Download Towards a theory of Hybrid Intelligent Autonomous Systems

Recent Researches in Circuits and Systems
Towards a theory of Hybrid Intelligent Autonomous Systems
OUARDA HACHOUR
Institute Of Electrical and Electronics Engineering
Signal and System Laboratory
Boumerdes University
Boulevard de l’independence Boumerdes 35000
ALGERIA
E-mail: [email protected]
Abstract: Today, researchers have at their disposal, the required hardware, software, and sensor technologies to build
intelligent autonomous systems. More, they are also in possession of some computational tool such as Fuzzy
Logic(FL), Neural Networks (NN), Expert system(ES), Genetic Algorithms (GA) and other more technologies that
are more effective in the design and development of intelligent autonomous systems than the predicate logic based
methods of traditional Artificial Intelligence. FL, ES, NN, and GA are well established as useful technologies that
complement each other in powerful hybrid system for example. Hybrid intelligent systems are now part of the
repertoire of computer systems developers and important research mechanisms in the study of Artificial Intelligence.
The integration of these technologies has proven to be a way to develop useful real-world applications, and hybrid
systems involving robust adaptive. More, these theories and applications provide a source linking all fields in which
intelligent control plays a dominant role. Cognition, perception, action, and learning are es²sential components of suchsystems and their use is tending extensively towards challenging applications.
Key-Words: - Expert System (ES), Fuzzy Logic( FL), Neural Networks (NN), Hybrid Intelligent Systems, Autonomous
Systems.
Most of the difficulties in this process originate in the
nature of the real world: unstructured environments and
inherent large uncertainties. First, any prior knowledge
about the environment is, in general, incomplete,
uncertain, and approximate.
The robot may have some perception capabilities,
reactive behaviors, and perhaps limited reasoning
abilities that allow it to handle an unstructured and
dynamic environment. But the user supplies the highlevel and difficult reasoning and strategic planning
capabilities.
In the context of autonomy, the main work here deals
with the navigation principle which is one of the most
vital aspect of an autonomous robot. In most practical
situations, the mobile robot cannot take the most direct
path from start to the goal point. So, path finding
techniques must be developed in these situations, and
the simplest kinds of planning mission involve going
from the start point to the goal point while minimizing
some cost such as time spent, chance of detection, etc.
Mobile Robot Navigation" covers a large spectrum of
different systems. requirements and solutions. Artificial
intelligence, including action, reaction, manipulating,
sensing, understanding, recognition and deciding, has
been actively studied and applied to domains such as
automatically control of complex systems like robot.
In fact, recognition, learning, decision-making, and
action constitute the principal obstacle avoidance
1 Introduction
The autonomous robot navigation problem has been
studied thoroughly by the robotics research community
over the last years. The basic feature of an autonomous
mobile robot is its capability to operate independently in
unknown or partially known environments.
Autonomous robots which work without interventions
are required in robotic fields. In order to achieve tasks,
autonomous robots have to be intelligent and should
decide their own action. When the autonomous robot
decides its action, it is necessary to plan optimally
depending on their tasks. More, when a robot moves
from a point to a target point in its given environment, it
is necessary to plan an optimal or feasible path avoiding
obstacles in its way and answer to some criterion of
autonomy requirements such as : thermal, energy, time,
and safety for example [8,9,11].
The autonomy implies that the robot is capable of
reacting to static obstacles and unpredictable dynamic
events that may impede the successful execution of a
task. To achieve this level of robustness, methods need
to be developed to provide solutions to localization, map
building, planning and control.
The robot needs the capability to build a map of the
environment, which is essentially a repetitive process of
moving to a new position, sensing the environment,
updating the map, and planning subsequent motion.
ISBN: 978-1-61804-108-1
146
Recent Researches in Circuits and Systems
problems, so it is interesting to replace the classical
approaches by technical approaches based on intelligent
computing technologies.
When the robot actually starts to travel along a
planned path, it may find that there are obstacles along
the path hence the robot must avoid these obstacles and
plans a new path to achieve the task of navigation.
The robot has to find a collision-free trajectory
between the starting configuration and the goal
configuration in a static or dynamic environment
containing some obstacles. Moreover, when a robot
moves in a specific space, it is necessary to select a most
reasonable path so as to avoid collisions with obstacles.
Several approaches for path planning exist for mobile
robots, whose suitability depends on a particular
problem in an application. The goal of the navigation
process of mobile robots is to move the robot to a named
place in a known, unknown or partially known
environment.
One of the specific characteristics of mobile robots is
the complexity of their environment, therefore, one of
the critical problem for the mobile robots is path finding.
Several approaches for path finding exist for mobile
robots, whose suitability depends on a particular
problem in an application.
Path planning in spatial representation for example
often requires the integration of several approaches. This
can provide efficient, accurate, and consist navigation of
a mobile robot. It is sufficient for the robot to use a
topological map that represents only the areas of
navigation (free areas, occupied areas of obstacles). It is
essential the robot has the ability to build and uses
models of its environment that enable it to understand
the environment’s structure. This is necessary to
understand orders, plan and execute paths [6,8,9].
For example, maps typically omit some details and
temporary features; also, spatial relations between
objects may have changed since the map was built.
Second, perceptually acquired information is usually
unreliable. Third, a real-world environment typically has
complex and unpredictable dynamics: objects can move,
other agents can modify the environment, and apparently
stable features may change with time. Finally, the effects
of control actions are not completely reliable, e.g. the
wheels of a mobile robot may slip, resulting in
accumulated zoometric errors.
Robot navigation can be defined as the combination of
three basic activities:
• Localization: this is the process of getting the actual
robot’s location from sensor readings and the most
recent map. An accurate map and reliable sensors are
crucial to achieving good localization.
• Path planning: This is the process of generating a
feasible and safe trajectory from the current robot
location to a goal based on the current map. In this case,
it is also very important to have an accurate map and a
reliable localization.
To evaluate the performances of systems one must
answer to all factors to be embedded with robot when it
executes its mission, this is summarized in how to
perform all tasks, such as intelligence and autonomy
requirements.
This paper draws together, and builds upon, a lot of
what is written in the domain of robotic field (navigation
process) and articles in Mobile Robot Navigation. This
paper deals with a review of some principles proposed
to an intelligent autonomous navigation process. This
paper deals with some computational tool such as Fuzzy
Logic(FL), Neural Networks (NN), Expert system(ES),
Genetic Algorithms (GA) and other more technologies
that are more effective in the design and development of
intelligent autonomous systems than the predicate logic
based methods of traditional Artificial Intelligence. FL,
ES, NN, and GA are well established as useful
technologies that complement each other in powerful
hybrid system for example.
Hybrid intelligent systems are now part of the
repertoire of computer systems developers and important
research mechanisms in the study of Artificial
Intelligence. The aim of this paper is to propose a plate
form of some definitions, basic rules and more some
principles which can be useful in some research papers.
In this context and to deal with the principle, we propose
our review of some factors before starting programs or
realization. This review is very useful to understand the
repertoire of hybrid intelligent systems.
2 Physical Scales
The physical scale of a device's navigation requirements
can be measured by the accuracy to which the mobile
robot needs to navigate - this is the resolution of
navigation. These requirements vary greatly with
application, however a first order approximation of the
accuracy required can by taken from the dimensions of
the vehicle it self. Any autonomous device must be able
to determine its position to a resolution within at least its
own dimensions, in order to be able to navigate and
interact with its environment correctly.
At the small end of the scale there are robots just a
few centimeters in size, which will require high
precision navigation over a small range (due to energy
supply constraints), while operating in a relatively tame
• Map building: this is the process of constructing a map
from sensor readings taken at different robot locations.
The correct treatment of sensor data and the reliable
localization of the robot are fundamental in the mapbuilding process.
ISBN: 978-1-61804-108-1
147
Recent Researches in Circuits and Systems
environment. To help in categorizing this scale of
requirements, we use three concepts:
4 The Hybrid Intelligent Systems
Today, researchers have at their disposal, the required
hardware, software, and sensor technologies to build
Intelligent autonomous systems . More, they are also in
possession of some computational tool such as Fuzzy
Logic(FL), Neural Networks (NN), Expert system(ES),
Genetic Algorithms (GA) and other more technologies
that are more effective in the design and development of
intelligent autonomous systems than the predicate logic
based methods of traditional Artificial Intelligence.
Fuzzy Logic FL and Expert System ES are well
established as useful technologies that complement each
other in powerful hybrid system for example.
Hybrid intelligent systems are now part of the
repertoire of computer systems developers and important
research mechanisms in the study of Artificial
Intelligent. The integration of ES and FL has proven to
be a way to develop useful real-world applications, and
hybrid systems involving robust adaptive.
2.1 Global navigation
Global navigation is the ability to determine one's
position in absolute or map-referenced terms, and to
move to a desired destination point.
2.2 Local navigation, the ability to determine one's
position relative to objects (stationary or moving) in the
environment, and to interact with them correctly.
2.3 Personal navigation, which involves being aware
of the positioning of the various parts that make up
oneself, in relation to each other and in handling objects
3 Autonomous mobile robots
Autonomous robots are robots which can perform
desired tasks in unstructured Environment without
continuous human guidance. Many kinds of robots can
be autonomous. In different ways, a high degree of
autonomy is particularly desirable in fields such as space
exploration, cleaning floors, moving lawns, and waste
water treatment. .
Some modern factory robots are “autonomous “
within the strict confines of their direct environment .it
may not be that every degree of freedom exists in their
surrounding environment but the factory robots
workplace is challenging and can often contain chaotic
,unpredicted variables. The exact orientation and
position of the next object and the required task must be
determined. This can vary unpredictably at least from the
robots point of view. One important area of robotics
research is to enable the robot to cope with its
environment whether this is on land, underwater, in the
air, underground, or in space for example [1,2,3,4].
A robot is a "device" that responds to sensory input by
running a program automatically without human
intervention. Typically, a robot is endowed with some
artificial intelligence so that it can react to different
situations it may encounter. The robot is referred to be
all bodies that are modelled geometrically and are
controllable via a motion plan.
A robotic vehicle is an intelligent mobile machine
capable of autonomous operations in structured and
unstructured environment. It must be capable of sensing
thinking and acting. The mobile robot is an appropriate
tool for investigating optional artificial intelligence
problems relating to world understanding and taking a
suitable action, such as , planning missions, avoiding
obstacles, and fusing data from many sources.
ISBN: 978-1-61804-108-1
4.1 Expert System
An ES is a computer program that functions, is in a
narrow domain, dealing with specialized knowledge,
generally possessed by human experts. ES is able to draw
conclusions without seeing all possible information and
capable of directing the acquisition of new information in
an efficient manner
4.2 Robot Control
Traditionally, motion planning and control have been
separate fields within robotics. However, this historical
distinction is at best arbitrary and at worst harmful to the
development of practically successful algorithms for
generating robotic motion. It is more useful to see
planning and control as existing on the same continuum.
In assistive robotics, a manipulator arm constitutes
one possible solution for restoring some manipulation
functions to victims of upper limb disabilities. The aim
of research work is to present a global strategy of
approach of an assistive mobile manipulator
(manipulator arm mounted on a mobile base). A
manipulability criterion is defined to deal with the
redundancy of the system. The aim is to keep the arm
manipulable, i.e. capable of moving by itself. The
strategy is based on human-like behaviour to help the
disabled operator to understand the action of the robot.
When the robot is far from its objective only the
mobile base moves; thus avoiding obstacles if necessary.
When the objective is close to the robot, both mobile
base and arm move and redundancy can be used to
maximise a manipulability criterion. The partial results
obtained with the real robot consolidate the results of
simulation. The work does not propose an autonomous
148
Recent Researches in Circuits and Systems
path planning and navigation of the mobile arm but
assistance to the user for remote controlling it.
The control task becomes more complex when the
configuration of obstacles is not known a priori. The
most popular control methods for such systems are based
on reactive local navigation schemes that tightly couple
the robot actions to the sensor information. Because of
the environmental uncertainties, fuzzy behaviour
systems have been proposed by researchers. The most
difficult problem in applying fuzzy-reactive-behaviourbased navigation control systems is that of arbitrating or
fusing the reactions of the individual behaviours, which
is addressed in this work by the use of preference logic.
Fuzzy Logic
To build intelligent systems that are able to perform
complex requiring massively parallel computation, a
knowledge of the environment structure and interacting
with it involves abstract appreciation of natural concepts
related to, the proximity, degree of danger, etc. the
implied natural language is represented through fuzzy
sets involving classes with gradually varying transition
boundaries. As human reasoning is not based on the
classical two-valued logic, this process involves fuzzy
truths, fuzzy deduction rules, etc. This is the reason why
FL is closer to human thinking and natural language than
classical logic.
Fuzzy models can obviously be made to work very well
indeed. The big advantage of a fuzzy model is that it is
relatively simple to construct and is in itself a simple
structure. It does not require the modeller to have a deep
mathematical insight, but relies more on intuition and
experience of the process. Its greatest value must be,
therefore, in those areas where such qualitative process
knowledge is predominant and essential for
understanding. As an example of fuzzy model, the figure
1 presents one model of navigation of an autonomous
mobile robots ( see the reference [7] for more details) :
A1
D1
FL
Model
Df
Af
D2
Fig.1: Fuzzy model
Where:
Ai: the direction of the robot. Di: intermediate distance
(see the figure2).
A1: the direction of the robot.D1, D2: intermediate distance
between current position, intermediate position and visual
point (see the figure2).
The direction A1 is calculated by :
V Visual
point
Intermediate
I position
C
Current
position
obstacle
Y
X
Fig.2 : Robot obstacle mode avoidance
Defuzzification is the output of the fuzzy system, it is
a decision-making logic ( written in a formula) adopted
for the compute of the real value of the output. The final
decision (defuzzification) is achieved to give the output
of fuzzy controls and to converts the fuzzy output value
produced by rules. The system must decide how to give
the right output using fuzzy logic from a fuzzy linguistic
formulation. The generation of rules operates itself
according to the distribution of the training whole in
fuzzy
linguistic
terms.
The
final
decision
(deffuzification) is accomplished to convert input of the
fuzzy system after treatments with the inference rules.
The final decision (defuzzification) is achieved to
give the output of fuzzy controls and to converts the
fuzzy given here for our example by:
G = (the sum of (ui * gi ) / the sum of (ui) )
A1= tan -1 (Yg-Yi)/(Xg-Xi)
Where : 1<=i<=m, m : number of rule, g: centroid of the
Where
ISBN: 978-1-61804-108-1
The P1(x1,y1), Pi(xi, yi), and Pg (xg, yg) are the
coordinates of respectively to initial point , intermediate
and visual point ( we calculate point to point until the
visual point become the target one). The vehicle must
learn to decided Af and Df using FL from a fuzzy
linguistic formulation of human expert knowledge. This
FL is trained to capture the fuzzy linguistic formulation
of this expert knowledge is used and a set of rules are
then established in the fuzzy rule « if A then B », for
more details, see the reference [5,7]. The expert system
represents a good part of activities of the Artificial
intelligence that makes call to knowledge on the domain
treaty, these systems are capable to reach human expert
performances for various types of tasks (diagnosis,
conception in restraint domains). As an example here,
the principle of the technique consists in verifying for
every degree membership function a whole of rules, or
each rule is the shape: IF <cond> THEN <name of the
stain>, Where <cond> is a combination of predicates
translating the spatial relations between the primitive of
the unknown shape (if the logic used by the ES is the one
of predicates).
149
Recent Researches in Circuits and Systems
unsupervised manner, i.e., with no “ supervisor or
teacher” required.
backend
membership function correspond
for each rule. U: factor of membership correspond for
each rule.
This intelligent task uses the fuzzy linguistic terms
and calculates for each degree of membership functions
under expertise of an expert system ES. An ES is a
computer program that functions, is in a narrow domain,
dealing with specialized knowledge, generally possessed
by human experts
4.3
4.4 Genetic Algorithms (GA)
Neural Networks
Neural Networks deal with cognitive tasks such as
learning, adaptation generalization and they are well
appropriate when knowledge based systems are involved.
In general Neural Networks deal with cognitive tasks
such as learning, adaptation generalization and they are
well appropriate when knowledge based systems are
involved. The adaptation is largely related to the learning
capacity since the network is able to take into account
and respond to new constraints and data related to the
external environments. Just as human being, a neural
network relies on previously solved examples to build a
system of “neurons” that makes new decisions,
classification and forecasts.
In designing a Neural Networks navigation approach,
the ability of learning must provide robots with capacities
to successfully navigate in the environments like our
proposed work maze environment, see the references
[12,13]. Also, robots must learn during the navigation
process, build a map representing the knowledge from
sensors, update this one and use it for intelligently
planning and controlling the navigation.
The general structure of the proposed Neural Networks
navigation is presented as follow:
Knowledge mapping: the model of the external
environment plays an important role in the intelligent
robot behavior. The human brain is able to create -simple
maps of the external environment by compressing the
huge amount of received sensory data, while preserving
the relationships between important facts.
Action: the different map sensory informations are
classified in several vectors where each component
responds to a particular situation. These situations must
be associated with the appropriate action taking
advantage of the topology preserving property of the
network.
Crossover
Parent
Point
Children
Figure 1 : Example of Crossover on
single point
Reinforcement learning: reinforcement learning allows
associations between detected sensory situations and
appropriate actions trough “ trial and error” learning. This
one uses only a priori knowledge such as “asked
response” is executed. These associations are formed in
ISBN: 978-1-61804-108-1
GAs are search algorithms based on the mechanics of
natural genetics. A genetic algorithm for an optimization
problem consists of two major components. First, GA
maintains a population of individual corresponds to a
candidate solution and the population is a collection of
such potential solutions. In GA, an individual is
commonly represented by a binary string the mapping
between solutions and binary strings is called a “coding”.
GA has been theoretically and empirically proven to
provide robust search capabilities in complex spaces
offering a valid approach to problems requiring efficient
and effective searching .
Before the GA search starts,
candidates of solution are represented as binary bit
strings and are prepared. This is called a population. A
candidate is called a chromosome (in the case of an
autonomous mobile robot: the path is a “chromosome”
and positions are the “genes”).
Also, an evolution function, called fitness function,
needs to be defined for a problem to be solved in order to
evaluate chromosome. As fitness function, we should
define distance for each chromosome to give an
evaluation function.
This evaluation is the goal of the GA search and goes
as follows: two (02) chromosomes are chosen randomly
from populations are mated and they go through
operations like the crossover to yield better chromosomes
for next generations. To determine execution of the GA,
we must specify a stopping criterion, in our case; it could
be determined, by a probabilistic function: as we have
four chromosomes and we choose randomly two
chromosomes, to combine and to compare one path with
itself. The crossover is the comparison operator (see
Fig.1).
Therefore, after several generations of GA search ( The
problem of mutation, see Fig.2), relatively low fitness of
chromosomes remain in a population and some of them
are chosen as the solution of the problem (the most
preferable path).ore detail, see the reference [10].
150
Recent Researches in Circuits and Systems
2002, pp. 113-126.
Origin chromosom
New Chromosome
1
0
1
1
1
1
[3]A.KRUPA and F.CHAUMETTE, Guidance of an
ultrasound probe by visual servoing, Advanced
Robotics,2006, Vol. 20, No. 11, pp. 1203–1218 .
[4]S.Florczyk, Robot VisionVideo-based Indoor
Exploration with Autonomous and Mobile Robots,
WILEY-VCH Verlag GmbH & Co. KGaA,
Weinheim, 2005.
[5] O.Hachour and N.Mastorakis, IAV : A VHDL
methodology for FPGA implementation, WSEAS
transaction on circuits and systems, Issue5,
Volume3,ISSN 1109-2734, pp.1091-1096.
[6] O.Hachour, Path planning of autonomous mobile
robot, International Journal Of System Applications,
Engineering
&
Developments,
issue4,
volume2,2008,pp.178-190.
[7] O.Hachour, The proposed Fuzzy Logic Navigation
approach of Autonomous Mobile robots in unknown
environments,
International
journal
of
mathematical models and methods in applied
sciences, Issue 3, Volume 3, 2009 , pp 204-218.
[8]O.Hachour, the proposed hybrid intelligent system
for path planning of Intelligent Autonomous
Systems, International journal of mathematics and
computers in simulation,Issue 3, Volume 3, 2009,
Pages 133-145.
[9]O. Hachour, ,“path planning of Autonomous
Mobile Robot”, International Journal of Systems
Applications, Engineering & Development, Issue4,
vol.2, 2008, pp178-190.
[10]O.Hachour, “The Proposed Genetic FPGA
Implementation For Path Planning of Autonomous
Mobile Robot”, International Journal of Circuits ,
Systems and Signal Processing, Issue 2, vol2
,2008,pp151-167
[11] Hachour Ouarda, “Autonomous Mobile Robots
Motion in Unknown Environments”, recent
researches in system science, proceeding of the 15th
WSEAS international conference on systems (part of
the 15th WSEAS CSCC multiconference” ISSN
:1792-4235;
ISBN : 978-1-61804-23-7, Corfu
Island, Greece, july2011, pp54-59.
[12] Hachour Ouarda, “The Proposed Neural Networks
Navigation Approach”, Applied Mathematics and
Informatics, European Society for Applied
Mathematics – EuroSAM, European Conference for
the
APPLIED
MATHEMATICS
and
INFORMATICS Vouliagmeni, Athens, Greece,
December 29-31, 2010 ISSN: 1792-7390, ISBN:
978-960-474-260-8,2010, pp 88-93.
[13] Hachour Ouarda , “Neural Path planning For
Mobile Robots”, INTERNATIONAL JOURNAL OF
SYSTEMS APPLICATIONS, ENGINEERING &
DEVELOPMENT , Issue 3, Volume 5, 2011, pp
367-376.
Figure. 2: Example of Mutation in the
second bit.
5 Conclusion
The theory and practice of IAS Intelligent Autonomous
Systems are currently among the most intensively
studied and promising areas in computer science and
engineering which will certainly play a primary goal
role in future.
In this paper, we have presented some previous
software implementations of navigation approach of an
autonomous mobile robot in an unknown environment,
Where the principle of using Fuzzy Logic(FL), Neural
Networks (NN), Expert system(ES), Genetic
Algorithms (GA) . These technologies are more
effective in the design and development of intelligent
autonomous systems than the predicate logic based
methods of traditional Artificial Intelligence. FL, ES,
NN and GA are well established as useful technologies
that complement each other in powerful hybrid system
for example.
These theories and applications provide a source
linking all fields in which intelligent control plays a
dominant role. Cognition, perception, action, and
learning are essential components of such-systems and
their use is tending extensively towards challenging
applications (service robots, micro-robots, bio-robots,
guard robots, warehousing robots).
This hybrid proposition has made the robot able to
achieve these tasks : avoiding obstacles, deciding,
perception, and recognition which are the main factors
to be realized of autonomy requirements. Hence; the
combination and hybrid principle are promising for
robust powerful movement task of an autonomous
mobile robot .
References:
[1]T.Willeke, C.Kunz, I.Nourbakhsh, The Personal
Rover Project : The comprehensive Design Of a
dometic personal robot, Robotics and Autonomous
Systems (4), Elsevier Science, 2003, pp.245-258.
[2]A.Howard, M.J MatariĆ and G.S.Sukhatme, An
Incremental Self-Deployment Algorithm for mobile
Sensor Networks, autonomous robots, special Issue
on Intelligent Embedded Systems, 13(2), September
ISBN: 978-1-61804-108-1
151