You are on page 1of 54

PATH TRACKING AND ANALYSIS OF

WHEEL MOBILE ROBOT USING


INTELLIGENT NEURAL NETWORK (INN)

Satya Sundar Sahoo

Department of Mechanical Engineering


National Institute of Technology Rourkela
May, 2019
PATH TRACKING AND ANALYSIS OF
WHEEL MOBILE ROBOT USING
INTELLIGENT NEURAL NETWORK (INN)
A Thesis Submitted to the
Department of Mechanical Engineering
National Institute of Technology, Rourkela
for the award of the degree
of
Master of Technology (Research)
By

Satya Sundar Sahoo


(217ME1228)

Under the Supervision


of
Prof. Dayal R. Parhi

Department of Mechanical Engineering


National Institute of Technology Rourkela
Orissa (India)-769008
May, 2019
DECLARATION
I hereby declare that this submission is my own work and that, to the
best of my knowledge and belief, it contains no material previously
published or written by another person nor material which to a substantial
extent has been accepted for the award of any other degree or diploma of
the university or another institute of higher learning, except where due
acknowledgement has been made in the text.

Date :- Satya Sundar Sahoo


NATIONAL INSTITUTE OF TECHNOLOGY, ROURKELA

CERTIFICATE

This is to certify that the work in the thesis entitled “Path tracking and
analysis of wheeled mobile robot using intelligent Neural Network” By
Satya Sundar Sahoo is a record of an original research work carried out
by him during 2018 - 2019 under my supervision and guidance in partial
fulfillment of the requirements for the award of the degree of Master of
Technology in Mechanical Engineering (Machine Design and
Analysis), National Institute of Technology, Rourkela. Neither this thesis
nor any part of it, to the best of my knowledge, has been submitted for any
degree or diploma.

Date Prof. Dayal R. Parhi


Rourkela SUPERVISOR
Dept. of Mechanical Engineering

ii
ACKNOWLEDGEMENT

I am mainly indebted to my guide Prof. D.R. Parhi who acts like a pole star for me during my
voyage in the research by his infusion, support, encouragement and care. I express my deep regard
to him for the successful completion of this work. The blessing, help and guidance given by him
from time to time made it possible for me to complete the work in stipulated time. His heart being a
great ocean of compassion and love not only created friendly environment during my work with
him but also enlightened my soul.

I express my sincere gratitude to all the faculty members and staff of Mechanical Engineering
Department, for providing me support and advice in preparing my thesis work. Last, but not the
least I would like to thank my parents and my friends who are involved directly and indirectly in
successful completion of the present work.

Date Satya Sundar Sahoo


Rourkela Roll no.217ME1228

iii
ABSTRACT

In this paper, using neural network based controller implemented for nonlinear
dynamic compensation along with adaptive trajectory tracking of a mobile robot
system, Basically using controlling law for mobile robot is a combined form of
back stepping controller which is designed to know all about the inverse
dynamic models of mobile robot along with to compensate the existing different
nonlinearities and uncertainties in the mobile robot system. These control
scheme implemented on the robot is a novel robust tracking controller which
allow providing advantages of dealing within-modeled with some disturbance in
the system. In the second proposed control scheme, the neural network is used to
tune the gain of the kinematic based controller in a back stepping structure.
Using the adaptive capabilities of the neural network and online learning
programmed helps to achieve a fast along with the smooth operation of the
mobile robot system. The comprehensive system includes robot kinematics,
dynamics path controller system in a wheeled robot system. Hence on
simulation result allow the tracking performance of the proposed control
algorithm over the classical back stepping controller.

iv
INDEX

Sl. No. Title Page No.


i
DECLARATION
CERTIFICATE ii
ACKNOWLEDGEMENT iii
ABSTRACT Iv
LIST OF FIGURES viI
LIST OF TABLES viii
Introduction 1
1 About Robot 1
1.1 Types Of Robot 1
1.2 Types Of Navigations 2
2 Robot Architecture 3
3 Sensors For Mobile Robots 4
3.1 Ultrasonic Sensor 4
3.2 Infrared Sensors 4
3.3 Laser Range Sensor 5
4 Applications Of Robot 5
4.1 Advantages Of Robots 5
4.2 Disadvantages Of Robot 6
5 Objective 6
6 Literature Review 6
7 Wheels Used In Wheel Mobile Robot 8
7.1 Conventional Wheels 8
7.2 Swedish Wheels 9
8 Kinematic Analysis Of 4-Wheel Mobile Robot 10
9 Forward Kinematic Model 12
10 The Kinematic Constraints 14
11 Tracking Control Based On The Robot Kinematic Model 14
11.1 Tracking Control By Back stepping Kinematic Into Dynamics 15
12 Introduction To Intelligent Neural Networks 15
12.1 Introduction 15
12.2 Biological Neural Network 16
13 The Neural Network Controller Design 18
13.1 Background On Neural Networks 18
13.2 Feed Forward Neural Network 23
14 Neural Network Training 23
15 Learning 25
16 Basic Building Blocks Of Intelligent Neural Networks 25

v
16.1 Network Architecture 25
16.2 Activation Function 26
16.3 Setting The Weights 26
17 Various Neural Network 27
17.1 Cascaded Neural Network 27
17.2 Elman Neural Network 28
17.3 Feed-Forward Back-Propagation Neural Network 29
17.4 Generalized Regression Neural Network 29
17.5 Radial Basis Neural Network 30
18 Robot Construction 31
18.1 Sensor 31
18.2 Actuator 31
18.3 Controller 32
18.4 Power Conversion Units 32
18.1 Programming Languages 32
19 Simulation & Training 32
20 Experimental Result 38
21 Conclusion 40
22 Future Scope 40
Reference 41

vi
List of Figures
Figure 1 Architecture of robot ..............................................................................................................3
Figure 2 Working principle of ultrasonic sensor ..................................................................................4
Figure 3 Working of infrared sensor .....................................................................................................5
Figure 4 Conventional wheel shape ......................................................................................................8
Figure 5 Types of conventional wheels ................................................................................................9
Figure 6 shows the schematic shape of a Swedish wheel ...................................................................10
Figure 7 Different kinds of locomotion types for wheeled mobile robots ..........................................10
Figure 8 Control Mechanism for mobile robot navigation .................................................................12
Figure 9 The differential drive mobile robot model ...........................................................................13
Figure 10 Shows a simple neural network ..........................................................................................16
Figure 11 Biological neuron ...............................................................................................................17
Figure 12 A neuron model .................................................................................................................17
Figure 13 Mathematical model of a neuron ........................................................................................19
Figure 14 One-layer neural network with s neurons ...........................................................................20
Figure 15 Two layer neural network ...................................................................................................21
Figure 16.Sigmoid function ................................................................................................................22
Figure 17 Common activation function ..............................................................................................23
Figure 18 Example of feed forward network ......................................................................................24
Figure 19 Algorithm for implementation of neural network training (NNT) using back propagation
.............................................................................................................................................................24
Figure 20 Flow network of Cascaded .................................................................................................28
Figure 21 Elman Neural network ........................................................................................................28
Figure 22 Feed forward neural network..............................................................................................29
Figure 23 Generalized regression neural network ..............................................................................30
Figure 24 Radial basis neural network................................................................................................30
Figure 25 neural network structure in MATLAB ...............................................................................34
Figure 26 The performance of the curve of the ..................................................................................35
Figure 27 Regression curves for training validation and testing ........................................................35
Figure 28 The movement of robot in the presence of multi no. of obstacles .....................................36
Figure 29 The movement of robot in obstacle ....................................................................................36
Figure 30 The movement of robot in the presence of multi no. of obstacles .....................................37
Figure 32 Robot crosses the obstacles presented in the environment .................................................38
Figure 31 Robot starts from the initial ................................................................................................38
Figure 32 Robot crosses the obstacles presented environment………………………………………38
Figure 33. Experimental view of multiple no of obstacle……………………………………………39
Figure 34.Experimental view robot moves in the multiple no of obstacle………………………… .39
Figure 35. Robot moves in the multiple no of obstacles……………………………………………..40

vii
List of Tables
Table 1.deals with terminologies of Biological neural networks and neural network........................18
Table 2. Comparison between artificial neural network and biological neural network ....................18
Table 3 Shows the input ,output and turning angle for the robot movement ......................................33

viii
INTRODUCTION
1. About Robot
A robot can be defined as an electromechanical machine that can resemble human action which
will do a series of actions automatically with the help of the program written in the computer.
1.1 Types Of Robot
There are basically six main types of industrial robots: Each of these types offers a different
joint configuration. The joints in the arm are referred to as axes.
a) Cartesian:-
These are also called rectilinear or gantry robots. Cartesian robots have three linear joints that
use the Cartesian coordinate system (X, Y and Z). They also may have an attached wrist to
allow for rotational movement. The three prismatic joints deliver a linear motion along the
axis
b) SCARA:
Commonly used in assembly applications, this selectively compliant arm for robotic assembly
is primarily cylindrical in design. It features two parallel joints that provide compliance in
one selected plane.
c) Cylindrical:-
The robot has at least one rotary joint at the base and at least one prismatic joint to connect
the links. The rotary joint uses a rotational motion along the joint axis, while the prismatic
joint moves in a linear motion. Cylindrical robots operate within a cylindrical-shaped work
envelope.
d) Delta:
These spider-like robots are built from jointed parallelograms connected to a common base.
The parallelograms move a single EOAT in a dome-shaped work area. Heavily used in the
food, pharmaceutical, and electronic industries, this robot configuration is capable of delicate,
precise movement
e) Polar:-
Also called spherical robots, in this configuration the arm is connected to the base with a
twisting joint and a combination of two rotary joints and one linear joint. The axes form a
polar coordinate system and create a spherical-shaped work envelope.

1|Page
f) vertically articulated:-
This robot design features rotary joints and can range from simple two joint structures to 10
or more joints. The arm is connected to the base with a twisting joint. The links in the arm are
connected by rotary joints. Each joint is called an axis and provides an additional degree of
freedom, or range of motion. Industrial robots commonly have four or six axes.

Navigation, that be the process of detecting as well as avoiding the obstacle in the path and to
reach the final point with the help of different sensor information in any environment like
outdoor, indoor, cluttered environment by using different soft computing techniques like
fuzzy logic, genetic algorithm, neural networks, particle swarm optimization and fuzzy logic
operation.

The main advantage of using mobile robot in the real life is to reduces the human effort. The
complete autonomous mobile robot should capable of do the things without human
involvement. In the simple way, it should be capable of act in the real environment. The
autonomous mobile should capable of do thing like capable of bring surrounding information
with no difficulty, without human involvement the robot should move from one point to
another, it should avoid the obstacles in the path and act according to the situation if
necessary.

1.2 Types Of Navigations


Navigation can be defined as the combination of the three fundamental competences
a) Self-localization
b) Path planning
c) Map-building and map interpretation

Controlling the speed has two features


a. Avoiding different obstacles
b. Avoiding the overturning in path

Navigation in the any environment can be categorized into two types


a. Controlling a heading
b. controlling speed

2|Page
Controlling the Heading has 4 features
a. Obstacle comes front it can avoid
b. Obstacle comes right it can avoid
c. Obstacle comes left it can avoid
d. Goal reaching.

Finally the above all given features used the information from the sensors and gain is goals.
Feature like avoiding the obstacle use the different sensors to find out distance from obstacle
to near to the robot. The feature like goal seeking uses the digital compass and it will measure
the direction of the target. The feature like avoidance of overturning uses the speedometer; it
gives the reading of the mobile robot speed. Now the focus is on how to develop a navigation
technique for mobile robot. For developing a navigational techniques many students,
researchers are get attracted and become the main trend in the navigational mobile robots.

This trend is favorable causing the present small gap between the available technology and
the new user application demands. The major problems in the industrial robots are lack of
autonomy, flexibility, frequent breakdown. These robots are generally performs only pre-
programmed sequence of operations in constrained environments and these robots cannot
perform the operations in the new environment if faces any unexpected situations occurs. At
present market, there will be a heavy competition for a complete autonomous mobile robot.
For the navigation mobile robot, there are many soft computing techniques available such as
genetic algorithm, Particle swarm optimization, fuzzy logic, Ant colony optimization etc.

2. Robot Architecture
Now we should characterize the design the exterior along with interior design to make a
perfect robot which enable to work and obey the laws of robotics .following figure-1 shows
the represents the map for transfer the data in the robot.

sensor Motion controller Mechanical


planner structure

Configuration sensor

Figure 1. The architecture of robot

3|Page
3. Sensors For Mobile Robots
Basically various types of sensors have been used for mobile robot navigation. It is
classified as three categories:

(i) Ultrasonic Sensors, (ii) Infrared Sensors, and (iii) Laser range Sensors.

3.1 Ultrasonic Sensor:-


A robot without an ultrasonic sensor is blind Ultrasonic sensor that allows our
robots to react to the world around them like to do with our eyes and ears. There are many
types of ultrasonic sensors, and they provide your robot with new or improved senses.
Ultrasonic sensors provide fast response times which allow your robot to respond quickly
to the changing environments around it. Our ultrasonic sensors provide your robots with
valuable information on the distance and direction to the target. Figure 2 shows the
working principle of ultrasonic sensor.

3.2 Infrared Sensors (Ir Sensor):-


An infrared sensor is an electronic device, that emits in order to sense some aspects of the
surroundings. An IR sensor can measure the heat of an object as well as detects the motion.
These types of sensors shows in the figure 3 shows the working of the infrared sensor that
able to measures only infrared radiation, rather than emitting it that is called as a passive IR
sensor.

Figure 2.Working principle of ultrasonic sensor [26]

4|Page
Figure 3. The working of infrared sensor [26]

3.3 Laser Range Sensor:


Laser light is transmitted and the reflected light is captured and analyzed. Distance is measured by
calculating the speed of light and time taken for the light to reflect back to the receiver. These sensors
are very useful for longer distances.

4. Applications of Robot:-
Robots are powerful machines that they are used in almost every application especially in
industrial applications, these keep safe us from dangers while doing different tasks that are
related to harmful to our health. These are some applications of robot.

a. In military applications, the robots are frequently used. For detecting and disposal of bombs,
monitoring the vehicle.

b. In automobile industries, for assembling the parts of the automobile is difficult for human
being so that robots ply a key role in this industry.
c. In space technology, space exploration, remote inspections of space stations.
d. For inspecting and fault detections of the parts in all industries.

e. In manufacturing industries, the robots are actively used where any repetitive work attains.

4.1 Advantages Of Robots :-

a. The robots that give the accurate results while inspection of the pars.

b. Robot can capture the image with in fraction of second’s time gap

c. Robot can capture the image with in fraction of second’s time gap

5|Page
4.2 Disadvantages of Robot :-
a. Maintenance of the robot is expensive.

b. Robot needs more power supply

5. Objective:
The aim of this project is to design an autonomous mobile robot which should be able to think
and decide its path while navigating in a cluttered environment. It should traverse its path in the
minimum possible route and at the same time must avoid collisions with obstacles on its way. It
can be able to detect obstacles, plan its path and move toward its goal safely.

6. Literature Review:-
The paper [27] is to diagnosis of fault for a cantilever beam subjected to transverse crack, radial
basis function neural network (RBFNN) method is adopted .The performance of the cantilever
beam gets degraded when there are no. of cracks present in it as well as affects the vibration
signature(mode shapes and natural frequencies).Basically graphite fiber reinforced polyimide
composite material used in the beam. To calculate the natural frequencies at different nodes and
mode shapes in a cracked and un-cracked beam, finite element analysis method is taken out
under the software usage of ANSYS, which generalize to forecast the crack position and its
depth by using smart based system on RBFNN. RBFNN controller generated to provide natural
frequency and mode function under different input parameter, finally using RBFNN controller,
we can get the result of crack depth and location of crack of composite beam under numerical
analysis. The paper [16] allows finding dynamic behavior analysis of a beam structure under the
influence of transverse crack using the method of neural network controller. basically natural
frequency and mode shapes can be calculated using theoretical, finite element method and
experimental method to analysis for a beam which being in cracked or un cracked. after then
results of those three analysis compared with each other from the result calculated vibration
signature is implemented for train the feed forward multilayer neural controller along with back
propagation technique in order to predict the cracks in the beam. On the source relative crack
depth and relative crack location are the outputs from neural network controller. Using
developed experiment setup results is obtained. so neural network controller (NNC) is the best

6|Page
method for obtaining crack location on different position using crack diagnostic parameter. The
article [1] for navigating the autonomous mobile robot, the research is carried out in order to
avoid various obstacles like structured or unstructured environments generated during movement
of mobile robot under a crowded area. It allows verify various robots and issues of different
robot behavior .All the actions for movement of mobile robot coordinate using fuzzy logic
.Basically inputs for autonomous mobile robot deals with fuzzy logic scheme which consist of
heading angle between robot and desired destination. and finally its distance from robot to the
obstacle whether it moves left, right or front direction observed by array of sensor. Basically
intelligent controller for mobile robot is applicable under complex environment. The result
obtained using simulation approaches along with experimental method. The paper [12] refers the
navigation of the mobile robots by using wireless sensor networking communication protocols.
This intelligence system is most suitable for the path planning of the robots in anywhere and
control by internet. it is represent the blockage of the induced more energy and effective path
planning technique for WSN named as survivable path routing. This system is works with high
traffic because more no. of mobile robots starts at the same time and the same target point. In
this methodology some algorithm are used for running three parameters like signal induction and
noise ration then survivability factor for the relative path of the robots. Finally the simulation
and analysis result of the survivable path planning of the mobile robots is more effective by the
online internet system with wireless sensor networking. [17]the paper Motivated by the way a
biological neural network functions, ANN consist of a massive parallel computing systems
comprising a huge number of simple processers with many connections in between them. ANN
structures try to utilize various characteristic principles on which our brain works. In this paper
(6) the mobile robots control by the intelligent approach. Here the mobile robots can drive safely
by avoiding the structural and unstructured area. This intelligent approach also navigates the
robots in an unknown and sudden change of the environment. By the intelligent approach the
robots also navigate in the real world application like air, water, underground and ground and
space. Sensor are used navigate the mobile robots and measure the distance of the left and right
hurdles, which are present around the robots. The research paper [18, 22] deals with the study
and analysis of biomechanics and mechanical behavior of the human femur bone. If human
femur bone is damage by an accident, it may replace by the use of artificial implants. All the
biological artificial implants are not suitable fit with the human body, so this paper is to analyzes

7|Page
and resolve the issue of vibration characteristic and frequency range, according to design aspect
[5]. A solution to the inverse kinematics is a set of joint coordinates which correspond to a given
set of task space coordinates (position and orientation of end effectors). For the class of
kinematic ally redundant robots the solution is generically no unique such that special methods
are required for obtaining a solution. The paper [21-25] presents a new algorithm for solving the
inverse kinematics which is based on a modified Newton-Raphson iterative technique. The new
algorithm is efficient, converges rapidly, and completely generalizes the solution of the inverse
kinematics problem for redundant robots. The method is illustrated by a numerical example.

7. Wheels Used In Wheel Mobile Robot


A wheeled mobile robot is a wheeled vehicle which is capable of an autonomous motion
without external human drivers, because it is equipped, for its motion, with motors that are
driven by an embedded computer. We can categorize the wheeled mobile robot systems into
two categories according to their wheel type:
(a) conventional wheels (b) Swedish wheels

7.1 Conventional Wheels


For a conventional wheel, the contact between the wheel and the ground is supposed to satisfy
the pure rolling without slipping constraint which implies the velocity of the contact point is
equal to zero and the components of this velocity parallel and orthogonal to the plane of wheel
are equal to zero. Schematic diagram of conventional wheel is specified given in Figure 4.and
various types of conventional wheels shown in Figure 5.[12]

Figure 4. the conventional wheel shape [12]

8|Page
Basically there are three types of conventional wheels:
a. fixed wheels,
b. center orient able wheels
c. off-center orient able wheels

Figure 5 shows types of conventional wheels [27]

7.2 Swedish Wheels :-


For a Swedish wheel, only one component of the velocity of the contact point of the wheel with
the ground is supposed to be equal to zero along the motion and the direction of this zero
component is arbitrary but it is fixed with respect to the orientation of the wheel shown in Figure
6.
Various types of wheels can be used according to the application and various environments. The
other important issue about wheeled mobile robots is their locomotion type. Wheeled mobile
robots can have the following five types of locomotion system
a) Differential drive

b) Steered Wheels: Tricycle, Bicycles and Wagon

c) Synchronous drive

d) Omni-directional

e) Car drive (Ackerman steering)

The differential drive wheeled mobile robot which is the simplest and most popular mobile
robot which has the simplest drive mechanism. It is a platform equipped with a front castor and a

9|Page
pair of rear co- axial drive wheels. Different types of wheeled mobile robot are shown in Figure
7. Each of these drive wheels are independently driven by a DC motor which is in turn energized
by a control voltage.

Figure 6 Schematic shape of a Swedish wheel[27]

Figure 7 Different kinds of locomotion types for wheeled mobile robots[14]

8. Kinematic Analysis Of 4-Wheel Mobile Robot


The purpose of this thesis is to analysis you to robotic kinematics, and the concepts related to both
open and closed kinematics chains. Forward kinematics is distinguished from inverse kinematics.
Kinematics is the study of motion without regard to the forces that create it. The forward kinematics
is about finding an end effectors or tool piece pose given a set of joint variables. The Inverse
Kinematics is the opposite problem. In this subsection, it will extracting the relationship between
joint movements and end effectors movements. More precisely, they will try to develop equations
that will make explicit the dependence of end effectors coordinates on joint coordinates and vice
versa. In inverse kinematics we can determine configuration of links provided the configurations of

10 | P a g e
end-effector are provided.

Basically the kinematics problem is to find a transformation matrix co-relating body coordinate
frame to a base coordinate frame. The relation between the bodies references coordinate to base
coordinate of reference is referred to by transformation matrix.
The transformation matrix is classified in to three types

a. translation
b. rotational
c. scaling

R(x, ø) = Rotation about x axis by an angle ø in anti-clockwise sense

(1)

Rotation about y axis by an angle μ in anti-clockwise sense:

(2)

Rotation about z axis by an angle r in anti-clockwise sense

(3)

If two co-ordinate frames are orthonormal and given by X and A, R to be the transformation that
maps X co-ordinate into A co-ordinate,
Then, X = RA (4)
and A maps with X by the relation
A = R-1X = RTX (5)

A general method used for solving the kinematical equations for a robot manipulator with different

11 | P a g e
joints is the use of “homogeneous transformations”. A generalized transformation is now described
by a single matrix that combines the effects of translation and rotation.

The entries consist of a 3x3 sub-matrix responsible for the rotation and a 3x1 sub matrix responsible
for translation. Two other partitions may be identified as perspective transformation and scaling
factor. In fig-8 represents the control mechanism for the mobile robot navigation

Figure 8 Control Mechanism for mobile robot navigation[32]

9. Forward Kinematic Model


Assume a differential drive mobile robot setup which has two wheels with the radius of Ra placed
with a distance L from the robot center as shown in figure-9.

The following notations will be used in this thesis:


A: The intersection of the axis of symmetry with the driving wheels axis

C: The center of mass of the platform

a: The distance between the center of mass and driving wheels axis in x-direction

L: The distance between each driving wheel and the robot axis of symmetry in y-direction

Ra: The radius of each driving wheel

Φr: The rotational velocity of the right wheel

Φl: The rotational velocity of the left wheel

12 | P a g e
v: The translational velocity of the platform in the local frame

ω: The rotational velocity of the platform in the local and global frames

Figure 9 The differential drive mobile robot model[14]

So the kinematic forward problem can be described as the problem of finding the following given
function:

Speed of wheel mobile robot of each wheel speed is , the translational speed in the robot frame
gives average velocity which is given as

V= Ra (6)

Finally velocity in global frame is given by

(7)

13 | P a g e
So this above equation represents the general forward kinematic equation for a differential drive of
wheel mobile robot.

10. The Kinematic Constraints


Some assumption for wheel motion makes a cause the mobile robot kinematic constraint.

a. The Point contact between the wheels and ground


b. Movement on a horizontal plane
c. No skidding, slipping, or sliding.
d. No friction for rotation around the contact points
e. Steering axes of robot is orthogonal to the surface

11. Tracking Control Based On The Robot Kinematic Model


The tracking control mainly proposed a stable tracking control rule for nonholonomic vehicles.
Stability of the rule is proved through the use of a Lyapunov function. The inputs to the
controller are the vehicle posture and the reference velocities and the output is the target linear
and rotational velocities. The control rule proposed in this work was based on the vehicles
kinematic model and it does not consider the effects of robot dynamics. This controller is robot
independent and can be applied to any kind of mobile robot with dead reckoning ability. The
appropriate controller parameters were found out by linearizing the systems differential
equations and finding a condition for critical damping. The need for velocity and acceleration
limitations is also discussed in this work. To compute the vehicles control inputs, it is assumed
that there is perfect velocity tracking. There are three problems with this approach: first, the
perfect velocity tracking problem does not hold in practice. Second, disturbances are ignored and
third, complete knowledge of dynamics is needed. In order to improve the performance of this
proposed control algorithm, other controllers which include the vehicle dynamics and can deal
with disturbances are proposed.

11.1 Tracking Control By Back stepping Kinematic Into Dynamics


The work which is a dynamic extension that makes possible the integration of a kinematic
controller and a torque controller for nonholonomic mobile robots. A combined kinematic/torque
controller is developed using back stepping and the stability is guaranteed using Lyapunov theory.
14 | P a g e
The proposed control algorithm can be applied to three robot navigation problems: tracking a
reference trajectory, path following and stabilization about a reference posture. This control
algorithm provides a rigorous method of taking into account the specific vehicle dynamics to
convert the steering system command into control input for the actual vehicle. First, feedback
velocity control inputs are designed for the kinematic steering system to make the position error
asymptotically stable. Then, a computed torque controller is designed such that the mobile robot
velocities converge to the given velocity inputs. This control approach can be applied to a class of
smooth kinematic system control velocity inputs and therefore, the same design procedure works
for all of the three basic navigation problems.

12. INTRODUCTION TO INTELLIGENT NEURAL NETWORKS


12.1 INTRODUCTION:-
The neural network consists of a very large no. parallel distribution processor which has the ability
to store experimental information and furnishing it according the requirement of the network.

It is classified into 2 types

1. The neural network learns through the learning process


2. The desired information is stored by means of neuron-neuron connection which is called as
synaptic weight.

It is also defined as a non-linear algorithm for image or signal or data processing.

Generally INN information process method, the information is processed by means of elements
which can be defined as a neuron. The output is passed through link connection and the output
signal is realized by using activation function. These are characterized by three basic function.

1. Activation function
2. Training & learning
3. Architecture

INN models try to use some characteristic principle behind the working of the human brain

.Neural network acts to help for driving conclusion from even imprecise data and as certain
trends to discover both computers as well as human being.
15 | P a g e
As to describe the intelligent neural network may become proficient in analyzing data if it is
given training.

So it can be predicted the solution to that information which has not been trained into other
advantages.
In neural network other advantages can be explained which included
1. Self-organization develops a structure of the data received while being taught.
2. Some network faculties may be kept when some significant network damage occurs
3. Operation at the real time.
4. Adaptive learning- has the ability to gain knowledge on the performing tasks based on data
provided while being trained in the above given neural network. In figure10 represents the
simple neural network under the input and the output layer.

Y OUTPUT

W1 W2

X1 X2 INPUT
Figure 10 Shows a simple neural network

12.2 Biological Neural Network:-


A biological neural network or a nerve cell comprises dendrites, cell body, the axon and synaptic
weight. The building blocks are discussed below

Dendrites:-It acts as to acquire information

Soma :- All the information adds to the neurons.

Axon:- when input reaches a predetermined threshold value, the information is passed onto
subsequent neurons.

16 | P a g e
A neuron is the basic element of a neural network. A biological neuron acquires input from
precious neurons, integrates them as per a rule, and perform non-linear manipulator on the result
and then provides the output. In fig-11 biological neuron model is specified with its different
function.

Figure 11 Biological neuron [5]

So some properties related to biological neuron pose the features on the artificial neurons which are
mentioned below
1. The weights at the receiving end have the capability to modify the incoming signal.
2. The neuron fires (transmits output), when sufficient input is obtained.
3. The output produced from one neuron may be transmitted to other neurons.
4. The processing of information is found to be local.
5. The weights can be modified by experience.
6. Neurotransmitters for the synapse may be excitatory or inhibitory.
7. Both artificial and biological neurons have inbuilt fault tolerance.
8. Signals are received by the processing elements.

Figure 12. A neuron model

17 | P a g e
Table 1.deals with terminologies of Biological neural networks and neural network

Biological Neural Network Neural Network


Cell Body Neurons
Dendrite Weights
Soma Net input
Axon Output

The major differences between artificial and biological neural network are given below

Table 2. Comparison between artificial neural network and biological neural network

Characteristics Neural Network Biological Neural Network


Faster in processing data. Slower in processing
Speed One cycle is operated in information. One cycle is
about nanoseconds. operated in milliseconds.
Instructions are operated
Parallel processing of
Processing sequentially, one after
information takes place.
another.
Due to the greater number of
Difficult to perform complex processing elements, more
Size and complexity
pattern recognition. complex pattern recognition
tasks can be accomplished.
New information in a
memory location replaces the Old information and new
Storage
old information. It is strictly information co-exist.
replaceable.
A control unit monitors all No specific control
Control Mechanism
the computing activities. mechanism.

13. The Neural Network Controller Design


In order to explain the details of the design of a neural network controller for a trajectory of this
wheel mobile robot system, a brief background on neural networks that includes different Neural
network topologies, properties of the network, weight selection and training.
13.1 Background on Neural Networks

Neural network topologies and overview


Intelligent Neural Network is modeled on the basis of biologically way for the information
processing also specifically the nervous system and its basic unit, the neuron. Generally, Signals
from the neurons are generated in the form of potential differences between the inside and outside
18 | P a g e
of neural cells. The mathematical expression for a neuron is formed on a one layer network.

Figure 13. Mathematical model of a neuro[27]

The input weights Wp which has the firing threshold b (also called the bias), the summation of the
weighted inputs and the nonlinear function (f) are shown in the above figure. If the cell of network
inputs are „n‟ no. of signals at the time instant k, x 1(k),x2(k), and x3(k)…..up to xn(k ) and the
output is the scalar y (k),

The mathematical equation of the neuron can be written as follow

∑ ) ) (8)

Basically, the activation functions which are selected for specific to the application through some
common choices which are as follows

Linear Threshold value at y= x (9)

Sigmoid (logistic curve) at y = (10)

The hyperbolic tangent at y = (11)

19 | P a g e
The hyperbolic tangent at y = (12)

The adverse intent of the neural network activation function is to prepare a model of nonlinear behavior of
the neural cell where there is no output below a certain values of the argument. The above Sigmoid
functions are general class of monotonically non decreasing functions that takes on the bounded values.
For n no. of neural network training algorithms including back-propagation algorithm, the derivative of the
activation function is needed, so the selected activation function should be differentiable. The given below
figure shows a one layer neural network having “S” no. of cells.

Figure 14. One-layer neural network with s neurons


The equation for the network can be defined as

Y= ∑ ) )[27] (13)

It is more suitable to write it down various weights as well as threshold values in matrix and vector
forms. So by defining in a matrix form of weights and thresholds as

20 | P a g e
We can write the output vector y(t)= [Y0 Y1 Y3 …….YN]T

As Y(t) = f(WTX +BW) (14)

So the two layer neural network which have containing two layer neural network with two neurons
along with one layer having L neurons feeding the second one having m no. of neurons in the
network.

Figure 15.The two layer neural network


The first layer in the neural network is called as hidden layer and the second layer is called as the
output layer.

The neural network with multiple no. of neurons called multilayer perceptron and its computational
power is focused significantly over the one layer network. The output of the above layer network is
given below.

yi = δ∑ (∑ ) ) (15)

The output in the hidden layer defined as Zl , then we can write the above function

zl = δ(∑ ) where l=1,2,3….L (16)

yi= δ(∑ )) where i=1,2,3……m (17)

21 | P a g e
The above two equation can be written in the matrix form and it is essential to write the input to
hidden layer weights can be selected randomly and hidden to output layer weight will turned into
approximate training methods.

As a result it will minimize the computational complexity of the network using neural network in
the control algorithm.

In neural network, these are composed of simple element operating in parallel which are came from
biological nervous system.

As in nature, the network function in neural network is determined largely by the connection
between element. We can train a intelligent neural network in order to show its performance for a
particular function by just adjusting different values of the connection called as weights commonly
neural network are trained and adjusted in order to reach its specific target output on a certain
particular input.

Figure 16. The sigmoid function [9]

The non-linear gain is basically calculated by finding the ratio of the change in the output in a small
change in NET. So to obtain the slope of the curve shown in fig-16 at a specific change in excitation
level of the sigmoid function. Basically it varies from the low values of the larger negative
excitation to a high value at zero excitation and it drops back as excitation. Small signals, while its
region of decreasing gain the positive the end negative extreme are appropriated for the large
excitation, at finally neuron in the network performs with appropriate gain over a wide range of
input levels. different types of function used and the variation in the graph shown in figure 17.

22 | P a g e
Figure 17.Common activation function [9]

13.2 Feed Forward Neural Network:-


In the feed forward neural network allows to travel in one direction only that is from input to
output and there is no feedback loops which the output of any layer does not affect the same layer.
These networks are called non-current networks and they don’t require any kind of memory as an
output are directly related to inputs and weights. They are extensively used in pattern recognition.
This kind of system specially referred to as bottom-up or top-down. The figure shown below
describes about simple feed forward neural network.

14. Neural Network Training:-


As it provided in the previous section, for an neural network (NN) to a function as desired, it
is most required to determine a suitable weights as well as thresholds values that ensure
performance as desired. Now a days those problems have for the most part been solved and there
are very good algorithms for NN weight selection and tuning. The back propagation training
algorithm is one of the best algorithms because of its simplicity and power.

In order to construct a neural network (NN) for the most approximation of a general multi input
as well as multi output function, there should be one algorithm to initialize the neural network, and
use the back propagation algorithm to train the network analysis and use the approximation error of
the neural network to show its performance. The algorithm which is used to construct the

Neural network (NN) and implement it through a suitable software package for function
approximation is shown figure below.

23 | P a g e
INPUT HIDDEN LAYER OUTPUT

Figure 18. The example of feed forward network[9]

Figure 19. Algorithm for implementation of neural network training (NNT) using back propagation

24 | P a g e
To obtain containing one output, two inputs and 10 hidden layer neurons, the forward propagation
will be as given bellow
Y= σ(∑ ) (18)
After forward propagation, we can able to find the error for each input data pattern and also the sum
of error squared (SSE)

SSE=∑ (19)

Where Np is the number of input data patterns in one cycle and Ep is the approximation error
corresponding to each pattern. Then MSSE=SSE/Np (20)

Where MSSE is the mean sum squared error by just finishing loop of the data cycles.

15. Learning:-
Learning is an essential part while designing these neural network architectures and hence
the choice of a algorithm is a central issue in the network development. Learning gives the
processing unit which is capable of changing its output or input value as in result of changes in
the environment. Since the activation rule is usually fixed when the network is designed and
since the input/output vector cannot be changed to change the behavior of input and output the
weight corresponding to that of input vector need to be adjusted. In a neural network, learning
can be supervised or to be unsupervised.

16. Basic Building Blocks Of Intelligent Neural Networks


The basic building blocks for an intelligent neural network are

1. Network Architecture
2. Activation Function
3. Setting The Weights

16.1 Network Architecture:-

The structure of the neurons into various layers and the trends of relationships intra-layer
and inter-layer is considered as the network architecture. The neurons in the network

contained either in fully interconnected or in partially interconnected. Various types of


network architecture like radial, basis, cascaded, back propagation networks etc. INNs can
25 | P a g e
be varied according to its size and shape. The network behaves as a vector-valued function
which takes in an input vector and returns another output.
16.2 Activation Function:-

The output of the neural network to a neuron generally is obtained by using activation
function. For neurons present in the same layer, same activation functions are used. In the
network activation function may be linear or non-linear.

eg:- linear activation function f(x) = x;


sigmoidal function is a non-linear function Which is represented by

y= (21)
16.3 Setting The Weights:-

Training process in the network is basically determining the values of the weights. Learning
(mentioned above) is the process of internally training a network. Basically there are 3 types
of learning methods as follows.
a. Supervised Training:-

In this process the network is provided with sample inputs and estimated outputs. The
training sets values to weights and computes outputs which are then compared with the
expected results. The learning runs till the output is equal to the expected results. These
weighs are determined as per the learning algorithm. This method is used in pattern
recognition. Some examples of supervised training are Back propagation network, Hebb
network.
b. Unsupervised Training:-
In unsupervised training the output vectors are not known for the respective input vectors.
Unsupervised networks are found to be more complex and difficult to implement. It
involves looping connections back into feedback layers and iterating through the process
until some sort of stable recall can be achieved. Unsupervised networks are also called self-
learning or self-organizing networks. Unsupervised is used in the areas of adaptive
resonance theory, self-organizing feature maps etc.

26 | P a g e
c. Reinforcement Training:-
In this training the expected answer is not provided to the network ,Instead it is given an
indication if the output wrong or correct. The network utilizes this data to refine its
interpretation. If sufficient information is available, the reinforcement learning can readily
handle a specific problem. In this method a trial and error method is used to map inputs
with outputs. The network does not know the correct output even if it knows output to be
right or wrong.

17. Various Neural Network:-


There are five major types of neural networks used:

d. Cascaded Neural Network


e. Elman Neural Network
f. Feed-forward Back-propagation Neural Network
g. Generalized Regression Neural Network
h. Radial Basis Neural Network

17.1 Cascaded Neural Network:-


A single neuron cascaded (SNC) architecture provides compact and simpler models with
less time consumed for performing tasks. Its unique network structure allows automation in
design. This network combines the strengths of multi-layer and single-layer neural
networks. The Cascaded architecture comprises an input layer, hidden layers and an output
layer. The input layer receives external information as input. Subsequent layers get external
information as well as outputs of all preceding layers. As the input to a neuron comprises
external inputs and output of all previous layers it is called as Cascaded. A cascaded neural
network can have any number of neurons in each layer shown in fig-20. It improves
mapping capabilities.

This network possesses the advantages of multi-layer and single-layer networks. The result
is a much simpler architecture that can undergo training and growth, of its own, to get the
expected response. It is also executes faster as compare to others.

27 | P a g e
Figure 20. the flow network of Cascaded[28]

17.2 Elman Neural Network:-


The ENN is one type of simple recurrent neural network. It was created in 1990 by Jeffrey
L. Elman. ENN comprises an input layer, a hidden layer and an output layer. Hence it is
akin to a 3-layer feed- forward neural network. Additionally, it also consist a context layer.
This context layer has its input as the output of the hidden layer. This output is then fed to
the hidden layer in subsequent iterations. This allows for the formation of a short-term
memory. ENNs are very helpful in forecasting time variation patterns. The image in fig-
21shows an ENN.

Figure 21. The Elman Neural network[29]

28 | P a g e
17.3 Feed-Forward Back-Propagation Neural Network:-

A FFNN has neurons only connected forward, but there are no backward connections.
The term “feed forward” refers to the way network processes and recalls functions. The
term “back propagation” refers to the way network is trained. It is an example of
supervised learning. By utilizing the expected responses, the back propagation training
algorithm makes calculations of errors and changes the weights of different layers
traversing backwards-output layer to input layer.

Figure 22. The feed forward neural network[30]

Complexity depends on the extent of the network, the quantity of patterns to be trained to,
the number of epochs, the tolerance of the minimizer and the speed of the computer. So
Back propagation ANNs are extensively used in diverse applications, such as pattern
recognition, location selection and performance evaluations.
17.4 Generalized Regression Neural Network:-
This network was proposed by D.F. Specht. GRNN is a type of probabilistic neural
network which requires simply a portion of the training inputs that a feed-forward
algorithm requires. This characterizes GRNN as a useful tool to predict as well as compare
performance of system. The probability distribution function used in Generalized
Regression Neural Network is the Normal Distribution. GRNN can solve any function
approximation.

Like back-propagation, it does not need an iterative algorithm. A Generalized Regression


Neural Network comprises four layers namely input, pattern-summation, summation layer.
and output which is shown in figure 23 below[13]
29 | P a g e
Input layer pattern layer summation layer output

Figure 23 shows the generalized regression neural network [31]

17.5 Radial Basis Neural Network

This neural network is an ANN which uses radial basis functions as activation functions.
The output of the network is a linear combination of these functions (functions dependent
on the norm w.r.t. a center vector and are radially symmetric about the vector) of the inputs
and the neuron parameters in Fig-24. RBFNs have applications in function approximation,

Figure 24. The radial basis neural network[33]

30 | P a g e
forecasting time series. It was proposed by Lowe and Broom head in 1998. RBFs have three
layers- input, hidden along with a non-linear RBF activation function and a linear output
layer. Euclidean distance and Gaussian function are taken to be the norm and activation
function respectively.

18. Robot Construction


a) Sensors
b) Actuators
c) Controllers
d) Power Conversion Units
e) Programming Languages

18.1 Sensor:-

Sensor is a device, module, or subsystem whose purpose is to detect events or changes in its
environment and send the information to other electronics, frequently a computer processor.
A sensor is always used with other electronics.

i. Accuracy
ii. Linearity
iii. Repeatability
iv. Measurement Range
v. Resolution: smallest observable increment in output.
vi. Response time
vii. Sensitivity

18.2 Actuator:-
An actuator is a component of a machine that is responsible for moving and controlling a
mechanism or system, for example by opening a valve. In simple terms, it is a "mover".
An actuator requires a control signal and a source of energy. The control signal is relatively
low energy and may be electric voltage or current, pneumatic or hydraulic pressure, or even
human power. Its main energy source may be an electric current, hydraulic fluid pressure,
or pneumatic pressure. When it receives a control signal, an actuator responds by
converting the signal's energy into mechanical motion.

31 | P a g e
18.3 Controller :-

A controller is a device which takes one or more inputs and adjusts its outputs so a
connected device functions in a controlled manner. We offer a wide range of controllers
from simple motor controllers to advanced microcontrollers that can be tasked with multiple
control outputs and perform closed-loop control. Different actuators require differing
control methods to achieve stable output, choosing an appropriate controller is therefore
important to the final operation of your project.
18.4 Power Conversion Units:-
Battery:-
A battery is a device consisting of one or more electrochemical cells with external
connections provided to power electrical devices such as flashlights, smartphones, and
electric cars Primary batteries, or primary cells, can produce current immediately on
assembly. These are most commonly used in portable devices that have low current drain,
are used only intermittently, or are used well away from an alternative power source, such
as in alarm and communication circuits where other electric power is only intermittently
available.
18.5 PROGRAMMING LANGUAGES

As we used MATLAB for the purpose of Simulation, for the purpose of simulations a great
help comes from the MATLAB Neural Network Toolbox. The toolbox gives a complete set
of functions and a graphical user interface required for the design, implementation,
visualization, and simulation of neural networks. One can use both supervised training as
well as unsupervised training and it also support a variety of training functions.

19. SIMULATION & TRAINING


In this work the training data sets were generated by using sigmoid and symmetric sigmoid
function. As the mobile robot moves in its environment and to reach its goal it is likely to
encounter obstacles. The robot task in such a condition is safely reach its goal without any

collisions with the additional data requirement for choosing the most optimum path. Basically a
threshold distance which need to be maintained as it moves towards the target. The wheel
mobile robot is being continuously fed with certain inputs distance from the obstacles ,by
means of sensors. By the data provided it has to undergo either on the given following
32 | P a g e
operation on encountering an obstacle. Back-propagation algorithm was used for training the
network and for updating the desired weights. In this work epoch based training method was
applied

 If left obstacle distance < threshold distance (safe distance) then look for other direction
which is in right or move ahead.

 If forward obstacle distance < threshold distance then look for other direction like left or
right.
By considering 15 no. of input given in Table no.1 in the network according the obstacles
followed in the real environment , threshold distance provided in the robot to avoid collision
with the obstacles and to the environment.

Table 3 The input ,output and turning angle for the robot movement
Turning
Sl.No. f.o.d.(cm l.o.d.(cm) R.o.d.(cm) T.A.(degrees)
1 16 80 16 90 left
2 16 16 80 -90 right
3 80 8 80 -30 right
4 20 60 40 55 left
5 30 80 50 55 left
6 12 80 80 90 left
7 20 40 80 -55 right
8 12 12 12 -85 right
9 40 12 12 0 straight
10 20 8 8 10 small left
11 80 12 12 0 straight
small
12 20 8 8 -10
right
13 80 12 80 0 straight
14 80 12 12 0 straight
15 80 80 80 -85 right
Where,
F.o.d.: front obstacle distance (cm)

L.o.d.: left obstacle distance (cm)


R.o.d.: right obstacle distance (cm)
T.a.: the angle by which the robot turns on detecting an obstacle(degree)
33 | P a g e
By taking an example from the serial no. 7 , if f.o.d , l.o.d and r.o.d are 20cm, 40cm and 80cm
respectively then the robot will shift in the direction of straight forward. If there is less than the
threshold distance then robot will turn clockwise direction by a turning angle of 550(motion for
anticlockwise is +ve and for clock wise ,its –ve). After turning through the robot continues to move
towards the target unless it encounters another obstacle where the above procedure is again
followed.

Using MATLAB neural network toolbox, we can able to train the input in such a case robot has to
take decision to reach its target without collision with any obstacle in the real environment

STEPS FOR TRAINING


1. At first give the input values by creating a 15x3 matrix in the workspace named , input‟ and then
transpose it.

2. now it provide the output values by creating a 15x1 matrix in the output named „output‟ and then
transpose it.

3. Open Command Window and then the toolbox by using nntool‟.

Now we need to import all the input to this network .The input will appear in front in the input data
box. The output is also fed by import then it will appear in target data box.
On applying the all input and output then train it nn tool.

Now the network is trained by allowing the option “train the network”

After training the value, the training activity has to be performed repeatedly for more than 80 times
to get lesser values of mean square error, regression=1, and desired output = actual output. The
figure-26 below shows the results of 1 training run.

Figure 25 Neural network structure in MATLAB


34 | P a g e
Best Validation Performance is 2.5383e-05
at epoch 10

Figure 26 Performance of the curve of the MSE

TRAINING R:1 VALIDATION R-1


Output ~= 1*Target + 0.00025

Figure 27 Shows regression curves for training validation and testing

Fig-27 represents the regression curve under certain training validation and testing given by
data input. After the training has occurred now the network is ready to be used for simulation.
Then we start to use network for the program where the robot moves towards the target in an
environment containing obstacles.

35 | P a g e
The algorithm is defined by

1. Input the start in the command and target the position.


2. Put the input position of obstacles.
3. Determine the angle in which robot should move
4. Load the neural network in MATLAB
5. Run the iteration according to requirement

a. Moving the robot towards the goal, determine the speed of movement.
b. Calculate distance of obstacles from the robot.
c. Give the threshold distance condition.
d. The robot obstacle distance is fed to the neural network and the network computes the angle
to be taken by the wheel mobile robot.

e. At finally robot reaches its target and stops the program.

Figure 29 Shows the movement of robot in obstacle Figure 28 Shows the movement of robot
in the presence of multi no. of obstacles

36 | P a g e
Figure 30 Shows the movement of robot in the presence of multi no. of obstacles

The following algorithm shows the movement of robot in multiple obstacles scenario:

1. Give the initial and target coordinates.

2. Give the initial positions of the obstacles.

3. Load the network.

4. Create an array- object-distance

5. Run an iteration
 Calculate the robot-obstacle distance and store in object-distance.

 Sort the array object-distance in ascending order and store in another array „p‟. The first 3
values of p give the location of nearest obstacles.
 Input these obstacles to the network and the network computes the angle to be taken by
the robot.
 As the robot reaches the target it stops.

37 | P a g e
20. EXPERIMENTAL RESULT: -
So various graphics below show the movement of robot towards its goal avoiding the obstacles
In this project we used feed forward back propagation neural network for the mobile’s path
planning. The functional aspects of the given neural network were to be experimented in some
workspaces. In the starting, experiments were to be conducted in a surrounding meant for the
training. So it was observed that the robot avoided all obstacles presented in the cluttered
environment and it traversed each path from the given initial position to the target point in
safely manner. Again experiment was conducted in the unknown surrounding for the robot.
Simulation results are shown in above. As these figures shows the navigation of robot in an
unknown environment. Simulation result has been shown in the previous one. The above
figure shows the navigation of mobile robot in an unknown environment also in experiment
showed that the algorithm is effective for obstacle of any shape, number and location. Also
the location of goal does not affect the effectiveness of the algorithm.

In fig 31 robot starts from the initial point towards target. and in fig 32 and 33, robot crosses
the obstacle according to threshold value.

Figure 31 Shows robot starts from the initial Figure 32 robot crosses the
obstacles presented environment

38 | P a g e
Finally in figure-34, the robot changes its orientation along with changing its obstacles position ,as a
result it followed the shortest path ,and in fig-35 the robot reached its goal avoiding any collision
with obstacles presented in the environment.

Figure 33. Experimental view of multiple no Figure 34.Experimental view robot moves in
of obstacle the multiple no of obstacle.

Figure 35. Robot moves in the multiple no of obstacles

39 | P a g e
21. CONCLUSION:-
Finally this project completed using the feed forward back propagation neural network under
proposing a path planning for navigating of wheel mobile robot by using certain no. of input
data used in the algorithm from the sensory data. And the from the simulation shows the
proposed was acceptable for the optimum path planning at the real time. In the experimental
data we found the same as that of simulation output done in MATLAB. By applying the
algorithm with simple controlling rule ,robot was able to staying away to un-optimized paths.
However, it may fails if the goal is closely surrounded by the obstacles on the both sides but
in last this issue can be judiciously controlled under different parameter

22. FUTURE SCOPE


 In this project, using feed forward back propagation algorithm used in a certain environment.

Also in future period, this work can be conducted using other types of neural network be like

radial basic function, cascaded network, regression neural network.

 different techniques just like particle swarm optimization, genetic algorithm, fuzzy logic etc.

Another variation could to be navigation could to be navigate the mobile robot in a more

complex environment moving target, moving trap, moving obstacles.

 Even in this algorithm one can try to reach the reach the obstacle instead of being heavily

surrounded by various obstacles.

 Here we can implemented multiple no. of robot with multiple target are to be considered

instead of single mobile robot with single target.

 So finally we can able to work using multiple robot to multiple no. of target problem

 Industrial application, defense sector, transportation, medical application this technology can

be implemented in future.

40 | P a g e
REFERENCE:-
1. Pham, D. T., & Parhi, D. R. (2003). Navigation of multiple mobile robots using a neural
network and a Petri Net model. Robotica, 21(1), :pp.79-93.

2. Kundu, S., Parhi, R., & Deepak, B. B. V. L. (2012). Fuzzy-neuro based navigational strategy for
mobile robot. International Journal of Scientific & Engineering Research, 3(6), :pp.1-6.

3. Pandey, A., Pandey, S., & Parhi, D. R. (2017). Mobile Robot Navigation and Obstacle
Avoidance Techniques: A Review. Int Rob Auto J, 2(3), 00022.

4. D. R. Parhi and Alok Kumar Jha,(2012) "Review and Analysis of Different Methodologies
Used in Mobile Robot Navigation", IJAAIES, 4(1), pp. 1-18.

5. Mohanty, P. K., & Parhi, D. R. (2012). Path Generation and Obstacle Avoidance of an
Autonomous Mobile Robot Using Intelligent Hybrid Controller. In SEMCCO,: pp. 240-247.

6. Mohanty, P. K., & Parhi, D. R. (2014). Navigation of autonomous mobile robot using adaptive
neuro-fuzzy controller. In Intelligent Computing, Networking, and Informatics, Springer, New
Delhi, :pp. 521-530

7. Sharma A., Chaudhary D.R .(2013). Character Recognition Using Neural Network,
International Journal of Engineering Trends and Technology, Volume4 Issue4- April : pp. 662-
667.

8. Huang.Y. (2009) Advances in Artificial Neural Networks –Methodological Development and


Application, Algorithms, 2,:pp. 973-1007.

9. Dadelahi S., Motlagh M.R.J., Shahri A.R.M. (2013). Obstacle Displacement Prediction for
Robot Motion Planning and Velocity Changes, International Journal of Information and
41 | P a g e
Electronics Engineering, Vol. 3, No. 3,: pp. 270-274.

10. Kumar T.S., Sivanandam S.N. (2012). An Improved Approach for Detecting Car in Video using
Neural Network Model, Journal of Computer Science 8 (10),: pp. 1759-1768.

11. Suresh A., Latha S.S., Nair P., Radhika N. (2014). Prediction Of Fight Or Flight Response
Using Artificial Neural Networks, American Journal Of Applied Sciences 11 (6),: pp. 912-920.

12. Hannan S.A., Manza R.R., Ramteke R.J.(2010) Generalized Regression Neural Network and
Radial Basis Function for Heart Disease Diagnosis, International Journal of Computer
Applications, Volume 7– No.13,: pp. 7-13.

13. Deepak, B. B. V. L., Parhi, D. R., & Praksh, R. (2016). Kinematic Control of a Mobile
Manipulator. In Proceedings of the International Conference on Signal, Networks, Computing,
and Systems ,Springer India,: pp. 339-346.

14. Parhi, D. R., & Deepak, B. B. V. L. (2011). Kinematic model of three wheeled mobile robot.
Journal of Mechanical Engineering Research, 3(9),:pp. 307-318.

15. Sharma A., Chaudhary D.R. (2013). Character Recognition Using Neural Network, International
Journal of Engineering Trends and Technology, Volume4 Issue4 : pp. 662- 667.

16. Parhi, D. R., Pradhan, S. K., Panda, A. K., & Behera, R. K. (2009). The stable and precise
motion control for multiple mobile robots. Applied Soft Computing, 9(2): pp.477-487

17. Mohanty, P. K., & Parhi, D. R. (2014). A new efficient optimal path planner for mobile robot
based on Invasive Weed Optimization algorithm. Frontiers of Mechanical Engineering, 9(4):
pp.317-330.

18. Suresh A., Latha S.S., Nair P., Radhika N.( 2014). Prediction of fight or flight response using
artificial neural networks, American Journal of Applied Sciences 11 (6),: pp. 912-920.

42 | P a g e
19. Sethi, R., Parhi, D. R. K., & Senapati, S. K., (2016) "Fault detection of cracked beams like
structure using artificial neural network (ann) approach", International Journal of Research in
Engineering and Technology, 5(13),: pp. 1-4.

20. Pezeshki S., Badalkhani S., Javadi A. (2012). Performance Analysis of a Neuro-PID Controller
Applied to a Robot Manipulator, International Journal of Advanced Robotic Systems, 9: pp. 1-
10.

21. Zakaria M.,AL-Shebany M.,Sarhan S.(2014) Artificial Neural Network : A Brief Overview,
International Journal of Engineering Research and Applications, Vol. 4, Issue 2( Version 1),:
pp.07-12

22. Kashyap, S. K., Parhi, D. R., & Sinha, A. (2009). Artificial neural network-a tool for optimising
mining parameters. In Proceedings of the ISRM-Sponsored International Symposium on Rock
Mechanics:" Rock Characterisation, Modelling and Engineering Design Methods"(2009),: pp
377-380.

23. Mohanty, P. K., & Parhi, D. R. (2014). A new real time path planning for mobile robot
navigation using invasive weed optimization algorithm. In ASME 2014 Gas Turbine India
Conference (pp. V001T07A002-V001T07A002). American Society of Mechanical Engineers.

24. Patle, B. K., Parhi, D. R., Jagadeesh, A. & Kashyap, S. K. (2017). On firefly algorithm:
optimization and application in mobile robot navigation. World Journal of Engineering,
14(1),:pp.65-76.

25. Deepak, B. B. V. L., & Parhi, D. (2013). Intelligent adaptive immune-based motion planner of a
mobile robot in cluttered environment. Intelligent Service Robotics, 6(3),: pp.155-162.

26. Rutt, H. N. (1994). Effect of visible and near-infrared illumination on the mid-infrared
transmission of silicon and germanium. In Window and Dome Technologies and Materials
IV .International Society for Optics and Photonics, Vol. 2286,: pp. 100-108.

27. Mohareri, O. (2009). Mobile robot trajectory tracking using neural networks. M. tech dissertation.
American University of Sharjah College of Engineering

43 | P a g e
28. Albarado, K. M., AL Arsenal, R., Wilson, A., AL Arsenal, I. R., & Wade, D. (2015). Advanced
neural-networks for mechanical health diagnostics. In Proceedings of the 5th American
Helicopter Society CBM Specialists Meeting.

29. Koskela, T., Lehtokangas, M., Saarinen, J., & Kaski, K. (1996). Time series prediction with
multilayer perceptron, FIR and Elman neural networks. In Proceedings of the World Congress
on Neural Networks,. pp. 491-496.

30. Glorot, X., & Bengio, Y. (2010). Understanding the difficulty of training deep feedforward
neural networks. In Proceedings of the thirteenth international conference on artificial
intelligence and statistics,. pp. 249-256.

31. Cigizoglu, H. K., & Alp, M. (2006). Generalized regression neural network in modelling river
sediment yield. Advances in Engineering Software, 37(2),.pp 63-68.

32. Gat, E. (1991). Integrating reaction and planning in a heterogeneous asynchronous architecture for
mobile robot navigation. ACM SIGART Bulletin, 2(4),pp 70-74.

33. Chen, S., Cowan, C. F., & Grant, P. M. (1991). Orthogonal least squares learning algorithm for
radial basis function networks. IEEE Transactions on neural networks, 2(2), 302-309.

-------------------------------------------------------------------------------------

44 | P a g e

You might also like