Professional Documents
Culture Documents
A SEMINAR REPORT
Submitted in partial fulfillment of the requirements
For the degree of
Bachelor of Technology
By
SARVESH SINGH
(Roll No. U09EE537)
Under the Supervision of
CERTIFICATE
This is to certify that the B. Tech. IV (7th Semester) SEMINAR REPORT entitled ROBOTICS AND
NEURAL NETWORKS presented & submitted by SARVESH SINGH, bearing Roll No.U09EE537, in
the partial fulfillment of the requirement for the award of degree B.Tech. in Electrical Engineering.
He has successfully and satisfactorily completed his Seminar Exam in all respect. We,
certify that the work is comprehensive, complete and fit for evaluation.
SEMINAR EXAMINERS:
Examiner
Examiner 1
Examiner 2
Examiner 3
__________________
__________________
__________________
DEPARTMENT SEAL
i | Page
ABSTRACT
This project gives an introduction about the robotics. It tells about why
robotics is important, where we need robotics and how we can build smart
robots.
Artificial intelligence using neural networks is also discussed in the
this report. Neural networks are special way of building algorithms that
are inspired by the structure of human brain.
Starting with the definition and application areas, working and
fundamental characteristics of robots are discussed. In working ,
fundamental sequence of steps are involved during the working of a robot
is discussed as :1) Perceiving its environment through sensors.
2) Thinking about the reaction.
3) Acting upon that environment through effectors
ii | Page
TABLE OF CONTENT
1
2
3
3.1
1
1
4
4
3.1.1
Properties of environment
3.1.2
Robotic sensing
3.1.2.1
Proprioceptors
3.1.2.2
Exteroceptors
3.2
3.2.1
Look up table
3.2.2
3.2.3
10
3.2.4
10
3.2.5
10
3.2.6
Neural networks
10
3.2.7
12
3.2.8
13
3.3
14
3.3.1
Effectors/Actuators
14
3.3.2
Types of actuators
14
3.3.3
Kinematics
14
3.3.4
Actuator Types
15
iii | Page
LIST OF FIGURES
1
12
13
A manipulator
14
A revolute joint
15
A prismatic joint
15
10
A spherical joint
15
iv | Page
1 | Page
Fig1 A robotically assisted surgical system used for prostatectomies, cardiac valve repair and gynecologic
surgical procedures
2 | Page
3 | Page
1) PERCEIVING ITS ENIRONMENT THROUGH SENSORS:Like any living body robots also need some kind of stimulus to react upon. Robots are
programmed to do certain things and are programmed to react in a certain manner
according as the programmer has programmed the robot.
Robots have sensors that receive a stimulus from the environment and reacts
according to the stimulus. Sensors acts as the input for robot. Robots think upon that input
and reacts according its thinking.
2) THINKING ABOUT THE REACTION :Now when the robot has received the stimulus from the sensors it has the responsibility to
act according to its knowledge provided by
the programmer. There are many ways how a robot can think and act accordingly. Here we
will give brief description about how the robot can act as a smart robot. We say a smart
robot is a robot that can learn things on his own and act. There are many ways of
programming the robot to make a robot smart.
One way to make a robot smart is to use neural network. Basically neural network
is the way of programming in which we give an opportunity for the program to learn. In this
programming style we imitate the way how a human brain works. We use the concept of
neurons and how each neurons are linked with other neurons.
4 | Page
An effector is any device that affects the environment, under the control of the
robot. To have an impact on the physical world, an effector must be equipped with an
actuator that converts software commands into physical motion. The actuators themselves
are typically electric motors or hydraulic or pneumatic cylinders.
Acting through the effectors requires the study of
1) kinematics of robot.
2) dynamics of robot.
3) force control.
4) motion planning(related with intelligence of the robot).
5) motion control.
DETAILED DESCRIPTION
3.1) PERCEIVING ITS ENVIRONMENT THROUGH SENSORS
As the robot is interacting with environment and it is interacting through sensors, thus we need to
understand environment and sensors
5 | Page
3.1.2.1) Proprioceptors
Robot consists of series of links interconnected by joints. Each joint is driven by an actuator which
can change the relative position of the two links connected by that joint. Proprioceptors are
sensors measuring both kinematic and dynamic parameters of the robot.
The usual kinematics parameters are the joint positions, velocities, and accelerations.
Dynamic parameters such as forces, torques and inertia area are also important to monitor for the
proper control of the robotic manipulators.
Kinematic parameters: Joint position sensors: They are usually mounted on the motor shaft.
Encoders are digital position transducers which are the most convenient for
computer interfacing. Incremental encoders are relative-position transducers which
generate a number of pulses proportional with the traveled rotation angle. These
gives relative position of the arms and in case of power failure it gives bad results as
it has lost the data of it relative position.
Absolute shaft encoders are attractive for joint control applications because
their position is recovered immediately and they do not accumulate errors as
incremental encoders may do.
Angular velocity sensors: is measured ( when not calculated by differentiating joint
positions) by tachometer transducers.
A tachometer generates a DC voltage proportional to the shaft' rotational speed.
Digital tachometers using magnetic pickup sensors are replacing traditional, DC
motor-like tachometers which are too bulky for robotic applications.
6 | Page
Acceleration sensors : are based on Newtons second law. They are actually measuring the
force which produces the acceleration of a known mass. Different types of acceleration
transducers are stress-strain gage, piezoelectric,capacitive, inductive.
3.1.2.2) Exteroceptors
Exteroceptors can be classified according to their range as follows:
- contact sensors.
- proximity (near to) sensors.
- far away sensors .
-Contact sensors are used to detect the positive contact between two mating parts and/or to
measure the interaction forces and torques which appear while the robot manipulator
conducts part mating operations.
Force/Torque Sensors
The interaction forces and torques which appear, during mechanical assembly operations, at the
robot hand level can be measured by sensors mounted on the joints or on the manipulator wrist.
This solution is not too attractive since it needs a conversion of the measured joint torques to
equivalent forces and torques at the hand level. The forces and torque measured by a wrist sensor
can be converted quite directly at the hand level. Wrist sensors are sensitive, small, compact and
not too heavy, which recommends them for force controlled robotic applications.
A wrist force/torque has a radial three or four beam mechanical structure. Two strain gages
are mounted on each deflection beam. Using a differential wiring of the strain gages, the four beam sensor produces eight signals proportional with the force components normal to the gage
planes.
Tactile Sensing
Tactile sensing is defined as the continuous sensing of variable contact forces over an area within
which there is a spatial resolution. Tactile sensors mounted on the fingers of the hand allow the
robot to measure contact force profile and slippage, or to grope and identify object shape.
The best known of tactile sensor technologies are: conductive elastomer, strain gage,
piezoelectronic, capacitive and optoelectronic. These technologies can be further grouped by their
operating principles in two categories: force-sensitive and displacement-sensitive. The forcesensitive sensors (conductive elastomer, strain gage and piezoelectric) measure the contact forces,
while the displacement-sensitive (optoelectronic and capacitive) sensors measure the mechanical
deformation of an elastic overlay.
Tactile sensing is the result of a complex exploratory perception act with two distinct
modes.First, passive sensing, which is produced by the cutaneous sensory network, provides
information about contact force, contact geometric profile and temperature. Second, active
sensing integrates the cutaneous sensory information with kinesthetic sensory information (the
limb/joint positions and velocities).
7 | Page
2. Proximity Sensors
Proximity sensors detect objects which are near but without touching them. These sensors are
used for near-field (object approaching or avoidance) robotic operations. Proximity sensors are
classified according to their operating principle; inductive, hall effect, capacitive, ultrasonic and
optical.
Inductive sensors are based on the change of inductance due to the presence of metallic
objects. Hall effect sensors are based on the relation which exists between the voltage in a
semiconductor material and the magnetic field across that material. Inductive and Hall effect
sensors detect only the proximity of ferromagnetic objects. Capacitive sensors are potentially
capable of detecting the proximity of any type of solid or liquid materials. Ultrasonic and optical
sensors are based on the modification of an emitted signal by objects that are in their proximity.
3) Far Away Sensing
Two types of far away sensors are used in robotics: range sensors and vision
a) Range sensors
Range sensors measure the distance to objects in their operation area. They are used for robot
navigation, obstacle avoidance or to recover the third dimension for monocular vision. Range
sensors are based on one of the two principles: time-of-flight and triangulation.
Time-of-flight sensors estimate the range by measuring the time elapsed between the
transmission and return of a pulse. Laser range finders and sonar are the best known sensors of
this type.
Triangulation sensors measure range by detecting a given point on the object surface from
two different points of view at a known distance from each other. Knowing this distance and the
two view angles from the respective points to the aimed surface point, a simple geometrical
operation yields the range.
b) vision
Robot vision is a complex sensing process. It involves extracting, characterizing and interpreting
information from images in order to identify or describe objects in environment. A vision sensor
(camera) converts the visual information to electrical signals which are then sampled and quantized
by a special computer interface electronics yielding a digital image
The digital image produced by a vision sensor is a mere numerical array which has to be
further processed till an explicit and meaningful description of the visualized objects finally results.
Digital image processing comprises more steps: preprocessing, segmentation, description,
recognition and interpretation. Preprocessing techniques usually deal with noise reduction and
detail enhancement. Segmentation algorithms, like edge detection or region growing, are used to
extract the objects from the scene. These objects are then described by measuring some
(preferably invariant) features of interest. Recognition is an operation which classifies the objects in
the feature space. Interpretation is the operation that assigns a meaning to the ensemble of
recognized objects.
8 | Page
9 | Page
10 | Page
e.g. suppose car is parked at a place and our robot driver has to decide where to go. It can
make an intelligent decision only if it knows the goal otherwise itll do whatever it is programmed
to do after the start. i.e. if it was programmed to take right when starts then itll take right. No
matter whether turning right is pushing it away from the goal or towards the goal.
Now reaching the goal from a particular position still has many possible ways. The question
arises how our robot will make intelligent decision to select the best way?? The answer to the
question is search algorithm and planning algorithm.
In search algorithm our robot thinks of all possible ways and maximize the parameters that
are concerned with the performance of the system. In our case of driver robot it will select the
shortest path or maybe consider about more parameters like fuel efficiency i.e. the shortest path
may have some hurdles due to which car cant be driven fast.
11 | Page
A layered feed-forward neural network. The circular nodes represent neurons. Here there are three
layers, an input layer, a hidden layer, and an output layer. The directed graph mentioned shows the
connections from nodes from a given layer to other nodes in other layers
The key concept of neural network are the weights that enables a program to get trained.
WEIGHTS
The weights used on the connections between different layers have much significance in the
working of the neural network and the characterization of a network. The following actions are
possible in a neural network:
1. Start with one set of weights and run the network. (NO TRAINING)
2. Start with one set of weights, run the network, and modify some or all the weights, and run the
network again with the new set of weights. Repeat this process until some predetermined goal is
met. (TRAINING)
TRAINING
Since the output(s) may not be what is expected, the weights may need to be altered. Some rule
then needs to be used to determine how to alter the weights. There should also be a criterion to
specify when the process of successive modification of weights ceases. This process of changing the
weights, or rather, updating the weights, is called training. A network in which learning is employed
is said to be subjected to training. Training is an external process or regimen. Learning is the
desired process that takes place internal to the network
12 | Page
2) ENCODING refers to the paradigm used for the determination of and changing of weights on the
connections between neurons. In the case of the multilayer feed-forward neural network, you
initially can define weights by randomization. Subsequently, in the process of training, you can use
the backpropagation algorithm, which is a means of updating weights starting from the output
backwards. When you have finished training the multilayer feed-forward neural network, you are
finished with encoding since weights do not change after training is completed
3)RECALL refers to getting an expected output for a given input. If the same input as before is
presented to the network, the same corresponding output as before should result. The type of
recall can characterize the network as being autoassociative or heteroassociative.
Autoassociation is the phenomenon of associating an input vector with itself as the output,
whereas heteroassociation is that of recalling a related vector given an input vector. You have a
fuzzy remembrance of aphone number. Luckily, you stored it in an autoassociative neural network.
When you apply the fuzzy remembrance, you retrieve the actual phone number. This is a use of
autoassociation.
neural network with a completely different type of neural network, the Hopfield network, and
present some simple applications for the Hopfield network
3.2)Types of Actuators
3.3)Kinematic
Fig7 A manipulator
14 | Page
Spherical Joint
3 DOF ( Variables - Y1, Y2, Y3)
15 | Page
3.4)Actuator Types
Electrical
Hydraulic
Pneumatic
Others
easy to control
from mW to MW
normally high velocities 1000 - 10000 rpm
several types
accurate servo control
ideal torque for driving
excellent efficiency
autonomous power system difficult
CONCLUSIONS :With the help of neural networks smart robots can be built that can help us in many ways
like in home, industries, hospitals, astronomy, military etc. Robots are our very good friend that
makes our life easier and happy. They can do works that would never be possible by human efforts
alone like in space exploration or industries etc. Robots can reach to many places that might be
dangerous for human like close to nuclear reactor chamber to check the safety.
Robots are fast and accurate. These key properties reduces human effort to a great
extent and increase the productivity and accuracy of the product.
Making robots that has its own thinking is still a matter of research. Although we
have made robots that can learn on its own , still it is the topic of interest to make a robot think on
its own. Whatever be the status of robots in our society but one thing is sure that in near future
well be greatly surrounded by robots.
16 | Page
references
1) C++ Neural Networks and Fuzzy Logic by Valluru B. Rao(third edition)( published by M&T Books, IDG
Books Worldwide, Inc. on 06/01/95)
2) Artificial Intelligence(A Modern Approach) by Stuart J. Russell and Peter Norvig(second edition) (published
by Prentice Hall, Englewood Cliffs, New Jersey 07632)
3) CEG 4392 Computer Systems Design Project
4) http://www.melbpc.org.au/pcupdate/2205/2205article10.htm (4 nov 2012)
5) Introduction to Robotics by Dr Suprava Patnaik
17 | Page