithm coupled with negative selection algorithms to develop the proposed AIS
controller. We have also developed a method for random generation of antibodies to
make the system more similar to the actual biological process. Finally, a generalized
architecture for representation of antibodies and antigens in a standard mobile robot
using proximity sensors for interaction with the environment has been introduced. The
results show that the proposed algorithm was able to explore the unknown
environments while learning from past behavior and look for the target. It was also
able to successfully map the traversed path and plot the obstacles based on their
type.
Keywords: Mobile Robots, Artificial Intelligence, Immune System, Path Planning, Mapping, Learning
1. INTRODUCTION
Classical control systems have been observed under dynamic changing environments and have shown
brittleness under such conditions. This leads to the increase in interest for biologically inspired
approaches; Artificial Neural Networks (ANN), evolutionary computation, genetic algorithms, and now
Artificial Immune Systems (AIS). Artificial Immune System is an optimization algorithm based on the

working of the biological immune system. This algorithm mimics and exploits the various behaviors of
the immune system such as learning, remembering and problem solving.
The vertebrate immune system is a highly evolved and coordinated system made up of various organs,
enzymes, cells and molecules. The human body is a powerhouse and hence it is targeted by many
foreign agents for its energy and chemical resources, called antigens. In order for effective functioning,
the immune system should be able to distinguish between antigens and its own molecules. In other
words, the system should be able to learn, recognize patterns, and be able to remember them. When
compared with other biological activities, the vertebrate immune system takes a small time frame as
short as a few days to achieve the evolutionary targets. All these capabilities make the immune
systems an ideal candidate for studies and implementation in computers as the system converges to a
solution faster than most biological functions of the body [2].
This study proposes an adaptive learning mechanism employing a behavior-based knowledge and the
online adapting capabilities of the immune system and applies to intelligent mobile robot. The Artificial
Immune System equations are transformed to calculate the values, affinities and concentration of the
antigens and antibodies which correspond to the conditions and actions of the robot. The calculated
values define the basic robot actions to browse through an unknown environment, mapping, obstacle
avoidance and target recognition. The dynamics of immune structure, employed to teach robot
movements, were engineered to achieve these types of behavior in spite of any changes in the robot’s
environment. The results obtained demonstrate the controller’s adaptability towards obstacle avoidance
and mapping in different environments.
Artificial Immune Systems was conceived around 1986 when Farmer, Packard and Perelson came out
with their paper on immune systems [2]. The AIS concept became a popular subject area only in the
mid-90s when Forrest et al came up with the Negative Selection theory in 1994 [27]. Dasgupta followed
on to conduct extensive studies in the field of negative selection algorithms [37]. Hunt and Cook came
up with Immune learning concepts [18] in 1996. Timmis and Neal made some improvements to the
existing AIS models by adding more biological functions to them [25]. De Castro & Von Zuben and
Nicosia & Cutello worked on Clonal Selection theory in early 2000 [3]. Also in 1994, Forrest [19-22],
applied the immunological adaptive theories to computer security systems devising a network intrusion
detection system that considered an intrusion attempt as an antigen and tried to evolve itself till it
reaches a solution to block the attempt.
Dasgupta also did extensive studies in the field of Negative Selection which stated that T-cells that
recognize self molecules (and peptides) with high affinity are deleted from the repertoire of cells. He
applied these theories on anomaly detection and in the application of germinal center dynamics in AIS
for improving the algorithm. He summarized that the immune system has the following features: self
organization, memory, recognition, adaptation and learning.
We begin by reviewing the basic concepts of the biological immune system covering its components
and basic mechanisms. A brief review of the Hardware and Software components used for
implementation of the desired immune algorithm is then followed by our approach for design and
implementation of the controller including the hardware interface. Finally, we discuss the results of our
algorithm implementation and conclude with a discussion about the contributions made by this paper
towards AIS research.
2. ARTIFICIAL IMMUNE SYSTEMS
The animal body can be compared to a powerhouse where constant chemical, physical and biological
processes generate energy and other valuable resources. Hence, it is almost obvious that foreign
agents will target the body to exploit its energy and resources. To counter such attacks, the vertebrate
immune system searches, identifies and destroys foreign bodies such as bacteria, virus and other
parasites, called pathogens or antigens. The immune system on the other hand is made up of
molecular agents called antibodies which lookout to seek and destroy antigens. Antigens are destroyed

in part when an antibody attaches itself to them, marking them in the process for eventual elimination
by lymphocytes, phagocytes and the complement system.
The human body contains almost 10
7
-10
8
different types of antibodies, each with a unique chemical
composition. These groups of cells that interact and defend the body against a variety of external
agents are formed out of a few precursor cells. Then, the cells of the immune system develop through
a constantly repeating cycle. The stem cells also reproduce themselves and in the process form many
specialized cells such as B-cells, macrophages, inflammatory T-cells, killer T-cells, and helper T-cells.
The immune system is formed of cells that are mobile and mostly remain unattached from any other
cells. This characteristic is crucial to their function.
2.1 Various components of the vertebrate immune system
2.1.1 Lymphocytes [9]: The cells of the adaptive immune system are special types of leukocytes,
called lymphocytes. B-cells and T-cells are the major types of lymphocytes and are derived from
hematopoietic stem cells in the bone marrow.
FIGURE 2.1: Association of a T-cell with MHC class I or class II, and antigen (in red) [9]
B-cells are responsible for the direct immune response where the antigen is directly encountered;
whereas T-cells are involved in inter cell based immune response. Both B-cells and T-cells are
designed to respond to only specific antigens. Antigens that are processed and combined with a “self”
receptor called major histocompatibility complex (MHC) [9] are deemed as “non-self” target or a
pathogen and are recognized by the T-cells as shown in figure 2.1. On the other hand, B-cells have
antibodies on their surfaces with receptors which recognize the whole antigen without the need of extra
processing. Each B-cell represents a unique antibody. Hence, a set of B-cells will represent all the
antibodies that the immune system can produce in a body.
2.1.2 Killer T-cells [9]: Killer T-cells are a type of T-cells that attack and destroy self cells that are
either infected with the pathogens, or are damaged and useless. Like B-cells, each type of T-cell
recognizes a different antigen. Killer T-cells are activated when their T-cell receptor (TCR) binds to this
specific antigen in a complex with the MHC Class I receptor of another cell. The T-cell then searches
through the body for situations in which the MHC I receptors bear the specific pathogen. As soon as
the killer T-cells come across such a case, they release chemicals called cytotoxins that cause the
formation of holes in the cells allowing water, ions and other chemicals to seep through them as shown
in figure 2.2. This causes the target cell to undergo apoptosis. This process of killing of infected self
cells is rather important to the body as it stops the replication process of viruses. T-cell activation is a
very tightly controlled process as it might cause harm to the system itself and hence only a very strong
MHC receptor signal can trigger it.

FIGURE 2.2: Killer T-cells directly attack other cells carrying foreign or abnormal antigens on their
surfaces. [9]
2.1.3 Helper T-cells: Helper T-cells can be considered as regulating bodies within the immune
system. They decide the type of response, innate or adaptive, that the system will perform on a certain
antigen. They do not have the capability to release cytotoxins or to attack and destroy an infected cell.
Rather, they direct the immune system towards the antigen and regulate the overall process of
intercepting and stopping a pathogen attack.
Helper T-cells have special receptors called T-cell receptors (TCR) that recognize the antigen attached
to class II MHC molecules. The MHC: antigen compound is also recognized by another special cell
called the CD4 cell help co-receptor, which initiates the molecules inside a T-cell that cause the
activation of the killer T-cells. Helper T-cells have a low association with the MHC: antigen compound
noted compared to the killer T-cells, which means that many receivers (about 200-300) on a helper T
lymphocytes have to be bound by an MHC: antigen compound to activate it, on the other hand, killer T-
cells can be activated by the participation of a single MHC: antigen molecule. The helper T-cells also
take a loner time to recognize an antigen-presenting cell. The activation of a helper T-cell causes a
chemical release that sets in action a lot of cells that make up the immune system. The chemical
produced by the helper T-cells improve the functioning of macrophages and activity of killer T-cells.
Also an activated helper T-cell regulates a lot of sensitive molecules on the surface of T-cell such as
the CD40 ligand (also known as CD154), which provide stimulus signals typically required to trigger
production of antibody generating B-cells. Overall, the helper T-cells are the regulatory bodies within
the immune system that orchestrate the response of the body towards a pathogen or cells that are
infected.
2.1.4 B lymphocytes and antibodies: B-cells recognize an antigen when the antibodies attached to
its surface are able to bind to the specific antigen. This antigen/antibody complex is processed using
proteolysis by the B-cell developed into peptides. These antigenic peptides then reflect on the B-cells
surface as MHC class II molecules. This combination of MHC and antigen is recognized by a helper T-
cell which then releases chemicals called lymphokines that activate the B-cell into production of
antibodies and cell division. Once the cell division starts, the B-cell starts mass production of the
specific antibody type that recognized the antigen in the first place.
The produced antibodies then flow through the blood stream on plasma cells or lymph cells, attach to
the antigen and ark them for eventual destruction by compliment activation of by phagocytes.
Antibodies can also attack antigens independently or mix with their receptors to mitigate their attack. [2]
2.2 How the immune system defends the body
The body is protected by a combination of millions of cells and molecules that work in tandem with
each other. The ultimate goal of all immune responses is an external agent that wants to exploit the
rich resources of energy or minerals that the body possesses. Some of these external agents or
pathogens are symbiotic while others cause harm, e.g. bacteria, virus etc. Specialized antigen
presenting cells, such as macrophages, travel through the blood stream, attacking the antigens and
breaking them up into antigenic peptides. Fragments of these peptides are combined together to form

MHC molecules that are visible on the cell surfaces. T-lymphocytes, another immune agent, have
special receptors that recognize unique MHCs-peptide combinations and activate accordingly. The
activated T-cells divide and secrete chemicals called lymphokines which are chemical signals which
direct the other agents of the immune system towards the pathogen. One of the agents that respond to
these lymphokines is the B-lymphocyte. The B-cell has receptors that recognize the anitigen based on
the specific antibodies that it can produce.
One of the major difficulties in making a perfect recognition system is that the antibodies may detect
and destroy the tissue of the host organism. To prevent this, the immune system can either (1) block
the production of antibodies that react with the molecules of the host organism; or (2) eliminate or
suppress those antibodies once they are produced. This is called the self-not- self recognition problem.
Plan (1) would not be feasible as any prior information of the antibodies which recognize the host tissue
must be genetically coded in to the organism. But, as the organism gets its genes from both mother
and father, the genes from the mother would lead to creation of antibodies that would recognize the
tissue from the father as foreign material, and vice versa. Hence, plan (2), which is regulation, is more
suitable. One means of regulating these antibodies, as suggested by Jerne [1], is called the idiotypic
network theory. This network type will be discussed in detail in section 2.5.
Antibodies are generated by special cells known as B-lymphocytes. Each B-lymphocyte has
somewhere around 1 million antibodies attached to its surface with identical paratopes that serve as
sensors to detect the presence of an epitope that this antibody type can respond to. When the
appropriate epitope is detected, the lymphocyte is stimulated to reproduce more lymphocytes (i.e. to
clone) and also to secrete free antibodies. This process of amplifying only those antibody types is
called Clonal selection.
2.3 Reinforcement learning and immune memory
To defend the body better, just recognizing the antigen is not sufficient; the immune system must also
have enough resources to attack the antigens better. The size of a certain lymphocyte population
specific to the antigen in hand with respect to the overall population of that antigen in the body is what
determines the final outcome of the battle between antigens and antibodies and eradication of the
infection. The immune system learning corresponds to the process of proliferation of the lymphocytes
within the body and their affinity towards the antigen that they have proven to be effective against. As
the overall population of all lymphocytes in the body is regulated, the growth in the concentration of one
type of lymphocytes means the reduction in the concentration of some other lymphocyte’s population. If
the immune system learns only by increasing the population sizes of specific lymphocytes, it must
either ”forget” previously learned antigens, increase in size, or constantly decrease the portion of its
force that is generated at random and responsible for responding to antigens that the system has not
encountered in the past.
If the immune system is evolving normally, then it is expected to encounter the same antigen
repeatedly throughout its lifetime. As soon as the immune system encounters an antigen for the first
time, it uses a small range cloned B-cells, each of which produce different antibodies having different
affinities towards the antigen and towards each other. How well the immune system responds to the
same antigen if produced a second time depends on whether it stores the high affinity B-cells from the
first encounter so as to be able to be ready with a large initial clone for encounters in the future.
Instead of starting over every time an antigen is encountered, the storing of high affinity antibodies will
lead to a much faster and efficient response of the immune syst1em which will increase exponentially
with each attack. This method is known as the reinforcement learning strategy, where the system is
continuously improving its capability to perform its task. To further demonstrate the immune response
(memory), consider that an antigen A is introduced at time zero and it finds antibodies that react with it
inside the body. After a lag phase, the antibody against antigen A appears and its concentration rises
up to a certain level, and then starts to decline (primary response). When another antigen B is
introduced, no antibody is present, showing the specificity of the antibody response.

FIGURE 2.3: Primary, Secondary and Cross Reactive Response to Immune systems [3]
On the other hand, one important characteristic of the immune memory is that it is associative: B-cells
adapted to a certain type of antigen, A presents a faster and more efficient secondary response not
only to A, but also to any structurally related antigen B. This phenomenon is called immunological
cross-reaction, or cross-reactive response. This form of memory is within the process of vaccination
and is called the generalization capability, or simply generalization, in other artificial intelligence fields,
like neural networks. Figure 2.3 illustrates primary, secondary and cross-reactive responses. After an
antigen has been seen once (primary response), further encounters with the same antigen, will lead to
a faster response.
2.4 Co-operation between B-cells and T-cells during immune response
B-cells are born within the bone marrow of a vertebrate body and are developed with stem cells. They
are capable of reproducing themselves and also able to clone themselves. Development of B-cells
starts with stem cells in the bone marrow. These clones pass through pro-B and pre-B stages to
become fully developed B-cells. Stromal cells generate chemical signals that B-cells must receive to
develop successfully. The cells start a process of mixing various antibody genes within the B-cell to
generate new receptors. All of this occurs in the plasma cell which secretes the antibodies generated
by the B- Cell to attack the external pathogens. As each B-cell responds to only a specific antigen type,
a population of B-cells is capable of identifying a large number of antigens. On top of that, constant
genetic mutations going within the B-cells allow them to create millions of unique antibody types. In
addition, each developing cell can modify the gene-splicing sites to further increase variability in the
DNA encoding the antigen-binding site. The B-cells can diversify further by inserting new nucleotide
sequences at the joint between fragments as they splice them together.
T-cells are produced in the thymus by stem cells that have traveled from the bone marrow through the
blood stream. The maturing cells go through stages that can be distinguished by the proteins that they
show on their surfaces. The cells whose receptors attach to class I MHCs on adjacent cells become
helper T-cells, and the ones whose receptors attach to class II MHC molecules become killer T-cells.
The cells that do not bind to MHCs and go for other cells will be eradicated from the system. The
immune attack consists of two systems that are inter-related to each other: the innate immune system
and the adaptive immune system. The body is born with the capability to recognize certain pathogens
and eliminate them immediately. This ability is called the innate immune system. The innate immune
system in our body can destroy a number of antigens as soon as it encounters them without the need
of any learning. As the antibodies can recognize all type of cells, the body’s own cells can also become
a target if there does not exist a proper distinguishable factor between a pathogen ant the self cell. The
innate immune system is able to identify self cells from non/self cells and this ability aids the body in its
adaptive response. The innate immune system is quite important to the system as it also aids the co-
stimulatory signals within antigen presenting cells (APCs) that will lead to T-cell activation, promoting
the start of the adaptive immune response. Adaptive immune recognition without innate immune
recognition may result in the recognition of lymphocytes that carry receptors involved in the adaptive

recognition as non-self cells and make them prone to attack and elimination. The adaptive immune
system uses antigen receptors that are genetically mutated and distributed among two major types of
cells: B-cells and T-cells. These antigen receptors are generated by random processes and, as a
result, the adaptive immune system works based on the Clonal selection of lymphocytes that are
specific to unique antigens. The antibody molecules (Ab) play a leading role in the adaptive immune
system. These antibodies act as receptors and are used in the adaptive immune response. They are
formed by shuffling the existing pieces of gene segments that are available in the body. Each antibody
uses a different combination of gene slices and hence forms a uniquely coded receptor. Collectively,
antibodies can identify and attach to a great variety of antigens. Adaptive immunity allows the body to
respond to any pathogen, even if it has been encountered for the first time. There are two basic types
of immunities in the body, humoral immunity and cell-mediated immunity. B-cells and T-cells both play
an important part in both of these immunities. B-cells participate in humoral immunity where antibodies
are secreted by the process of Clonal proliferation. T- Cells on the other hand, take part in cell
mediated immunity.
Killer T-cells destroy infected and functionless cells whenever they are recognized. Helper T-cells
orchestrate the stimulation and suppression of antibody formation through Clonal selection. One class
of T-cells, called Killer T-cells, destroys the infected cells whenever they recognize the infection.
Lymphocytes roam around freely in blood, plasma cells and lymph nodes looking for foreign
pathogens, ingest them, create MHC compounds and come back to the lymphatic system to start the
process all over again. When the body is attacked by an infectious agent, the innate immune system is
activated first to see if the immune system already has a solution for the agent. Innate immunity is not
antigen specific; it acts the same way towards all encountered pathogens entering the body. It is called
non-specific immune response. Phagocytes are the most important cells in the innate immunity. They
include monocytes, macrophages, etc. They are the cells that actually neutralize external attack by
pathogens in the human body. Once the attack is over, they create protein peptides called MHC
making the phagocyte an antigen presenting cell (APC). APC extracts the features out of an antigen by
processing and presenting the peptides on surfaces of B-cells and T-cells. MHCs also help the immune
system to distinguish self molecules from non-self (pathogen) molecules. The lymphocytes become
sensitive to the antigens and are activated when in contact after which, the helper T-cells secrete
chemical signals, called cytokines, which trigger the B-cells to start dividing and producing antibodies.
B-cells activate based on recognition of the right antigen. Recognition happens by intermolecular
binding dependant on the shape of the receptor and the electrostatic charge it carries. The antibodies
are actually a soluble version of the receptors of the B-cells which enables them to be freely distributed
throughout the body. An antibody attaches to the antigen by means of paratope – epitope connection
according to their affinities. Moreover, B-cells are also affected by helper T-cells during the immune
responses. The helper T-cells also help the immune system in determining whether it should act using
cell mediated immunity (by Th1 helper T-cells) or humoral immunity (by Th2 helper T-cells), making the
overall system response connected and efficient. When a B-cell clone matures more and faster as a
response to the matching between the clone and the antigen, the affinity is matured as well. Those
mutant cells are bound more tightly and stimulated to divide more rapidly. This process of affinity
maturation dynamically balances recognition versus reaction in adaptive immunity.
2.5 Jerne’s Idiotypical Immune Network Theory
Jerne had proposed the idiotypic network hypothesis based on the mutual stimulation and suppression
between the antibodies [1]. Because antibodies, like other molecules, have epitopes they are capable
of being recognized by other antibodies. Different antibody types can be distinguished and regulated
just as though they were antigens. An antibody type is thought to be “enhanced” when its paratopes
recognize other types, and “suppressed” if its epitopes are recognized. Even if self-destructive antibody
types are produced, their numbers can be regulated by other antibodies which recognize them and
cause their destruction.
There is experimental evidence indicating that such recognition processes can be extended to more
than one level, with A recognizing B, which recognizes C, etc., allowing for the possibility of complex
reaction networks forming.

FIGURE 2.4 Jerne's Idiotypic Network [1]
Jerne’s hypothesis [1] is modeled as a differential equation simulating the concentration of a set of
lymphocytes. The concept of immune network states that the network dynamically maintains the
memory using a feedback mechanism within the network. The lymphocytes are not isolated, but
communicate with each other among different species of lymphocytes through the interacting
antibodies. Each antibody recognizes only a very specific antigen that it can attach. This can be
visualized as a key fitting into a keyhole. The antibody recognizes the antigen using a part called the
epitope which is seen as the key and the corresponding part on the antibody, used to attach itself to the
antigen, is called the paratopes and is seen as the lock. Recent studies [17] in immunology have
shown that each type of antibody also has its specific antigen determinant, called the idiotope. Jerne
proposed a hypothesis [1] which he has called the ‘idiotypic network hypothesis’, sometimes called
‘immune network hypothesis’. The concept of this network hypothesis is that the antibodies are not
isolated, as a matter of fact they communicate to each other based on the species of the corresponding
antibody. The structure of the network is not constant, but varies continuously. It organizes itself
according to dynamic changes of environment. This function is called meta-dynamic function and is
mainly attained by adding newly generated antibodies in the system and removing the non functional
ones. The new cells are generated by both gene recombination in bone marrow and mutation in the
growth process of activated cells. Although many new cells are generated every day, most of them
have no effect on the existing network and soon die away without any stimulation. Due to such
enormous loss, the meta-dynamic function works to maintain an appropriate repertoire of cells so that
the system can cope with environmental changes. The meta-dynamic function would be expected to
provide feasible ideas to the engineering field as an emergent system. The dynamic equations of
idiotypic network proposed by Farmer [2] are applied to calculate concentration of antibodies. The
concentration of ith antibody, which is denoted by a (i), is calculated as follows: The dynamic equations
of idiotypic network proposed by Farmer are applied to calculate concentration of antibodies.
( ) ( ) ( ) ( )takmtamtamtA
dt
d
i
N
j
N
k
iikikjjii 







−+−= ∑ ∑
= =1 1
βαα
………2.1
Where, in equation 2.1, N is the number of antibodies, ai is the concentration of the ith antibody and α
and β, are positive constants. mji and mi denote affinities between antibody j and antibody i i.e. the
degree of interaction, and between the detected antigens and antibody i, respectively. The first and
second terms of the right hand side denote the stimulation and the suppression from other antibodies,
respectively. The third term represents the stimulation from the antigen, and the fourth term ki the
dissipation factor (i.e. natural death).
2.6 Negative Selection
The basic concept behind negative selection algorithms is the self-nonself problem within the immune
system. It is one of the purposes of the immune system to recognize all cells within the body and

categorize those cells as self or non-self. The non-self cells are further categorized in order to induce
an appropriate defense mechanism. The immune system learns through evolution to distinguish
between foreign antigens (e.g. bacteria, viruses, etc.) and the body’s own cell molecules. The purpose
of negative selection is to provide tolerance for self cells. During the generation of T-cells, receptors are
made through a pseudo- random genetic rearrangement process. Then, they undergo a censoring
process in the thymus, called negative selection. There, T-cells that react against self-proteins are
destroyed; thus, only those that do not bind to self-proteins are allowed to leave the thymus. These
matured T-cells then circulate throughout the body to perform immunological functions and protect the
body against foreign antigens.
The negative selection algorithm Forrest [27], is one of the computational models of self/non-self
discrimination, first was designed as a change detection method. It is one of the earliest AIS algorithms
that were applied in various real-world applications. Since it was first conceived, it has attracted many
AIS researchers and practitioners and has gone through some phenomenal evolution. In spite of
evolution and diversification of this method, the main characteristics as described by Forrest [27] still
persist.
FIGURE 2.5: Basic concept of the Negative selection Algorithm [27]
The major steps in such an algorithm, as shown in Fig.2.5 are:
• In the generation phase, the detectors are generated by some random process and censored by
trying to match self samples.
• Those candidates that match are eliminated and the rest are kept as detectors.
• In the detection stage, the collection of detectors (or detector set) is used to check whether an
incoming data instance is self or non-self.
• If it matches any detector, then it is claimed as non-self or anomaly.
This description is limited to some extent, but conveys the essential idea.
Two important aspects of a negative selection algorithm are: 1. The target concept of the algorithm is
the compliment of the self set; and 2.The goal is to discriminate between self and non-self patterns, but
only samples from one class are available (one-class learning).
3. HARDWARE AND SOFTWARE COMPONENTS
3.1 Hardware Setup
The hardware setup of our project is as follows:
The robot is attached via cable to a serial break-in box provided by the robot manufacturer.
FIGURE 3.1: Hardware Setup Block Diagram
Robot with
Sensors and
Camera.
Serial
Break-in
Power
Source
Computer for
Image
processing,
Command
Signal
Generation.
Easy CAP adapter
Video
Command
Command

International Journal of Robotics and Automation, Volume (1): Issue (1)
10
The serial break-in box separates the command signal from the video signal and also acts as a
continuous power source to the robot. The serial break-in box is connected to the computer by means
of the USB port for bidirectional communication. The robot sends sensor, encoder motor information
and the computer sends command signals. Video is extracted from the robot by means of a RCA cable
which in turn is connected to the EasyCAP video adapter and to the computer by means of the USB
2.0 port. The video signal is read by the computer and processed and that leads to the Webots
software generating control signals for the robots movements. A block diagram representation is shown
in figure 3.1.
3.2 Software Setup
The software flow in our project is as follows:
Sensor information is received from the robot and is sent for processing to the Webots Software. The
video signal is received through the watcher software and is converted to a
FIGURE 3.2: Software Flow Block diagram
usable format by 2Bitmap. The converted image is sent to webots for processing. All data is run
through the AIS algorithm and then control signals for robot movement are sent to the robot by webots
through the serial port. Hence the serial port is a medium for bidirectional communication between the
robot and the computer.
4. IMMUNE BASED INTELLIGENT ALGORITHM
4.1 Immune Networks and Negative selection based algorithm formulation
Our system is a mixture of Negative selection and Ab-Ab interactions. This system learns on basis of
both priori behavior and online adaptation mechanisms of the immune system. As the immune system
is made up of various cells that interact with each other to make the system function properly, the level
of interaction between these cells is what makes the system learn and adapt. In our system, the
interactions between B-cells and T-cells make the system robust and adaptable to any given
environment. The introduction of negative selection makes the system faster to react for certain
conditions which are defined as self conditions.
An antigen here is the environment perceived by the robot’s sensors and camera at any given time.
This provides information about the current location of the robot and the obstacles. As the antigen is a
direct derivative of the environment, the same representation can be used for any environment making
the system tolerant to environmental changes. As the antigen is a combination of various situations, it
can interact with various antibodies but only one antibody can bind with the antigen. This is similar to
the B-cell interactions as they interact with themselves and the antigens.
Initially, the system is initialized by using a random subset of 20 antibodies out of a total of 64 possible
antibodies to act on the system. As the model concerns itself with the variations of the antibody types,
the process is closer to a model of genetic changes that occur during cloning than it is to production of
new types in the bone marrow through reshuffling the gene library. Hence generation of new antibodies
by means of randomly picking out of the pool of available antibodies is a possible solution [1].
Robot
Webots running AIS
algorithm
Watcher 2Bitmap
Command signal and sensory information
Image
Acquisition
Format ConversionVideo Signal

11
The dynamic equations of the idiotypical system as proposed by Farmer [1] are applied to calculate the
concentration of the antibodies. The concentration of the i-th antibody denoted by a(i) is calculated as,
( ) ( ) ( ) ( )takmtamtamtA
dt
d
i
N
j
N
k
iikikjjii 







−+−= ∑ ∑
= =1 1
βαα
(4.1)
( )
( )( )tA
ta
i
i
−+
=+
5.0exp1
1
1
(4.2)
Where, in equation 4.1, N is the number of antibodies, ai is the stimulus of the ith antibody, Ai is the
concentration of the ith antibody and α and β, are positive constants. mji and mi denote affinities
between antibody j and antibody i (i.e. the degree of interaction), and between the detected antigens
and antibody i, respectively. The first and second terms of the right hand side denote the stimulation
and the suppression from other antibodies, respectively. The third term represents the stimulation from
the antigen, and the fourth term ki the dissipation factor (i.e. natural death). Equation 4.2 is a standard
squashing function to make sure that values of concentration stay within 0 to 1.
Affinities are calculated using a basic matching function and a threshold value using the following
equation,
( )
( ) ( ) 0^,^
0^,0
00
0
>−−
≤−
=
∑∑
∑
nn
n
i
thresholdYXifthresholdYX
thresholdYXif
m
(4.3)
As per equation 4.3, mi is the affinity of the ith antibody to the agent. Threshold represents the
minimum number of bits required to match for the candidate to be eligible to interact with the antigen. X
and Y are the agents which are needed to be matched. A difference of summation of the bitwise XOR
and the threshold level shows the level of matching. In simple terms, the exact binary opposite of the
antigen will be the best match to the antigen.
The antibody with the highest concentration amongst the triggered one is selected and presented to the
critic. The critic checks as per the matching function if the presented solution is the one that fits the
problem best. If yes, then the system behaves according to the winner antibody, other wise the worst
antibodies are removed from the system and are replaced by new ones which have been inactive.
The system performance can be managed by means of feedback. A reward or a penalty can be factors
to train an antibody incrementally. At a given time step, if the antibody chosen to act has the highest
affinity, the critic will give a reward. On the other hand, if this condition is not satisfied, a penalty signal
is sent to the helper T-cells.
The concentration of the helper T-cells, CT varies as follows:
( )( )np
CT
*exp1
1
η−+
=
(4.4)
Where η is the growth factor and np is the number of penalties (np≥0). When no penalty signal occurs
(i.e. np=0), the value of CT is equal to 0.5. A reinforcement signal will be secreted to active adaptive
mechanism if CT is higher than an allowable threshold value δ (δ ≥ 0.5). Adaptation function is realized
by the B-T cell cooperation and modifying the affinity among antibodies with a certain learning rate
according to the following criterion.
( ) ( ) )0(01.0 ==≥ λλδ ElseThenCTIf
(4.5)
If a reward is acquired then the affinity of the winner towards other antibodies, mw,o in the subset is
increased according to equation 4.6

12
( )
( ) ( )( )tm
tm
ow
ow
,
,
15.0exp1
1
1
γ+−+
=+
(4.6)
When the same antigen is recognized next time, the correct response will be made efficiently as a
result of this procedure. The suppressive interaction is enhanced using the equation below when a
penalty is the result as per equation 4.7. As a result, other antibodies will be active when the same
antigen is encountered next time, mo,w.
( )
( ) ( )( )tm
tm
wo
wo
,
,
15.0exp1
1
1
γ+−+
=+
(4.7)
An added feature of both these equations is that they ensure that the values of antibody affinities stay
between 0 and 1. The network formed as a resultant of these interactions leads towards formation of a
memory that stores reactions of the system towards antigens. Unconnected relations between
sequences of antibody reactions within the system are stored in this memory. As the antibodies are
directly related to antigens which are actually combination of sensor outputs, the memory hence
formed is independent of environmental changes.
Another feature of this algorithm is the utilization of the negative selection algorithm. Negative selection
can be thought of as an immune alternative of the classical control system’s observer system. The
Negative selection algorithm uses a self set from which it constantly compares the input to the system.
If the system receives a self input, it is functioning normally and generates a standard output, but if the
input deviates from the normal, which is the self set, then the system is passed through a correctional
system to solve the problem and bring the system back to normal. Similarly, in our case, a self set is
initialized at the beginning of the algorithm. If the antigen matches the self set, then the innate memory
takes over and the antigen is directly assigned a solution antibody without the need of concentration
calculations and system dynamics, otherwise the system passes the antigen through the system
dynamics to obtain a solution of the problem and bring it back to normal self. This greatly reduces
computational requirements and makes the process a lot faster.
4.2 Antigen and Antibody Representation
4.2.1 Antigen Representation: The robot interacts with the environment based on its sensors. The
values that the sensors get is what the robot perceives of its situation at the given time step. The
Khepera robot has six sensors to the front and two sensors in the back as shown in figure 4.1.
FIGURE 4.1: The Khepera Robot Sensor arrangement

13
In our project, the rear sensors didn’t play a significant role in deciding the movements of the robot. So,
only the front six sensors are used to obtain information of the environment from the robot. Thus, the
antigen is decided to be a six bit binary string based on the sensor location.
FIGURE 4.2: Antigen Representation
As shown in figure 4.2, the antigen is a representation of 6 bits where each bit corresponds to a sensor.
If a sensor crosses the threshold value then the corresponding bit will be 1 or else 0. The names s0 –
s6 represent the actual sensor number assigned to that bit. Here, s0 is sensor 0; s1 is sensor 1 and so
on, on the khepera.
4.2.2 Antibody Representation
As the antigen is a 6 bit binary string number the overall number of possible combinations that can
occur are 64. Also, the antibody that best matches an antigen according to our formulation is the exact
binary bitwise NOT of the antigen string. Hence an antibody is also represented as a 6 bit binary string
and the system has an antibody set consisting of a total of 64 antibodies.
4.3 The Immune based Control Algorithm
In this section we describe the step by step formulation of our immune based algorithm:
1. Initialization: Initialize a network of immune cells. As we know that the antibodies will lie from a 0 to
63, we initialize a superset of antibodies. We initialize their initial concentrations, reset the robot.
After that we make a subset of 20 randomly picked antibodies. We define stimulation and
suppression between antibodies using a basic matching function. Also, we define a self condition
which in our case is when the front two sensors are not ON.
2. Population Loop: If base 2 is not found, do:
(i) Antigenic Recognition: The information from the sensors is collected and an antigen is
formed based on that information. The matching between the antigen and the subset of
random antibodies is calculated and affinities are assigned based on matching. Each antigen
triggers multiple antibodies but only one antibody perfectly matches the antigen.
(ii) Self-Non Self Check: The current antigen is matched if it belongs to the self set in which
case the innate memory takes over and assigns the system a standard solution and starts the
loop again. Otherwise, the system moves on to the next step.
(iii) Network Interactions: Based on the selected antibodies, we calculate the interactions
between them.
(iv) System Dynamics: The difference between sums of stimulation and suppression between
antibodies added to the affinity between the corresponding antibody minus the natural death
factor or the decay factor of inactive antibodies gives the overall stimulation to the system. The
product of this stimulation to the concentration corresponding antibody gives its rate of change
of concentration with respect to time. The antibody with the highest concentration is sent to the
critic who decides whether the antibody should be penalized or rewarded. Based on the critic’s
decision, the affinities are modified.
3. Feedback Loop: A penalty activated Helper T-cell is activated and its concentration is calculated at
each step. Adaptation function is realized by cooperation between network cells and T-cells by
modification of affinities between antibodies employing a suitable learning rate.
4. Cycle: Steps (2) and (3) are repeated until a convergence criterion is met.
5. RESULTS AND DISCUSSIONS
5.1 Experimental Setup
We implemented our algorithm on a Khepera II robot with a K2D-c NTSC analog camera top turret with
Webots as an interfacing environment. Webots was set in remote control mode for execution of the
compiled program onto the robot. Remote Control mode is a special mode in which the robot receives the
s5 s4 s3 s2 s1 s0

14
command signals from the computer and all the processing is done on the computer. As the video
processing takes place on the computer, it was required that the other processing occurs on the computer
so as to maintain synchronization between both processes. The program was coded in C and was
compiled in Webot’s built-in IDE which is a feature recently introduced on Webots. A USB port for the
EasyCAP video converter was required as digital video transfer requires high baud rates which are not
supported by a USB 1.0 port. The serial break-in box was attached through a USB port as well to the
computer and also connected to the robot. A power adapter was fixed to the serial break-in box to provide
Khepera continuous power as shown in figure 5.1.
FIGURE 5.1: The Experimental robotic setup
The arena used for the robot’s testing is a 100 x 100 cm grid. The size of robot is 6 cm in diameter and
the maximum sensing region of the robot is 15 cm.
5.2 System Parameters Setup
We had to manually setup some parameters of the system to ensure proper functioning. One of them
was the Infra Red sensor object detection threshold value. Through observations, we found out that a
value of 150 at all sensors made sure that the robot detected an object as soon as it comes within a
range of 1.5cm of the robot which was more than enough to keep the robot from bumping into an
obstacle because of a previous movement. The immune parameters defined were, natural death
coefficient ki = 0.01, learning rate γ = 0.01, and growth rate η = 10. Green color blocks, figure 5.2 are
considered as home base and blue color corner, figure 5.3 is considered as target base.
FIGURE 5.2: Home Base FIGURE 5.3: Target Base
5.3 Results for Traversing and Mapping Maze
Two different types of environments were used to test the capabilities of the controller in adapting,
learning and mapping of real life environments. One was the structured maze type environment which
represents indoor features and the random scattered environment which represents outdoor situations.
The robot was made to look for the target which is the blue corner and simultaneously map the
environments. The results are explained and discussed in the following subsections.

15
5.3.1 The Structured Maze environment
The structured, maze type environment, as shown in figure 5.4, was used to test the capabilities of the
controller in adapting, learning and mapping. This maze comprises of walls, and a fixed structure. This
map was used to simulate an indoor environment and to evaluate how the robot controller fares in such
an environment. Initially, the robot was kept facing towards the home base or the green blocks. As the
logging starts only when the robot comes across the home base, this was a necessary step. Moreover,
from a practical point of view, the robot will always start from home base.
FIGURE 5.4: Two views of the structured environment
The robot was made to go through two runs of the environment. At the beginning of each run the robot
was made to face the home base at different angles. This changed the overall path of the robot. Figure
5.5 shows the number of steps taken in both runs to find the target base.
FIGURE 5.5: Number of steps taken to find target base in structured maze
One run is supposed to be over for the robot if it starts from the home base and reaches the target
base.
Concentration plots of the 64 antibodies at each time step were recorded and the graphs show that
concentration increased for the antibodies that were used during exploration and the ones that got
penalty were slowly eradicated from the system. Figure 5.6 is the concentration plot of the first run and
figure 5.7 is the concentration plot of the second run.

16
FIGURE 5.6: Concentration plot of Antibodies for the first run in a structured maze
FIGURE 5.7: Concentration plot of Antibodies for second run in structured maze
Figure 5.8 is the T-cell concentration for the first run and figure 5.9 is the T-cell concentration for the
second run.
FIGURE 5.8: T-cell Concentration for first run in structured environment

17
FIGURE 5.9: T-cell concentration for second run in structured environment
5.3.2 Unstructured Environment
The unstructured environment, as shown in fig 5.10, was used to test the capabilities of the controller in
outdoor environments.
FIGURE 5.10: Two views of the unstructured environment
As seen in the figure 5.10, the environment comprises of the home base at one corner of the arena and
the target base is the exact diagonally opposite corner. The blocks were placed randomly without any
fixed plan. Once again the robot was made to start with it facing the home base.
Two runs of the environment were recorded. Figure 5.11 shows the number of steps taken in the
environment to find the target base during the first and second runs of the environment.
FIGURE 5.11: Number of steps taken to find target base in unstructured environment

18
FIGURE 5.12: Concentration plot of antibodies for first run in unstructured environment
Figure 5.12 and 5.13 show the change in concentration of the total 64 antibodies present in the system
with respect to time steps for the first and second runs of the environment respectively.
FIGURE 5.13: Concentration plot of antibodies for second run in unstructured environment
FIGURE 5.14: T-cell concentration for first run in unstructured environment
Figure 5.14 and 5.15 show the change in concentration of helper T-cells within the body with respect to
time.

19
FIGURE 5.15: T-cell concentration for second run in unstructured environment
It can be concluded from the results that our robot is able to explore large areas and search for target
base effectively in all environments due to the assistance of self-organized immune network. The
immune based controller has no planning of movements, only the selection of the next action. Based
on the immune memory, the affinity connects chains of the recognition-triggered antibodies. These
associated connections guide the robot to traverse through the maze, avoid obstacles and look for the
target base in different environments realistically. The robot is employed to learn useful characteristics
from every changed environment to survive. The learning process will transitorily stop whenever no
penalty is given by the critic. This means that we change the environment whenever in both simulation
and real time once the effect of feature selection becomes highly localized. The procedure is repeated
until a stopping criterion is satisfied.
The helper T-cell activation, as shown in figures 5.8, 5.9, 5.14 and 5.15, depends on whether there is a
specific interaction of penalty signals. That signal happens if an antigen cannot be recognized correctly.
Once activated, helper T-cells consequently trigger the activation of B-cells, by specific cell to cell co-
operation and affinity modification in the proposed mechanism.
5.3.3 Mapping
Mapping of the maze starts as soon as the robot sees the home base. In case the robot turns back and
sees the home base again, the log till then is deleted and a new log is started. The log has a simple
format and is calculated at each time step after the winner antibody is chosen to act over the system.
The log contains 4 numbers. Which are namely, angle turned, left wheel speed, right wheel speed,
color encountered. So, if the robot sees a wall and speeds of -4 rad/s and 4 rad/s are assigned to the
left and right wheels respectively, i.e. a turn of 35 degrees which was calculated from observations, the
system records:
35 -4 4 1
Where, 35, is the angle turned, -4 and 4 are the left and right wheel speeds and 1 stands for the color
of the wall. 2 and 3 stand for the colors of target base and home base respectively. In case the robot
doesn’t see anything, the color is taken as 0 and the wheel speeds and angle turned are recorded. As
far as the plotting is concerned, basic mobile robot Forward Velocity kinematics formulas are
employed. We know that the axle length of the robot is 0.053 m and the wheel radius is 0.008 m. As we
have the left wheel and right wheel velocities in terms of rad/s, a multiplication of velocities with the
wheel radius will give us the speed in m/s. Then using equations 5.1, 5.2 and 5.3, we can obtain the X
and Y coordinates of the robot along the local coordinates of the robot.
( )
( )
( )
( )



−
−
+
= lr
lr
lr
VV
l
t
VV
VVl
tx sin
2 (5.1)

20
( )
( )
( )
( )



−
−
+
= lr
lr
lr
VV
l
t
VV
VVl
ty cos
2 (5.2)
( )
l
VVt lr −
=θ
(5.3)
Where Vr and Vl are the right and left wheel velocities respectively, l is the axle length and θ is the
angle of rotation of the robot.
When both wheel velocities are the same, the robot moves forward. At this time, the angle of
movement remains zero. When both wheels are same in magnitude and opposite in signs, the robot
makes a zero radius turn. In this case the change in position is zero and the angle changes according
to the time the robot was allowed to move at constant velocity.
In our algorithm, we made sure that when the robot sees something to its front, then it looks for the
best direction to turn, i.e. there is no forward movement when the robot detects objects to its front.
Hence, wherever the robot takes a turn means that it encountered an object, otherwise it will move
forward.
FIGURE 5.16: Comparison of Arena 1 and the map obtained
The initial plot of the robot’s movement is processed further to extract the turn points and that’s how the
map of the region it traversed is obtained. Figure 5.16 shows the map obtained from the structured
maze, and figure 5.17 shows the map obtained from the unstructured environment.
FIGURE 5.17: Comparison of Arena 2 and the map obtained
These maps clearly show that the accuracy and clarity of the plot depends on how much of the
environment the robot has explored and how much time it spent in the environment before finding the
target location. Also, the amount of training the Ab-Ab network receives also decides the accuracy of
the plot.

21
This shows that the proposed learning algorithm is capable of traversing though an unknown
environment, able to adapt to its surroundings, explore large areas, decipher between various features
within the environment and reach the target while mapping the environment.
6. SIGNIFICANCE OF OUR RESEARCH CONTRIBUTIONS
A. Generalized architecture: The previous research on AIS applied to robotics [5] [8] [11] [16] used
fixed or pre-designed antibodies. These antibodies were conceived by the programmer and were then
related to the robot’s perception of the environment based on cases considered. This does not help in a
general architecture for designing an AIS controller for mobile robots employing proximity sensors to
perceive the environment around them. The representation of an antigen used in our implementation of
immune algorithm can be used as a general structure for mobile robots which employ proximity
sensors for navigation and mapping. In our implementation we made use of six sensors in the robot
which led to the antigen being a six bit binary string. If generalized, this method can be used to
implement any number of sensors as the antigen will be a combination of the proximity sensor
readouts. Hence for a mobile robot with “n” sensors, the resulting antigen will be an “n-bit” binary string.
Similarly, the antibodies can be within the range of 0 – 2
n
. This means that our proposed algorithm is
capable of being ported to a different mobile robot using any number of proximity sensors with very
minor changes to our code.
B. Random generation of antibodies in our proposed project: This method will make the AIS more
similar to the actual biological process and give way to new possibilities. Farmer et al. [2] in their paper
stated that antibody generation is more like genetic mutation where instead of creating new antibodies,
a set of antibodies are chosen from the existing gene pool, mixed and matched to form new
antibodies.. Also, the authors in [11] had considered a fixed set of antibodies, 13 in total. At each step
of the robot, all 13 antibodies would act on the system and the ones which got triggered were
compared for the winner to be chosen. This works fine if the number of antibodies is small. In case of a
larger number of antibodies, since the system makes computations for all the antibodies and then
produces a solution, this would not only eat up a lot of computational time but also will pose a problem
for implementation on some general microcontrollers due to memory limitations. In our implementation,
we have a superset of 64 antibodies out of which we randomly select 20 antibodies to act on the
system. This greatly reduces the computational load on the system that it would have had to face if it
would have had to process 64 antibodies every time. This makes the system a lot more similar to the
real immune system as well.
C. Real-time replacement of inactive antibodies: During execution, if the antibody with highest
concentration doesn’t meet the critic’s criterion, the worst antibodies of the 20 that are acting on the
system are removed and replaced by the same number of random antibodies and the system is run
through the dynamic equations of the system again. This is done till the critic’s criterion are met or a
limit to iterations is reached. In the latter case the next best option is allowed to act on the system. This
means that the system will deviate very little from the actual solution and approximately be on the right
track at all given times. This process has been theoretically discussed [18] but is being implemented for
the first time.
D. Combination of negative selection and Ab-Ab interactions: Our controller employs both negative
selection and Ab-Ab interactions to help the system learn, explore and map the environment. Negative
selection checks the system for anomaly or a deviation from the norm. As soon as the anomaly is
detected, the Ab-Ab network is employed to find a solution to the anomaly and bring the system back to
normal. When the system is in normal state, the antigen is presented with a default antibody solution
which is stored in the body’s innate memory hence saving the system computational time. In our
system, we have employed self condition as nothing in front of the robot and the default antibody as
move straight forward. When an anomaly is detected i.e. something is in front of the robot, the antigen
is sent through the Ab-Ab network to find a solution and bring the system back to normal. This helps in
mapping as well, as the system moves straight in self state and makes fixed angle turns based on the

22
anomaly antigen, a simple mapping technique can be used to plot the robot’s movements and
obstacles accordingly.
E. Machine vision integration: The integration of machine vision into our project enabled the robot to
look and decipher between significant and non-significant components of the maze which allows the
robot to adapt better in the given environment. The recognition of obstacles also allows the robot to set
priorities between the jobs at hand and future processes.
F. The results of our findings can be used in real-life applications such as rescue robots in natural and
man-made disasters where rescuers have limited access and control in finding survivors. The entrance
to affected area can be identified as the home base and the survivors as the target base. The robot can
navigate and look for target base and create a map. The log of the map is just a four-digit integer which
can easily (unlike transfer of video signal) be transmitted wireless. Other applications are in the
household robotics industry where the robot can initially map the entire house. Once the house is
mapped, the robot can utilize external Simultaneous Localization and Mapping (SLAM) algorithms to
retain only the key features of the map for further use.
7. CONCLUSIONS
We have designed an immune based controller for the Khepera II robot. We have created a mixture of
concepts of Negative Selection, self-non self recognition and Ab-Ab binding to come up with this
algorithm. Implementation of this controller was done in real time in three different environment
situations.
The robot had to first locate the home base and from there start mapping the whole environment till the
target base is found using both its proximity sensors and machine vision. During that it was supposed
to explore as much of the arena as possible and have a collision free navigation of the area while
performing straight motion. The dynamics of the immune system used to train the robot were designed
to achieve this behavior even if there were changes made to the environment. The results obtained
demonstrate the controller’s capability to adapt in different configurations of the environment. The
implementation of negative selection and immune network interactions together made the system to
easily map the environment and give out a log which was used to plot the environment as perceived by
the robot.
This controller was designed to focus on exploring the self-organizing, adaptive and learning properties
of the immune system. Adaptation and learning are realized by modifying the affinities between the
antibodies. Once learning is achieved, the affinity corresponds to long term memories until newer
antigen is met which starts the learning process all over again.
The base finding and mapping problem was employed to evaluate the performance of the immune
based controller. The mobile robot utilizing the immune based adaptive learning mechanism is capable
of browsing through a workspace and exploring unknown environments. In addition, the robot learns to
distinguish between useful and useless components of the environments and adapts itself with the
environment. Once the immune system has learned, it switches from immune response mode to
immune tolerant mode, the decision making strategy changes from short term learning to utilization of
the learned data. Hence, the proposed controller provides a principled way to organizing an intelligent
robotic system.
In the near future, the authors will use fuzzy logic techniques for the conversion of an antibody to wheel
speed in order to improve and bring a smoother, efficient and interpolated movement of the robot within
the maze. A drawback and challenge for this future proposed work is expected to be that the differential
drive system is already non-holonomic and curves in the path would make mapping and trace-back
even tougher.

23
8. REFERENCES
[1] N. K. Jerne, “The Immune System”. Scientific American, 229(1):52-60, 1973
[2] J. Doyne Farmer, Norman H. Packard and Alan S. Perelson, “The Immune System, Adaptation, and
Machine Learning”. Los Alamos National Laboratory, Los Alamos, NM 87545, USA.
[3] De Castro and Von Zuben, “Artificial Immune Systems: Part 1-Basic theory and applications”. TR-
DCA 01/09,
[4] R. Brooks, “Intelligence without reason”. Proc. of IJCAI, pp. 565-595, 1991
[5] Guan-Chun Luh and Wei-chong Cheng, “Behavior based intelligent mobile robot using an
immunized reinforcement learning mechanism”. AEI, Elsevier, 2002
[6] R. Brooks, ”A robust layered control system for a mobile robot”. IEEE Journal of R and A, 2(1):14-
23, 1986
[7] P. Maes, “The dynamic action selection”. Proc. of IJCAI, pp. 991-997, 1989
[8] Guan-Chun Luh, Wei-Wen Liu, “An immunological approach to mobile robot reactive navigation”.
Elsevier, Advanced Engineering Informatics, 2006
[9] http://en.wikipedia.org/wiki/Immune_system
[10] Forsdyke, D.R., ”The Origins of the Clonal Selection Theory of Immunity”. FASEB Journal
9(1):164-66, 1995
[11] S. Ozcelik, N. Mathur, R. Challoo, “Immune System Based Artificial Intelligence for a mobile robot”.
In Proceedings of the 2005 International Symposium on advanced Control of Industrial Processes
(AdCONIP’05), Seoul, Korea, 2005
[12] D. Dasgupta, S. Forrest, “Artificial Immune Systems in Industrial Applications”. 1996
[13] Guan-Chun Luh, Wei-Chong Cheng, “Behavior -based intelligent mobile robot using an immunized
reinforcement adaptive learning mechanism”. Elsevier, Advanced Engineering Informatics, pp.85-98,
2002
[14] Dong Hwa Kim, “Parameter Tuning Of Fuzzy Neural By Immune Algorithm”. Fuzzy Systems,
FUZZ-IEEE'02, 2002
[15] A.M. Whitbrook, U.Aickelin, J.M. Garibaldi, “Idiotypic Immune Networks in Mobile Robot Control”.
IEEE Systems, Machine Intelligence and Cybernetics, 2006
[16] A. Ishiguro, T. Kondo, Y. Watanabe, Y. Uchikawa, “Dynamic Behaviour Arbitration of Autonomous
Mobile Robots Using Immune Networks”. In Proceedings of IEEE International Conference on
Evolutionary Computation, 1995
[17] Dipankar Dasgupta, “Advances in Artificial Immune Systems”. IEEE Computational Intelligence
Magazine, pp.40-49, November 2006
[18] John E. Hunt, Denise E. Cook, “Learning Using an Artificial Immune System”, Journal of Network
and Computer Applications 19:189-212, 1996

24
[19] A. Somayaji, S. Hofmeyr, and S. Forrest, “Principles of a Computer Immune System”. New
Security Paradigms Workshop, pp. 75–82, 1998
[20] S. Hofmeyr and S. Forrest, “Immunity by Design: An Artificial Immune System”. In Proceedings of
GECCO Conference, 1999
[21] A. Hofmeyr, A. Somayaji, and S. Forrest, “Intrusion Detection using Sequences of System Calls”.
Journal of Computer Security, 6:151–180, 1998
[22] C. Warrender, S. Forrest, and B. Pearlmutter, “Detecting intrusions using system calls: Alternative
data models”. IEEE Symposium on Security and Privacy, 1999
[23] J. Kim and P. Bentley, “The human Immune system and Network Intrusion Detection”. In
Proceedings of the 7th European Congress on Intelligent techniques—Soft Computing (EUFIT’99),
Aachan. Germany, Sept. 13–19, 1999
[24] S. Forrest and M.R. Glickman, “Revisiting LISYS: Parameters and Normal behavior”. In
Proceedings of the Special Track on Artificial Immune Systems, WCCI.-CEC. May 2002
[25] J. Timmis, M. Neal, and J. Hunt, “An Artificial Immune System for Data Analysis”. In Proceedings
of the International Workshop on Intelligent Processing in Cells and Tissues (IPCAT), 1999
[26] F. Gonzalez and D. Dasgupta, “An Immunogenetic Technique to Detect Anomalies in
Network Traffic”. In Proceedings of the International Conference Genetic and Evolutionary Computation
(GECCO), New York, July 9–13, 2002.
[27] S. Forrest, A.S. Perelson, L. Allen, and R. Cherukuri, “Self-Nonself Discrimination in a Computer”.
In Proceedings of the 1994 IEEE Symposium on Research in Security and Privacy, Los Alamitos, CA:
IEEE Computer Society Press, 1994.
[28] S.A. Hofmeyr and S. Forrest, “Architecture for an Artificial Immune System”. Evolutionary
Computation, 8(4):443–473, 2000
[29] Kim and W.O. Wilson, Uwe Aickelin, and Julie McLeod, “Cooperative Automated
Worm Response and Detection ImmuNe ALgorithm (CARDINAL) Inspired by T-cell Immunity and
Tolerance”. In Proceedings of the Fourth International Conference, ICARIS 2005 on Artificial Immune
Systems, Banff, Alberta, Canada, pp. 168–181, August 2005
[30] U. Aicklen, P. Bentley, S. Cayzer, J. Kim, and J. McLeod, “Danger Theory: The link
between ais and ids”. In Proceedings of the Third International Conference, ICARIS 2003 on Artificial
Immune Systems, Edinburgh, UK, pp. 156–167, 2003
[31] H. Bersini and F. Varela, “Hints for Adaptive Problem Solving Gleaned from Immune Network”. In
Parallel Problem Solving from Nature, pp. 343–354, 1990
[32] Y. Ishida, “Fully Distributed Diagnosis by PDP Learning Algorithm: Towards Immune Network PDP
Model”. Proceedings of International Joint Conference on Neural Networks, San Diego, pp. 777–782,
1990
[33] Y. Ishida, “Distributed and Autonomous Sensing Based on Immune Network”. In Proceedings of
Artificial Life and Robotics, Beppu. A.A.A.I. Press, pp. 214–217, 1996

25
[34] D. Dasgupta, K. KrishnaKumar, D. Wong, and M. Berry, “Negative Selection Algorithm for Aircraft
Fault Detection”. In Proceedings of the Third International Conference, ICARIS 2004 on Artificial
Immune Systems, Catania, Sicily, Italy, Sept., 2004
[35] J. Hunt, J. Timmis, D. Cooke, M. Neal, and C. King, “The development of an Artificial Immune
System for real world applications”. Applications of Artificial Immune Systems, pp. 157–186, 1999
[36] J. Kim and P. Bentley, “Toward an Artificial Immune System for Network Intrusion Detection: An
Investigation of Dynamic Clonal Selection”. In Proceedings of the 2002 Congress on Evolutionary
Computation (CEC 2002), Honolulu, Hawaii, pp. 1244–1252, 2002.
[37] D. Dasgupta, “Immune-based Intrusion Detection System: A General Framework”. In Proceedings
of the 22nd National Information Systems Security Conference (NISSC), 1999.
[38] P.D. Williams, K.P. Anchor, J.L. Bebo, G.H. Gunsh, and G.D. Lamont, “Cdis: Towards a Computer
Immune System for Detecting Network Intrusions”. Lecture notes in Computer Science, vol. 2212, pp.
117–133, 2001.
[39] http://science.kukuchew.com
[40] http://www.irvingcrowley.com
[41] http://www.k-team.com
[42] http://www.cyberbotics.com
[43] http://www.library.thinkquest.org

Navigation Control and Path Mapping of a Mobile Robot using Artificial Immune Systems

More Related Content

Viewers also liked (15)

Similar to Navigation Control and Path Mapping of a Mobile Robot using Artificial Immune Systems (20)

More from Waqas Tariq (20)

Recently uploaded (20)

Navigation Control and Path Mapping of a Mobile Robot using Artificial Immune Systems