Foundations of Cognitive Robotics

INDEX
S. No Topic Page No.

Week 1
1 Introduction to Cognitive Robotics (Private) 1
2 Smart Materials-I (Private) 19
3 Smart Materials-II (Private) 47
4 Smart Materials-III (Private) 75
Week 2
5 Architecture of the Brain 101
6 Architecture of the Brain- Continued 129
7 Nerve Cells 163
Week 3
8 Introduction to Synchronisation Models 186
9 Synchronisation models (continued) 213
10 Introduction to EEG 250
Week 4
11 Theories of Intelligence-I 280
12 Theories of Intelligence-II 295
13 Kuramoto Model 322
14 Child Robot Interaction 348
Foundations of Cognitive Robotics
Prof. Bishakh Bhattacharya
Department of Mechanical Engineering
Indian Institute of Technology, Kanpur
Lecture – 01
(Refer Slide Time: 00:21)
Welcome to the course on Foundations of Cognitive Robotics. I will first tell you about
the course evaluation procedure. You will have a final exam, which is optional, but with
a fee, you can actually go for these. E-certificate will be issued to only those who would
register and write the final exam provided the average assignment score will be 25
percent of the average of the best 3 out of the total four assignments. The exam score
will be 75 percent of the proctored certification exam score out of 100. The final score
will be the average assignment score and the exam score; however, it should be greater
than or equals to 40 out of 100.
Now, you will be eligible for an E-certificate only if the average assignment score is
greater than equal to 10 out of 25, and the exam score is greater than equal to 30 out of
75. If one of the above two criteria is not met, you will not get an E-certificate even if the
final score is greater than equals to 40 out of 100. So, having said the course evaluation
procedure, let us start the course on cognitive robotics. (Refer Slide Time: 02:01)
1
I will first talk about the reference books that I will be using in this course. The first
important book is by Hooman Samani, which is published by Taylor and Francis; the
title of the book is Cognitive robotics. And then we also would follow the basics of
Neuroscience from a standard book like Dale Purves. And Cognition, Brain and
Consciousness some Introduction to Cognitive Neuroscience by a beautiful book from
Bernard J. Baars and Nicole M. Gage, which is from the Academic Press.
And the other important book is from Rolf Pfeifer of MIT Press and Josh Bongard that is
on How the body shapes the way we think – A New View of Intelligence. And also, you
need some basic knowledge of Control Systems, particularly the modern controller AI-
Based Approaches. So, you can follow books like the book of Jitendra Raol, Ayyagari,
or other similar books in this direction.
I will now tell you a little bit about the motivation beyond these subjects of cognitive
robotics. As you can understand, cognitive robotics is an interdisciplinary subject so, it
draws inspiration from computer science, from psychology, and, of course, from
robotics. How this intermixing has created this new subject, cognitive robotics, and the
primary motivation of it what we have new in this particular course, I will give you this
basic introduction in the first lecture. (Refer Slide Time: 04:11)
2
The motivation behind this cognitive robotics would come from this very famous quote
from Virgil, which is Felix, qui potuit rerum cognoscere causas, which means fortunate
was able to know the cause of things. As homo sapiens, you would little understand the
significance of how fortunate we are that we can know or we can pursue to know the
cause of the things.
Now, if you watch carefully that these to know something in Latin, it is called
cognoscere, the root of the word of cognitive, which has two parts in it that the cogn part
of it is the learning knowledge and the process of it. So, cognitive referred to the process
of knowing or learning something. In fact, until the mid-15th-century cognicioun usually
3
would mean the ability to comprehend, mental act or process of knowing. In 17th
century the meaning started to get extended to include things like perception and
sensation. (Refer Slide Time: 05:49)
Now, when we talk about cognitive robotics, what we mean by these? Well, cognitive
robotics would actually mean if we literally try to translate, would suggest that how to
create the thinking robots. Alternatively, if you try to, the word thinking will sound a
little bit vague to you. So, if you try to dilute it, it would mean how to endow intelligence
to a robot.
Now, while doing it, we would not only consider homo sapiens. We would consider that
there is this process of thinking presently even in animals. So, as a starting point, while
developing automated information processing, we will consider how this cognition can
happen in animals and the interesting characteristics of that cognition that we can
actually impart in robots. This is opposed to a more traditional artificial intelligence
technique.
4
The other important aspect here is that when we say about the robotic cognitive
capability, it is not only in terms of some sensing or some actuation; we aim to include
characteristics like perception processing, attention allocation, anticipation planning, and
complex motor coordination, reasoning about other agents. Here, this word agent would
mean broadly about living and nonliving things you know nonliving say robots and
perhaps even about their mental states.
That is our objective: Can a robot comprehend its mental states, and can a robot have
these kinds of you know the animal or human-like capabilities of attention allocations?
So, that is what is the goal of how we can develop such a thing in cognitive robotics.
Now, I would just like to show you a small video. Many times we think that intelligence
is only possible in higher-order animals; I mean mostly mammals, and even among the
mammals, I would say primates and the homo sapiens, but that may not be the case. Let
us look into the simple example of a much lower order animal biologically that is about
an octopus and let us see that how this octopus can have attention can have a perception
about a problem, and can think about a strategy to escape.
So, you see here is a bottle where we have kept an octopus, and you can see that I have
taken it from an interesting YouTube link, and you can see that the direction is showing
that if there is a movement from these angles as you can see already that movement is
5
happening. The tentacles of the octopus are moving; it has understood which direction it
has to move, and it is doing some reverse direction also.
So, to understand that which way it has to apply the torque and it has wholly understood
it is applying it perfectly now if you wait and see that how it is now going through this
entire process of unlocking itself, it is fascinating. You can see that it is moving its limbs,
and it knows that this unturning is over. So, it can take the whole lid out, and you see it is
taking out all its limbs, and it is fascinatingly escaping.
Can a robust AI robot plan for such a fascinating escape? I will give you a one more
fascinating example. This time it is about the fish, and it is about the spawning of the fish
eggs. This particular example I have taken from an Amazon rain forest and just observe
the amount of intelligence. In this case, there are two fishes, the male and female, which
are cooperating in order to make the spawning of the fish possible not in the water, but
outside the water. Let us look into that example.
This will tell you that intelligence is not just the property of higher-order animals. So,
this is copella arnoldi, commonly known as splashing tetra, which lays its eggs outside
of water, and the advantage of doing it is many. Like you know, the eggs can develop
faster in outside water where you are getting more oxygen supply; because the many
times in such rain forests, the oxygen demand biological oxygen demand will be quite
low because of algae covers. Also, it helps to avoid predators.
6
Now, let us see we know the advantage of how it is actually executing it. So, you can see
that this water this splashing tetras the spool of these splashing tetras you can see both
males and females. Now, you see how this spawning is happening in them. So, they are
still inside the water and then looking for a target, they have to jump out of the water. So,
they are looking for a suitable target with a fisheye view. They are searching which plant
will be suitable for spawning.
And, after some movements, various directions, they have fixed it fixed the leaf. You can
see both the male and the female are spawning their legs are spawning their eggs. The
male would remain for little more time. They will go back. Some eggs you can see
already. Again they will be back. They will be spawning the eggs and you can see how
they are actually spawning the eggs phase by phase; they are populating it on the leaves.
So, they have chosen the target, they have come out of water and they have spawned the
eggs all over the leaves. Now, that is not all, you know a fish cannot live without water.
The same thing is true for the eggs. You have to keep it moist, atleast. So, what it is
going to do is that it is going to find out the position of these leaves precisely and you see
how it is splashing water. So, that all these eggs would remain moist and they can nicely
grow inside. You can see already the life has started.
7
You can see these small small the fishlets are actually inside.
The moisture is helping them, and these male tetra is continuously moisturizing until
these fishes as growing to the extent that with the help of further water, they can start
their journey back into the stream. So, do we not call it as an excellent example of
intelligence, but this is coming from fish from a relatively biologically lower-order
animal.
8
So, that is the basis when we say that when we talk about intelligence and when we are
talking about imparting intelligence, it has to happen from the animal level itself. Now,
let us talk about some basic definitions of cognitive robotics that you will find in the
open literature.
The general perception of cognitive robotics is given very nicely by Hooman Samani,
and he has said that you can consider it to be an approach of creating artificial
intelligence in robots by enabling them to learn from and respond to real-world
9
situations. You are not preprogramming the robots that, if this happens, you react in this
manner. You are giving it some clues that are based on the real-world situation about
how you can make your decision.
So, this is sometimes also called that how to interact with the unconstrained
environment, and you need, of course, a suitable control architecture in order to interact
with such an unconstrained environment, acquire knowledge and use this experience for
your further action. In other words, you should generate predictive capabilities to
augment the immediate sensory-motor experiences. So, what kind of control structure
can actually give us this?
Well, one very famous control structure was proposed by John Boyd which is based on
observe-orient-decide-act; in other words, the OODA loop. If you look at these carefully
that there are these four sections observed section in which you are carrying out
observations unfolding the circumstances. Check what is the information that is coming
to you and then take this information's take it forward.
Do a feed-forward and look at it from the prism of various say cultural traditions, various
analysis, and synthesis that is available with you the previous experiences that you have
understood, any new information whether you perceive, whether if there is a genetic
heritage of this kind of information is that you have already acquired. Based on these
orientations, you feed yourself forward, and you take individual decisions.
10
Now, whatever decisions you take, you also feed them back in the observation process.
Let us say you have decided that this is observation is more like a cultural observation.
Then you modify your observation towards the cultural direction, or you want to
correlate it with your previous experiences. So, you observed more keeping your
previous experiences in mind, and also, based on your decision, you feed it forward
further and take some actions.
Now, as soon as you are taking the actions that also you have to feed it back so that you
are having more unfolding of your interactions with the environment. As you are
unfolding your interactions with the environment, you are learning more, and you are
getting newer observations, and you are feeding it through this entire circuit. And, doing
this process, particularly in the orientation phase, you continuously give the implicit
guidance and control that are there based on the system you have.
Now, these are some of you know control architectures that can help you in an
unconstrained environment. We will talk more about this type of control architecture in
the future.
Now, there are three directions of development for the robot. The first question is that
can we consider this kind of cognitive robots as a quasi-person? Many people think that
since these robots have a short of intelligence, thinking process, perceptions, in other
11
words, they operate in a manner that is identical to that of humans. So, should they not
have some legal rights, and should they not have some duties?
You know, even though in certain philosophies like Judeo-Christian philosophies, this
kind of humanism outside the homo sapiens we call it transhumanism is generally treated
as a taboo, but certain other cultures like Japanese scientists they are perfectly
comfortable with the idea of robots as like living entities. You might be knowing that in
Japan and the robots have participated even in social marriage. So, this is a concept that
is coming up that with the increased level of intelligence, human-like properties like
attention perceptions, you may consider a robot to be a quasi-person.
Some people are a little less optimistic they can they are ready to consider it like a quasi-
animal, and they consider that well we need to give an intelligence like animals.
However, the robot must have an owner, and the owner will be held responsible in case
of any accident or mischief, just like if you have a dog and this dog creates some
problem and is the owner of the dog who is held responsible, similarly is the owner of
the robot who will be held responsible.
Now, some more people think that it is neither an animal nor a person, but it is simply a
product, and hence like for other products, the manufacturers are responsible; similarly,
for the robots, the manufacturer or the operator will be responsible. So, there are these
three different viewpoints towards the direction of the cognitive robot. I will now show
you one example of one of the latest humanoid robots, which has been developed by
Honda, also called the ASIMO robot.
12
Now, look at this particular work, which is reported by Honda very carefully and think of
it that in which category you can place this robot based on this action that you are going
to see. You see, the robot here ASIMO is bringing something that looks like a flask,
which is having some drinks in it and a glass. Look how carefully it has come to the
unconstrained environment without any human hand went to a proper position.
And, then, without falling, it was taken a sidewise movement and you standing in front
of the table that he has brought the cart with the drinks that he has brought. Every time
you will be now seeing this, you should also compare the performance of it with the
octopus example that I have shown. You see how it has held; it now knows how to
unscrew the bottling process; it is opening up the bottle perfectly, keeping in mind that it
may have some fluid that should not spill out.
So, it is applying just a good amount of torque to open it up. It has opened up and kept
the lid first very safely, and then it has (Refer Time: 24:12) picked up that glass. Now, it
has to decide about the proper angle proper orientation and gradually fill the glass. That
is what it has done with the help of its fingers and the visual scanning and the touch
sensors and the actuators. So, now that it has filled the glass, it is doing a social duty. So,
it has given this water or the drinks to a person.
So, based on these interactions, you have to tell me that in which class you will be
keeping this ASIMO robot? Would you like to give it a right just like a homo sapiens,
13
transhumanism, or would you like to treat it like an animal? Suppose, if it does anything
wrong, suppose if you know, spill off this fluid on the lady when it is serving.
So, you know, would you help the operator the people who have actually developed the
robot responsible? Or would you help the person who has designed who is running the
robot the operator to be responsible? Or you would think no, you know the robot is
autonomous on his own so, it has his own personality so, it is the ASIMO robot who
himself or herself is responsible for that?
So, this as the intelligence is propagating. As we can embed more and more human-like
behaviors into the robots, these are becoming our fundamental legal issues.
Now, as you can see that in the last example of the robot, some embodiment has taken
place in the robot. This word embodiment is very important because there was a group of
people who are from the strong AI group, and for them, the artificial intelligence does
not necessarily require any hardware to be present. Let us say if there is a network of
computers according to the strong AI group, it can be an intelligent agent.
However, according to the embedded group, this embodiment is very very important in
order to impart intelligence to the system. So, here the psychology and computer science
people have to merge with the robotics brand in order to give this embodiment into the
14
cognitive robots. Now, there is a very famous scientist Searle, was given a strong
argument in favor of this type of embedded AI's which is also known as weak AI's.
And he has proposed a sort of a Chinese game in which he has considered that let us say
there will be a person in the Searle's Chinese game. He has considered that let's say there
is a monolingual person who only knows English and he is in a closed room, and you are
supplying him with all the Chinese letters and some Chinese scripts and some
instructions which tell that the syntax says that how each one of these letters is part of the
script.
Now, if you write something based on these Chinese symbols in the form of a script, can
we call him to be a Chinese writing person? If he speaks something based on such
instructions, can we call him to be a Chinese speaking person? Because the person does
not know anything about these symbols. He only knows that the syntax says that is given
to him.
So, the fallacy is that if you have a robot which is having these symbols as the first batch
and these script as the second batch and these third batch as how to join them the
instruction set, based on that, we cannot actually call that the robot has some Chinese
personality or similarly the robot has intelligence.
Today you have, for example, Alexa with us, very popular. Now, can we call Alexa to be
an intelligent being because, in the back end, it does not understand each one of the
symbols with which it is trained? So, Searle's weak AI logic is that in order to have an
intelligence you need to explore with the environment. So, you cannot call this kind of
instruction based intelligence to be a proper intelligence, you need embodiment in order
to have intelligence.
So, if you look at Searle's arguments, you can put it in a way that every organism with
this intelligence should have a particular biological structure, and this structure, under
certain conditions, is causally capable of producing perception, action, understanding,
learning and such other intentional phenomena. So, this is the significance of
embodiment in cognitive robotics.
15
Now, there are five such possible categories of an embodiment; the first one is just
simple structural coupling between the agent and the environment. Moreover, when I
told you when I would talk about an agent, it can be any living or nonliving thing which
has the capability of interactions. So, in this sense, if you consider Alexa so, there is a
structural coupling between Alexa and the environment because if the environment
changes, you give some instructions, Alexa understands and changes its behavior. So,
there is structural coupling.
There can be further improvement of structural coupling in terms of historical coupling.

So, you can improve your AI up to that level that you can keep a track record of a history
of agent environment interactions. So, you can learn from it, and it can adapt itself based
on it. But, with that Alexa example, there is no physical embodiment; on the other hand,
for the ASIMO, there is a physical embodiment that is there.
Now, this physical embodiment can have a sort of an initial group of embodiments like
organismoid embodiment, which is not precisely an organism, but organism like bodily
form like the humanoid robot. So, here the physical bodies have at least to some degree it
has a similar form like an organism and it has the sensory-motor capacities with it. So,
that should be the starting point of cognitive robots.
And, the ultimate point is like organismic embodiment, which is like autopoietic;
autopoietic are the systems which can self create itself which can maintain itself on its
16
own, just like a living system. So, that is the final goal of developing this kind of
cognitive robots. However, in order to have cognitions, you must have a physical
embodiment, and it must have organismoid embodiment, and the final goal is to develop
an organismic embodiment.
So, this is what we will perceive in our course; that means this building blocks will be
discussing the organismoid embodiment and how to develop each for cognitive robotics.
17
Now, for the cognitive robotics development, we will talk about three different types of
strategies like physical interaction of the robot; how can a robot develop complex
coordinated movements while exploiting the fundamental dynamics of the system, the
robot has to be aware of its basic dynamics. Furthermore, then the social driving factors
and can provide a robot bootstrap itself and try to know on its own.
And, finally, all these three properties how can you integrate them in order to get a kind
of a complete system out of it which is capable of having physical interactions, social
interactor and actions, and developments. So, this is what we will be exploring in the
course as we will have the future lectures in these directions.
Thank you.
18
Lecture – 02
Welcome to the 2nd lecture on Foundations of Cognitive Robotics. In the last class, I
talked about a basic introduction of cognitive robotics, where I told you that what is the
source or inspiration of these term cognitive. And how cognitive robotics is different is
unique from other branches of robotics. I also told you if you remember that what are the
different types or characteristics of interactions that are possible in a robotic system in
order to develop an intelligent robotic system.
And we have made these graduations from strong to weak artificial intelligence strong to
weak AI. Like for a robust AI system, I told you that embodiment is not a precondition
you can have simple coupling; you can have historical coupling with the environment,
and that will be good enough for a strong ai system. Whereas, for a weak AI system
where we are talking about things like cognitive robotics, the physical embodiment is
very much required.
Now, there are two subcategories under a physical embodiment, which are organismic
embodiment and organismoid in embodiment. And I will talk about these different types
of embodiments in this particular lecture. Also, I will tell you that how we can develop
such cognitive robots using a certain special class of materials called smart materials.
And thus, we will be able to build up organismic robots, and there will be a hope of
developing cognitive robots through that.
19
So, in this second lecture on the foundations of cognitive robotics, we will focus on
organismoid embodiment. So, all other couplings in a structural coupling or historical
embodiment and the very futuristic embodiment, which is organismic embodiment these
are the things which will not focus at this moment. Organismoid embodiment, which is
the basic requirement for cognitive robotics, will be in focus first.
Now, when we talk about organismoid embodiment, what is it that comes into your mind
first? Remember, in our last lecture; I have shown you a small video from Honda on the
20
performance demonstration of a robot, which is a humanoid robot; that is the ASIMO
robot.
Now this particular robot, how do you classify it would you classify it to be an
organismoid robot. And if you classify this to be an organismoid. So, how much human-
like is this particular robot?
So, let us look into Honda specifications and let us see how much humans like is this
ASIMO robot.
21
So, we will talk about the building blocks, and in this direction, let us look into what
goes into the ASIMO robot. Now Honda engineers have created ASIMO with 34 degrees
of freedom. This 34 degrees of freedom essentially means that there are 34 independent
ways in which the ASIMO robot can move it is different limbs or body parts. And that
actually helps it to walk and perform many tasks much like a human, exactly not like a
human, but much like a human. You have seen in the last class in the last video that I
have shown that how it was delivering a glass of dreams to the lady.
Now, this ASIMO, apart from the fact that it is made of 34 degrees of freedom
structurally it is made of a lightweight material, which is a magnesium alloy structure,
and it is combined with powerful computers.
And it is these two combinations; that means, by choice of lightweight material and by
choice of these 34 degrees of freedom, we are able to see the performances of the
ASIMO robot. Well, you will be interested to know that in comparison to the ASIMO
robot, what the degrees of freedom to a real human are. And then, only we will be able to
judge the performance of the ASIMO robot.
If you find out how many degrees of freedom are there in human. You will see that, for
example, our head in the neck joint can have 3 independent movements; up or down, left
or right, and rotation, so it has 3 degrees of freedom.
22
Similarly, in the shoulder, it has 3 degrees of freedom again. The elbow joints have 1
degree of freedom, the wrist joints, in comparison, have much higher degrees of freedom
7 degrees of freedom in each one of the wrist, so for the two wrists that we will make it
14 degrees of freedom.
The maximum number of degrees of freedom is there on the fingers. That is why our
fingers know the in comparison to ASIMO’s; ASIMO looked like so much like a robot.
Because if you look at our fingers, we will be having 13 degrees of freedom in each
hand, so that means a total of 26 degrees of freedom.
Then the hip is having 2 degrees of freedom, the crotch joint in the leg is having 3
degrees of freedom, the knee joint 1 degree of freedom, and the ankle joint will be
having 6 degree of freedom each, which means; two ankle joints will give you 12
degrees of freedom. So, in all, it is about 57 degrees of freedom.
So, compare our 57 degrees of freedom to the ASIMO robots 34 degrees of freedom, and
you to be able to know that how much steel we have to achieve in the robotics. There is
also some more thing to compare; how about the weight of the ASIMO robot with these
34 degrees of freedom and how much of power it requires. Let us go back to the
specifications of the Honda robot and let us check that.
23
If you look at some interesting specifications of ASIMO, which I gathered from the
Hondas website; it’s height is about 4 feet 3 inches; it is just like the height of a child.
Weight is 50 kg well, that is, little high in comparison to a child, walking speed and
running speed are slightly on the higher side as we can see 2.7 kilometers per hour.
Walking speed and running speed is about 7 kilometers per hour, gasping force a little bit
in the lower side, about 0.5 kg or 500 gram per hand. I am sure that a kid can lift slightly
more weight than 500 grams. Actuator wise this 34 degrees of freedom comes in the
expense of 34 servo motors.
In fact, that is what gives the maximum cost to this robot. And the control unit is having
the walking and operating control unit wireless transmission units. It has sensors, for
example, in the foot, it has torso level sensors in terms of gyroscopes and acceleration
sensors.
And again, the negative side is the power it needs a 51.8 volt lithium-ion battery; it is
very substantial. Also, lithium-ion battery is quite dangerous due to its weight. And the
operating time that we get that is also very low if you look at it is just 1 hour. So, what
we can figure out from this specification is that there are at least three points in which
ASIMO is still far away from a good organismoid robot. One is in terms of course, the
degrees of freedom it has 34 degrees of freedom whereas you know in nature the human
that it mimics that has 57 degrees of freedom.
And it has a substantially high weight of more than 50 kg despite using a very advanced
class of materials also, it needs a very high power, and it can work only up to an hour or
so. So, these are some of the limitations of the best current available robot, which is close
to an organismoid robot.
So, all these things, what does it point to? It tells us that we need to improve in terms of
the overall weight or the bulkiness of the robot and also in terms of the overall power
consumption of the robot. So, what is the solution in front of us?
24
If we look into the new developments of muscles in this new age, we will see that there
is a new branch of robotics first expanding, which is also known as soft robotics.
Now, under soft robotics, the robotic arms are meant to stretch and squeeze at every
point along their length. Furthermore, their movement cannot be described with simple
geometric relations because it is too complicated it is a large deformation motion much
unlike the way we are used to doing it; let us say for a rigid robot system with the help of
25
kinematic chains. It is very difficult to describe this kind of a flexible robotic link with
this type of robotic system.
So, here you can see an octobot or a octopus morphology you know kind of thing which
is mimicked in an octobot system. Now one of the thing if we compare between the
robots and the animals, we would see that; robots traditional designs of steels is far
stronger than any natural naturally living material. Let us say the nearest will be the teeth
or the nacre and if you go for softer materials like wood or ABS plastic, then possibly it
will be closer to the cortical bones or a squid beak.
If you make your robot even softer like; high-density polyethylene or

polytetrafluoroethylene or polyurethane, then we will come closer to cancellous bones,
ligaments, tendons, etcetera. If you can make your robot further soft like silicone
rubbers, then you will get some of the muscles like cardiac muscles, and even in a softer
range almost close to hydrogels, you will start to get the skeletal muscles.
So, thus this you know comparison is aptly telling us that the direction to which we have
to take our soft robotics, but Is weight is the only concept? No, we have to also keep in
mind that we have a constraint of the power; so many servo motors, which makes it very
expensive to afford.
26
Now you look at the different variations of soft robots like; this one is a caterpillar
inspired locomotion and then a multi gait quadruped robot again based on soft material.
And active camouflage one you can actually play with metamaterials and optically
coupled properties to develop active camouflage.
Walking in hazardous environments type of robots and then worm inspired locomotion
in robots and then particle jamming based actuation rolling powered by a pneumatic
battery and hybrid hard and soft robot kind of combinations.
And then we have this snake inspired locomotion, jumping powered by internal
combustion. Then there is a manta ray inspired locomotion and an autonomous fish. All
these are various types of soft robots that are coming up very fast and with the intention
of how we can reduce the weight of the robot.
But also as I told you that functionally you need to improve or replace these classical
servo motors; in order to make it more sustainable, feasible and low power consuming.
So, we need to think about the development of muscle-based flexible robots. The first
stage towards that is the development of pneumatic artificial muscles or PAM’s.
Moreover, this PAM’s were the first step where the air pressure is used in terms of
compressing and getting a muscle like motion from the system.
27
Now, however, the compression itself is quite heavyweight. So, we need to change from
a compressor-based system to let us say a solid-state kind of actuator system. And one of
the very interesting candidates towards this is ionic polymer matrix composites or
IPMC’S.
These are a general class of electroactive polymers like here you can see; a four-finger
electroactive polymery gripper which can grip up to the extent of 10.5 gram of rock and
this is used by NASA in an EAP Gripper for you know extra outside the earth in one of
the planets planetary missions to collect the debris; I can show you that how this EAP
would work.
You can see that such a large deformation you are getting from the electroactive
polymers. And then how each one of the four fingers has to work together with the last
deformation, and then they have to grab the object.
28
Which is white, you can see that it has to it has deformed a lot. So, that is something
without the use of any motor is challenging just by deforming it to the help of the direct
transformation of electrical energy to mechanical energy and when placing it at a
different location.
You can do much more complicated works with the help of this kind of IPMC rocker.
Today we have an even more impressive group of materials that can perform amazing
fits, and they can be easily embedded and integrated in order to form actuators that
would go very well with soft robotics. And can make our dream of developing cognitive
robots this are the groups of materials which are known as smart materials.
29
Now, before we talk about such smart materials, we have to keep in mind that the
mechanisms also have to be made compliant. For example, here you can see that this is a
gripper and the gripper ends you can see a large deformation you can make with the help
of these closing down between the two.
Now what here has been done using the fingers can also be done with the help of smart
materials, and then you can get a substantial deformation. So, even though the smart
materials can be used at the back end in order to get a large amplified motion, you need
to make such contraptions of compliant mechanisms for grippers.
30
Now, let us talk about smart materials. In this group of materials, the features of such
materials are that; they are functional in nature, and the basic energy forms that get
interchanged in such materials are between thermal energy to mechanical energy,
between electrical energy to mechanical energy, magnetic energy to mechanical energy,
or sound energy to mechanical energy.
As I told you earlier also that energy can only be transformed. So, in this case, we have
to manipulate between these transformations in order to get our intended type of
31
actuation or type of sense. Now this material's behavior, which is why they are so much
important for soft robotics, is that they are analogous to biological materials.
Like there the nature of adaptivity, cellular function, self-sensing, actuation, and control.
And moreover, these sensors and actuators are highly embeddable. So, as different
energy forms are interchanged in smart material. Let us look into these different types of
smart materials from an input-output classification point of view.
There is a broad group of smart materials like; piezoelectric. In this case, the input is
electrical energy and the output is mechanical deformation mechanical energy. And then
we have magnetostrictive materials where the input is for the use of for the group of
ferromagnetic materials; the input is magnetic energy.
And the output is once again mechanical energy in the form of mechanical stress and
deformation. And then for shape memory alloys, it remembers it is original shape after
being deformed and it can go back to the original shape.
So, this is by actually exploiting the change from thermal energy to mechanical energy.
And we also have some special groups of rheological materials which are like electro
magneto rheological materials where; it is the viscosity that changes in the material. So,
thus there are these different types of materials and if we look at these from an input
output point of view.
32
The inputs can be categorized in terms of the electric field, magnetic fields, stress, heat
or light for that matter even light can be used as input. And outputs can be in terms of
charge, magnetization, strain, temperature, and light. Now, if you look at these, let us say
like in a matrix form, then the diagonal terms are something which are the material
properties which is very common in the material.
So, they are not very smart properties, but it is the off-diagonal terms. Particularly this
green row column and this gray row, which is of very useful nature to us. For example,
the electric field as an input and as an output it is creating strain is what is known as
reverse piezoelectricity that is what is used in piezoelectric materials or even in the
IPMC example that I am showing you a few minutes before. And this is one such kind of
an off-diagonal property of a smart material.
Similarly, if you can achieve the deformations using magnetic energy, then it is called
Joule effect magnetostriction so this is also another off-diagonal term. For a heat for the
shape memory alloys, it is thermal you know thermal-induced phase transition, and for a
light, it is photostriction. Now, this green column this is corresponding to the actuator
development.
Many of these materials can also be used as sensors, and here this gray row is significant.
Like if you use, for example, stress as the input mechanical stress and by exploiting the
33
direct piezoelectric effect, you can build up a sensor that will give current or charge as an
output.
Similarly, using the Villary effect in some materials, you can get magnetic field change
magnetization change as an output. Change of stress creating strain is a very familiar
thing of elastic modulus; there is no smartness here, but if I change the stress and if the
temperature changes, there is a thermomechanical effect that is related to. Similarly, if
the electric property changes, there is a photoelastic effect, which is also equally
attractive in terms of developing sensors.
Now, if you compare these materials in terms of actuators and compare them with the
traditional actuators like; motors, like pneumatic actuators or hydraulic actuators, we are
interested to know that in which way smart materials are better in which way these smart
material based actuators will be outstanding in comparison to all the traditional actuators.
If you look at the response times, you can see that all these greens are the traditional
actuation system, which is in terms of seconds or milliseconds. And here, for smart
materials, you get it in terms of 0.1 milliseconds. So, it is even one-tenth of a
millisecond, and that shows why they are swift.
So, their response time is faster, which makes them closer to this kind of organismoid
robots. And also, another exciting property their accuracy, they are far accurate then
34
most of these actuators that we are talking about at least 10 to 100 times more accurate
these smart materials are. So, in terms of the speed, in terms of accuracy, they are several
orders of magnitude better than the traditional actuators.
Now the flip side is that the displacement is lower or the force is also lower, but these
two things you can actually compensate by using suitable mechanical amplifiers or by
distributing the actuators the smart material based actuators. You know, in terms of just
like our muscles, you can distribute it, and you can extract more energy from the system.
Now, if you look at the piezoelectric materials because in this talk and today’s lecture I
am mostly focusing on piezoelectric materials. As you can see here that if I apply force
and if I deform the piezoelectric material, you are getting a voltage that is produced in
the system.
So, this is what is the direct piezoelectric effect. Where you are applying the force
mechanically, and you are getting the voltage out of it.
35
If we look at these piezoelectric materials close, we will first see that they are not that
very new; it was developed around 1880 as you can see that Pierre and Jacque Curie has
developed the first found out this piezoelectricity which is electricity from pressure and
contemporary electricity.
Where to like; electricity forms static electricity from friction; it was contact electricity
or electricity by heating up crystals, which was pyroelectricity. So, when they found out
that even pressure in a crystal can generate electricity, this piezo means pressure; that is
where this new group came into the picture.
36
And as a result of that Pierre Curie and Paul Jacques curie both of them got Nobel Prize
in physics in nineteen o three on the direct piezoelectric effect. The direct effect signifies
if you apply pressure on the piezoelectric material, you are getting electricity out of it.
Similarly, at a later stage, Gabriel Lippmann theoretically predicted that the reverse is
also possible; if you apply electric field, you will get a mechanical deformation for which
he got his Nobel Prize in 1908 that is on the reverse piezoelectric effect.
37
Now why piezoelectric material is so exotic what is there? So, special in a piezoelectric
material that actually gives it this property that as you apply the mechanical force, you
are getting an electrical charge out of it.
So, one of the very important groups of piezoelectric material is these perovskites,
named after a Russian geologist which was discovered roughly around 1950 or so. And
these perovskites are generally ternary structures; that means, it is made of three
components like in this example barium titanium and oxygen. So, there are three
components that form this piezoelectric crystal.
Now, what is so special about this component is that; if you look at the crystal structure
of this material, you would see that barium atoms are at the vertices of this tetrahedral
these tetragonal structures. And you would see that the oxygen atoms are at the face-
centered everywhere, and the titanium, which is the source of this kind of
piezoelectricity, is deep inside the geometric center of this tetron.
Now, this titanium is very precariously positioned because if I apply a little bit force on
this crystal from any direction, the titanium will get displaced from it is center at the
moment. It will be displaced, as the titanium has a high degree of electro positivity; it
will actually generate a dipole moment in this crystal. And thus a little mechanical
deformation the crystal is charged. And if you imagine that in any bulk material, there
are hundreds or millions of crystals, each one of them will get charged because of this
deformation.
Although the dipoles could be random and you need to electrode it to make it especially
moving towards a particular direction, but those are the things I will talk about at a later
stage now; this is about the barium titanate structure. The similar piezoelectric effect you
will be able to see in lead zirconate titanate, PZT in the lithium niobium family, in the
lead niobium family, in yttrium manganese family, and in the ammonium cadmium
family.
The thing that would differ maybe the Curie temperature tells us that, beyond that
temperature, the piezoelectricity would not be there. So, that is what is the Curie
temperature, which you have to keep in our mind that you will not see the muscle the
artificial muscle that we are trying to build in action if the temperature goes a particular
beyond a particular level.
38
Now, what is the constitutive equation of such piezoelectric structure as I told you that
for the direct effect where you are applying mechanical stress, X.
So, you are applying stress in the crystal means increasing X. And as you are increasing
X, you will see that the charge is appearing in the system; if you apply or increase the
electric field, then also there will be a charge appearing, but that is pretty common is not
it for a capacitor type of material you know things like a dielectric material.
So, that is not what the smartness is where you are applying the mechanical stress and
getting a charge out of it; only a very few groups of materials will show these. And the
degree up to which it will show is governed by this d, which is the electro mechanical
coupling coefficient. The higher the d, the more will be the electric charge that will be
generating in the system.
How can we use this that you are applying stress and you are finding charge will they
have great use in terms of the development of sensors. So, the direct effect is generally
used for sensor development. Now for the converse effect where you are applying the
electric field now, not the mechanical field creating x that is the mechanical strength is
quite natural for any material.
But you are applying an electric field and finding the mechanical strain that is something
new is not it and that is what is the basic you know equation that we will be using for
39
developing these artificial muscles and by applying electric field mechanical
deformations and that is what is the converse effect or the reverse effect the electrical
stimulus creating the strain.
Once again, here, you can see the same electromechanical coupling coefficient. That
means the more the materials d will be; the more will be getting the strain out of that
same electric field. So, these are the two elementary equations that will continuously
guide us in terms of the development of actuators and sensors by using these two effects.
I will now show you some applications of piezoelectricity. The first application as an
actuator is pretty standard in terms of every day you might have seen it but might not
have noticed that is in printers.
40
One smart one nonsmart like these two things together, one can make a bending.
41
42
43
44
So, this is one application another application is in terms of amplification of the piezo
actuator.
So, if we look at the piezo actuator here, the central part is where the actuation is taking
place.
45
And this actuation in the central part is very small, but because we have used a unique
acolyte structure, a little bit of deformation here can create a large deformation. And all
the semi-minor axis of the ellipse, which is what is these amplified piezo actuators do.
And look at the very last one; this is where the piezoelectric material is not offset with a
composite. Here also because of this offset by actuating these, you can get a large
deformation out of the system. All these three things you can use somewhere or the other
in terms of developing the actuators.
So, in this talk, I have told you about one basic block of a smart material, which is the
piezoelectric material, and how this piezoelectric material can be used in terms of
development in an actuator. And this actuator entire can be used to replace those servo
motors that I have showed you in that ASIMO and to make a robot more organismoid or
organic like a robot there are many other smart materials. And in the next few lectures, I
will quickly cover some very prominent smart materials to you.
Thank you.
46
Lecture – 03
In the last class, we have talked about one of the basic building blocks for organismoid
robots, which is piezoelectric smart material. We have discussed how this piezoelectric
smart material, with the help of the direct and the reverse effects, can build up different
types of actuators and sensors, which can be very effective in terms of developing some
of the parts of these robots.
Now, there are certain disadvantages of these piezoelectric materials; one, of course, is
that it needs slightly high voltage and secondly, it is basically a kind of a capacitor in
nature. So, just like a capacitor gets discharged with respect to time; similarly,
piezoelectric material loses its efficiency because it gets discharged, and hence, it needs
to be reported after particular durations of time.
So, it can be used for a very sophisticated type of action, let us say for the finger
manipulations or for certain motions or rotations, but for more robust and rugged action,
we need more powerful actuators. Now, one of the most powerful actuators is the
magnetostrictive actuator. Also, we will talk about some of the artificial muscles that we
can develop with the help of another smart material called shape memory alloy. So, let us
have a look into both of these different these two different types of smart materials; the
magnetostrictive materials and the shape memory alloy.
47
We will first talk about the magnetostrictive materials, and the name itself is somewhat
suggesting to you that these are materials which have a magnetic connection. Yes,
indeed, that these are materials that respond to the change in the magnetic field.
So, whenever at the advantage of the magnetic field is that you can essentially generate
the changes without physically in touch with the object, which means you can develop
the force at a distance because the magnetic field can get transferred without physical
contact. So, that is the attractive part of it, and the second part the strictive part to
magnetostrictive materials; it essentially refers to that the magnetostrictive material can
respond with the change in the magnetic field, and that is what is the strictive refers to.
Now, if we look into the availability of magnetostrictive materials, we will see that one
of the most popular magnetostrictive material is Terfenol - D, which is made of rare
earth material called Terbium and Iron that is the Fe and it is first developed in Naval
Ordinance Laboratory.
So, that is why their name goes, and of course, there is a D that stands for Dysprosium.
So, essentially, it is a kind of an alloy of terbium iron and Dysprosium. Now, what it
imparts this alloy is a very high energy density.
48
If you can look at this video of a Terfenol - D material sample here and you can see that
if it is hit, you know just impact was so much of energy gets dissipated from the system.
So, it has a very high energy density. Now, naturally, these actually tells us that it will be
very good in terms of an actuator.
49
Now, how does this magnetostriction happens in these kinds of materials? Well, you can
think of that as if each of these magnetostrictive elements will be having millions of this
type of small magnetic domains just like beads, with its own, you know dipole direction,
with its own magnetic field directions which can be random, to begin with. So, that is
what is the general state of magnetostrictive material when the magnetic field applied is
0.
Now, the moment from a distance, I apply a magnetic field. Each one of these magnetic
domains they starts to align themselves with the magnetic field, and you can see. As a
result, the entire group of materials they expand in one direction, creating one
magnetostriction in the system.
So thus, by applying a magnetic field, it can actually expand, but by playing certain
tricks, you can also make it contract, and thus, you can make a forward and backward
movement with the help of the magnetic field by using this type of a system.
50
Now, there are a whole lot of materials ferromagnetic materials, mostly which show this
behavior, and its initially it was observed, of course, in iron and nickel by Joule long
before in 1880.
But today, the material, as I already told you, we use one of them is terbium dysprosium
and iron and also, there is another material in which instead of terbium, we use gallium
actually. So, these materials have large magnetostriction; that means, you know they
show large deformation in comparison to let us say iron or nickel. So, something like
2400 parts per million that is the level of strain that you will see.
Now, this is, of course, a dimensionless quantity, and the other important thing is that it
will show these two up to a very high temperature, which is good means they can really
work in a very rugged environment. Now, how much is this 2400 or 2500 parts per
million, it is even though you know it looks like a large number; but in terms of actual
change of length, it is not that large.
51
In fact, if you look at this curve that we have here the length of Terfenol - D rod, let us
say we consider a 100-millimeter length rod, then with about 2500 parts per million, the
change will happen is only 0.24 millimeter. So, which is even smaller than 1 millimeter.
It may not be visible by the naked eye. But it actually has a lot of power, and you can
also amplify just like for piezoelectric material and then, you know you can make good
use of this kind of system.
52
So, the other important thing for the Terfenol - D material is, as I told you, that if I
change the direction of the magnetic field, no matter positive or negative, you will
always get expansion out of it. It always increases.
But there are two interesting things to note here that if you can somehow keep us keep it
stress, to begin with, you apply some initial stress, you can get actually larger strain out
of the system and the second thing is that this is how you can a get a bi-directional
movement. Let us say if you keep it already under a biased magnetic field, let us say
somewhere here; then, if you increase the magnetic field, it is going to expand; if you
decrease, it is going to come down.
So, which means you can actually get a bi-directional movement out of the whole thing.
The only thing that you are sacrificing is that the range of strain in the earlier case, if it is
only unidirectional, is quite large; whereas, here, it will be with respect to the bias, and
here is that range will be smaller. It will be about half of what you can get otherwise. So,
this is what is the famous butterfly curve that one has to keep in mind for designing
systems using Terfenol - D.
Now, the attraction of such Terfenol - D material is manifold. For example, you can use
it for large force generation up to 50 MPa stress levels you can create. You can create
large deflection up to 200 millimeters, and you can get a large energy conversion
53
efficiency and most importantly, it does not decay with the time that is a very good
advantage of the magnetostrictive transducer.
And it is cost-effective in terms of a low-frequency band, it is not as high as a frequency

band like a piezoelectric material and firstly, it was used in deep-sea underwater
applications like this TALON you can see Tactical Acoustic Littoral Ocean Network, the
entire sonar system of TALON is made out of magnetostrictive Terfenol - D materials
which is sourced by Etrema products.
Now, I told you about direct and reverse effects if you remember. The direct effect is
useful as a sensor. When you are applying force, you can get the voltage or in this case, a
change of magnetization, in a magnetostrictive material that is the direct effect and the
reverse effect is used for the development of actuators, wherein this case as you will be
changing the magnetic field, we will be able to see the change in terms of the length of
the system.
Now, for magnetostrictive material, there are many different effects you need not only
always see the change in length. You can actually generate the change in terms of
twisting in the system or a change in terms of the bulk volume in the system. And
similarly, the same material can respond to if there is a twisting happening or if there is a
bulk change of volume that is happening, it can respond to that through the change
concerning the change in magnetization. So, there are various effects of the
magnetostrictive material.
54
Some of the magnetostrictive effects which are used for actuation and sensing are for the
reverse effect joule effect is, of course there, where there is a change in dimension. But
also sometimes, we use the Wiedemann effect, where a helical magnetic field will create
a torque so that if you have a torque like you know motion generation system required,
you can use the Wiedemann effect.
And also, in some other cases, you can use the Magneto volume effect so that the
magnetic field can create a change in volume. And from the sensor perspective, there is
the Villari effect, which responds to the change of stress and in the axial direction, and
that is reflected in terms of change in magnetostriction.
But there is also the Matteucci effect, where if there is a torque, it senses, and it changes
in terms of a kind of a magnetic field induced EMF, and there is something which is
known as Nagaoka Honda effect, where if there is a change in volume, it actually senses
heat in terms of a change in magnetic state. So, there are three varieties of reverse effects
and three varieties of direct effect which you can use for building up actuators and
sensors using magnetostrictive material.
55
Now, the constitutive equation of such materials is actually very similar to the
piezoelectric material. As you can see that for the actuators point of view, we generally
use the Joule effect, and in the Joule effect, if you apply the magnetic field through H,
you will be able to see the strain, and that is the deformation, and that is the deformation
that we can use in terms of either by restricting it.
So, that you get a force or by a change of length of the system. Remember if there is a
change of stress, then also deformation occurs, but that is not like a smart material like
behavior because that happens for any system.
For the magnetostrictive material, this actually happens because of once again the
coupling called magneto-mechanical coupling here like piezoelectric coupling. So, here
this d creates a coupling between H and ε. So, that if there is a change in H, you get a
change in ε.
Now, that is as far as the actuator building is concerned. Similarly, for the Villary effect,
if there is a change in stress sigma, then there will be a change in magnetization, which
can be sensed in terms of a change of voltage in a pickup coil and once again, the
magneto-mechanical constant d plays an important role here because as the d increases
for a material with a smaller change of stress, you will be able to get a large change of
magnetization.
56
Of course, the magnetization can be changed by the magnetic field also, but that is
something that happens for the regular solenoid, and it is governed by permeability, and
that is not what we are focusing on a smart material. It is this d sigma part which we will
be capitalizing on for the development of the sensors for the magnetostrictive material.
Now, there are many applications of it.
For example, if you want to control the noise and vibration, let us say during the motion
of a robot, this is used. Machine tools, if you have control of machine tools. Controlling
of servo valves just like I told you that. So, many servo valves are required, for example,
the different types of pneumatic actuation.
So, you can actually use these magnetostrictive materials instead of you know
compression based pneumatic materials; it will be much less bulky. So, for servo valves,
you can use it. You can use it for hybrid motors, for Sonar devices, for brake systems,
micro-positioner, and different types of cleaning and welding type of applications.
57
Now, one of the very common magnetostrictive actuators is called mini magnetostrictive
actuator. So, if you look at this mini magnetostrictive actuator, in this case, you can see
that there is a Terfenol - D rod which is sitting inside as you can see it here and then,
there are this magnetization coils and there is a set of a permanent magnet.
So, what will happen is that the moment I will apply, and because of this, permanent
magnets are already biased. Now, the moment I will apply an external magnetic field,
then depending on the direction of it, this Terfenol - D rod will expand or contract and as
a result, this piston will move forward and backward, and that can be used in terms of
motion generation in a robot for various types of applications.
58
Now, here I will show you a fascinating application of magnetostrictive material, and
this is in terms of its motion as an earthworm in a conduit. So, you can see this is also
known as inchworm motion and there are you know inside, these there are these
magnetostrictive materials, and there is a stick-slip happening.
And because of the expansion and contractions and you can see that you know that is
happening and that is creating, these stick slips and the whole thing is actually
propagating.
59
So, you can make it wired or wireless, but this has excellent applications in terms of the
development of pipe health monitoring robots, and you see this is an application that is
entirely wireless, and you see the whole thing is moving vertically.
So, if you want to let us say create an animalistic motion, let us say a kind of an ant-like
motor which are actually miniature, as you can see here, particularly in miniaturized
circuitry.
60
So, an ant-like motor and inside a conduit, this particular type of robot is in
magnetostrictive actuator can be very good in terms of generation of such motions as you
can see that how small you can make the whole thing.
61
And here, you can see that how nicely it is going inside a pipe, and if you have sensors, it
can actually monitor, in that case, the nature of changes that are happening in the pipe,
and it can do that both upward and downward.
62
So, you know you can have a versatile movement that you can generate out of this
system. So, this is you know, magnetostrictive material. I told you that we would be
discussing two materials today; one is the magnetostrictive material, and for which I
have shown you that their most generous application for our purpose will be in terms of
generating locomotions and that you can do for different, very complex environments
you can do it. Of course, you can use such materials for sensors as well; but mostly, it is
used as actuators.
Now, other than these two, there is also a third way of creating motions, which we will
discuss today, and that is with the use of certain artificial muscles. Now, one of the
artificial muscle categories is actually metallic alloys, and it is known as popularly shape
memory alloys. So, let us have a look into how these shape memory alloys behave, how
we can actually use them in terms of actuations, and in some cases, as sensors as well.
63
Let us look into the shape memory material. So, we look into these smart materials based
on shape memory alloys and electroactive polymer, and today, we will specifically focus
on shape memory alloys.
Now, what is a shape memory effect? It is some memory right, and it is a memory, as the
name suggests, is related to shape. So, precisely, this material has a memory in its
parent's stage, which is called the austenite stage, which happens at a high-temperature
phase.
64
So, generally, what we do is that we cool this material. So, it comes to a low-temperature
phase on a martensite phase, then you carry out all the deformations, and then, whenever
you are heating it, it will go back to its memorized shape at the austenite phase.
So, that memory is called a shape memory effect. Now, in some cases, it can memorize
both the austenite and martensite stage, if you properly train it. So, this case is called a
one-way shape memory effect, whereas if one can adequately train it so that both the
phases, it can memorize its shape. It is called a two-way shape memory effect.
So, you can develop both the type of such shape memory effects. Now, the materials
which show such shape memory effect it was first observed that in 1932 in Gold
Cadmium Alloy.
Later on, it was found that there are other alloys like ternary alloys of copper, zinc and
aluminum or copper, aluminum and nickel, or the best one is called nickel-titanium. In
fact, again, Naval Ordinance Laboratory discovered it. So, their name goes with the
whole thing, and it is popularly known as NiTiNOL today. So, it is a particular ratio of
about 45 to 55 percent of nickel and titanium, which actually makes this NiTiNOL
material, and it shows a substantial shape memory effect.
65
Now, as I told you that the shape memory material actually has two different types of
structures. In the austenite phase, it has a body-centered cubic structure. You can see the
atoms at the corners and at the center of the body; whereas at the low-temperature phase,
it has a face-centered cubic structure, you can see that instead of just at the center of the
body, there are these atoms which nicely comes into the center of the faces.
Now, something, if it is inside the center of the body actually inherently, makes it harder
and if it is an FCC, it inherently the face-centered cube is actually a more flexible
structure. The other thing to note is that at the BCC in the austenite phase, these crystal
structures of SMA is nearly cubic in nature.
So, all the lattices are equally dimensioned, whereas, in this case, so in the martensite
phase, it becomes a tetragonal structure. So, where you can see that the sizes are not the
same, you know it would be different essentially.
66
Essentially, these directions it would be different in comparison to the other two

directions. So, this is what is the crystal structure of SMA and why we go for an SMA is
that you get considerable actuation stress, for example, something like 0.1 mega Newton
meter per kg in comparison to PZT, it gets about you can see here that 100 to 1000 times,
in fact, more you can get you to know I mean some in between you can get for the
SMAs. And also, in comparison to human muscle also you can get a very high
amplification, not as high as PZT, but still, it is relatively high.
So, once again, you know you are you know coming after 0.5 here. So, it is slightly less
than 100 times that you are getting through this; the strain is also very high. So,
essentially you are getting immense stress and strain out of the system, and that is what is
very lucrative, and you can get these materials both in wire form as well as in a
springform. Spring forms give better stress and deformation, but the only catch is that the
spring form is actually more complex to analyze.
67
So, modeling is a little bit difficult in the springform. Now, if you look at this diagram
very carefully, it actually describes the entire process. As you can see here that in these
cases, you know we have the stress, three-axis strain, and temperature. So, let us say the
material is first starting from this level when everything is 0.
Now, if I apply the stress and to begin with, let us say that it is at the austenite phase. If I
apply the stress, it will deform until beyond a point, it will behave almost like a plastic
deformation and then, if I reduce the stress, this plastic deformation will get you to know
reversed.
So, this can happen inside the austenite phase; but if you cool this material, let us say you
are in the martensite phase, then if I increase the stress, it will then go to the plastic
deformation, it will go to detwinned martensite. I will once again reduce the stress, now
it will not go back to 0 strain, it will be somewhere here and then, if you heat it up, only
then it will come back to its parent phase.
So, that is what is the shape memory effect and also, the hysteresis in this effect the
stress-based effect is also called superelasticity. So, these are the two effects that are
generally observed in the shape memory alloy.
68
Applications, as I told you, is phenomenal; it is increasing every day; mechanical,

robotics, biomedical, civil structures, and composite. Specifically, for robotics, it is used
in mini actuators sensors, biomimetic robotic hands, and crawler robots. So, there are
very many applications of the shape memory alloy today.
Now, here there are some examples as I am showing you about the SMA type of
artificial muscles which is joined using the springs, you can see it in the two views. Now,
this can be used for developing the fingers.
69
So, there were ASIMO fingers which are controlled by Servo Motor can be, in fact,
better controlled with less power with the help of SMAs, if you can model it properly
and you can actually determine the forces there so that it does not get crushed.
So, depending on the type of element you are holding you can control the force. So, that
is the beauty of the SMA actuated prosthetic hand. SMA is also used in terms of sensors.
So, there are not many applications, but we have devised some applications for SMA as
sensors as well.
You can see here is an SMA based sensor, where there is a rotating drum which is
having the contact endpoints, and these two points are connected with the help of
styluses and SMA wires operate these styluses.
Now, the moment the two styluses are on conductive tape, there is a current that will
pass through, and as soon as there is a current that will pass through those, there were
heating effects and the switches A and B will operate. So, that you will get a signal out
of them, and once one of the stylus you will miss during the rotation, it will not be able
to operate, and hence, you will see that the signal will come down to 0.
So, by measuring these two counter outputs, you will be able to find out how much of
rotation is happening on this particular drum. Thus, you can actually make a superior
70
sensor a very robust sensor out of it. The details of it are again available in one of our
papers in sensors and transducers.
Now you can also use these SMAs in terms of muscles; let us say if there are two
compliant links. So, there are two links, and I have attached them with two SMA wires,
and you can see that by doing that, you can very nicely manipulate the tip of this whole
system.
So, this is a fabricated two-link mechanism, where SMA wire is controlling essentially
two SMA wires, are controlling the two link deformation and by that, it is controlling
these point of interest, the topology of this point so that it can be used in terms of various
types of manipulations. For example, now, this is once again is reported in one of our
papers in the Journal of mechanical design.
71
Now, you can generate many different types of shapes. So, here, for example, you can
see that there is a figure of 8 that is generated by using the shape memory alloy-based
two-link mechanism.
In fact, one can actually find out that what kind of current through simulation will be
able to generate it, and you can feed that same current a loop into the system like this
type of a system, you feedback the currents in the two SMA wires related to the two
links; link one and link two and you will be able to get the desired deflection in the
system. So, this is you know this is one way of using SMA for developing nice
manipulators, which can be used for finger movements things like that.
72
Now, it is used in general for control of aerodynamic surfaces like for control of motion,
for vibration isolation, for grasping of robotic fingers, space exploration, and for the
deployment of solar array hinges. Indeed many applications, it is increasing day by day.
What is the advantage? I am just giving here a quick comparison PZT, which was a
piezoelectric material. Its strain level is 1000. PVDF, which is the plastic piezoelectric
material, let us call it; its strain level is 700. Terfenol – D has a strain level of 2000,
which means it gives more strength than these two.
73
NiTiNOL will give you about 20000; it gives much more than all the three. The only
thing is that in terms of elastic modulus, this one is a little bit softer; whereas this one
Terfenol - D and PZT both of them are equally high and then, from the bandwidth point
of view, Terfenol - D is the SMA is the smallest, then Terfenol - D and followed by the
piezoelectric material.
So, this is just the conglomeration of the properties so that we can make a quick decision
in terms of the use of the smart material. Now, this is where we will put an end to this
lecture. In the next lecture, we will look into another kind of a smart muscle which is
made of shape memory or which is made of electroactive polymers, shape memory alloy
we have discussed today.
Thank you.
74
Lecture – 04
Welcome to the course on foundations of cognitive robotics. In the last lecture, we have
talked about one of the very important smart materials, which is needed, which is
essential for organismoid robot development and that is Shape Memory Alloy.
Today, I will talk about another interesting smart material which is based on polymers
and it is predicted that this is the smart material that will be inevitable in terms of the
development of cognitive robots. So, let us have a very good discussion on this type of
polymeric smart materials today.
Today, we will talk about the Electro Active Polymers, which is also abbreviated as
EAP. Now, this electroactive polymers, they belong to these smart materials group,
which is commonly known as active smart polymers.
75
So, what are these active smart polymers? Well, the polymers that respond to external
stimuli by changing the shape or size actually known as active smart polymer. So, the
output has to be a change of shape or size and the external stimuli could be of different
types. In fact, the best way you can differentiate between the external stimuli is that
whether it is electrical in nature or non-electrical in nature with respect to these, there are
two categories of these active smart polymers.
76
The first group to be active polymers that respond to non-electric input stimuli, such as
pH magnetic field and light one of the example is polyanionic cellulose. Electric field-
based actuation of polymer is an even more important step in terms of polymeric
development, because you can easily apply electric field on a polymer instead of using
other stimuli like; let us say, light or some chemicals and hence in a quick signal based
on electricity actuated polymer is a very special class of polymer. This special group of
polymer, we will call them as Electro Active Polymers that respond to the change of
electrical input it is also known as EAP
Now, one of the examples of an active smart polymer which is not EAP, but simply
active and that is a polymer here as you can see in this particular case and this is a based
on Azobenzene actually, and there is a ring-like structure you can see and this response
to you know light specifically, light or visible light or UV light and because of the
presence of this Azobenzene group in this case which actually contains N=N double
bonds. Under visible light, these double bonds have a cis configuration in which the
polymer is bent. So, this is the cis configuration you can see these active groups
Azobenzene groups here, which are bent.
Now, this is under visible light. The moment we apply UV on it, the part becomes flat.
Now, once again, so that means the polymer will be flattening under the UV light. Once
77
again, you put it under visible light, it will start to bend, and again, if you apply the UV
light in the next joint, it will get flattened and again.
So, this is under the visible light and then the UV light, it will once again, it will get
actually the flattened and again the visible light the bending will start. So, that is how
this whole ring can actually make its movement from one side to the other by
successively making its joints transforming between cis and trans configuration. So, this
is an interesting case of an active smart polymer.
Now, in terms of the electroactive polymers, if we compare the electroactive polymers

with the other polymers like shape memory alloys and electroactive ceramics, you would
see that electroactive polymers keep the maximum amount of strain that is why we have
the highest actually potential of application related to these. It gives about 10 percent of
strain as you can see here.
On the other hand, shape memory alloys can give about 8 percent strain and it has a short
fatigue life also. So, which means you cannot really use it for a very long time. Active
electroceramics that we have earlier discussed gives a much smaller actual percentage of
strain even though it does not have, and it does not suffer from a shorter fatigue life.
Now, the only catch is that the force is less here; for example, in terms of stress
maximum stress it can generate is about 3 MPa compare that with the shape memory
78
alloys, which has about 700 MPa and you can easily understand the changes in
electroactive polymers is somewhere in between it is having about 30 to 40 MPa.
If you compare the reaction speed, the reaction speed is also equally fast like the electro
active ceramics that is between the micro second to second. On the other hand, the shape
memory alloys are slower.
They have second to minute reaction speed. Density wise they are very low density from;
that is why there is a very low weight penalty by using this type of actuators they are
almost close to water density of water in comparison to that SMAs are heavier
electroactive ceramics are also heavier.
The driving voltage is not very high it is about two to seven volt whereas, for SMA also
it is not very high volt all these not mention here and electroactive ceramics are you
require actually quite a very high large voltage.
Now, in terms of the consumed power, you can see that these ones are having about
milliwatts of power, whereas all others are required watts of power, which means this
leads very low amount of power.
Fracture toughness wise also they are resilient and elastic. On the other hand, ceramics
are very-very fragile. So, an overall comparison tells us that these kinds of electroactive
polymers are far better placed in comparison to the SMAs or in comparison to the
electroceramics as far as we are considering their application in terms of robotics
manipulation devices, etcetera. They also have a very interesting history of development.
79
Let us try to see how these kinds of materials have evolved. We will talk about the
evolution of electroactive polymer. Now, one of the very important name in this
evaluation is Rontgen. You know about Rontgen from his discovery of X-ray for which
he got this Nobel prize. Now, in 1880 Rontgen carried out a unique experiment in which
he has applied electric field through the thickness of a rubber band having one end fixed
like this end is fixed and a mass attached to its other end, the free end and what he found
is that as he is applying the electric field, the rubber band is thinning in this section.
You can see that with the help of needle-like probes, he has applied the electric field, and
it is thinning in this region and as a result of it, because the rubber is actually
incompressible in nature. So, the total volume has to remain the same. So, as it is
thinning here, it actually expands in the longitudinal direction.
So, it is thinning and it is expanding, thickness direction its thinning, and it is expanding
in the longitudinal direction in order to keep the volume constant. So, that was a very
interesting phenomenon that was observed and you know re-discussed by Keplinger in
this very interesting PNAS paper in 2010.
Now, that 1880 experiment possibly is the fast on any electroactive polymer. It is also
unique ah in terms of the experiment because, in this case, the rubber was you might say
like a naked rubber. There is no electrode that was actually there. So, simply directly, the
charge is actually provided to the system.
80
Now, in such systems, there is one danger, however, and the danger is also known as
pull-in instability. So, what is this pull-in instability? If I consider an elastomer actuator
which is subjected to high voltage, the elastomer will thin down; however, the same
voltage induces an increasing electric field in the elastomer and so an increasing
attractive force between the oppositely charged electrodes and this oppositely charged
electrodes they are basically bringing the two sides closer.
So, that is why it is called the pull-in voltage, and at the pull-in voltage, there is positive
feedback that will happen and the elastomer will thin down drastically. Finally, it will
result in an electrical breakdown of the system.
81
Let us look into some of the other developments. So, as has been shown here by
Keplinger in his unique PNAS paper that you can actually develop a 3-dimensional
structure based on an electroactive polymer and in these, you do you do not apply any
electrode essentially, because the electrodes will create the pull-in instability on the other
hand, you apply a high voltage through these needles much the same way as originally
Rontgen had planned.
So, you apply the high voltage to the needle and as you can see that as you are applying
it, there is expansion happening in the system and you can see that as you are increasing
the voltage, you can almost get from an angle to a very flat structure. So, this is
something that is kind of a rediscovery of the Rontgen's experiment on an electroactive
polymer.
There are many other electroactive polymers which was discovered after Rontgen's
discovery. An interestingly, each one of them has a kind of a 50 year approximately time
of a frequency between successive discoveries.
82
Let us look into those other discoveries of electroactive polymers. In the evolution of
EAP, Eguchi's discovery plays a very interesting role. In 1995, Eguchi prepared a new
material which is produced from natural resin; it is also known as a Brazil resin from
carnauba wax. So, it is a Brazil wax that is how it is generally known as and beeswax.
So, he has mixed these two and was solidified by cooling while subjected to the DC bias
field and what he noticed is that if this kind of solid electrets. It will generate a voltage
when a stress is applied, and it will have the reverse behavior of being deformed.
So, that is what was the discovery of the Eguchi of a new type of you know kind of a
polymeric system which is a mix of carnauba and the beeswax which actually shows a
high degree of deformation and also, the other effect that is if you know how to apply
stress you are going to generate a voltage. You remember we call this to be a direct
effect, and we call this one to be a reverse effect. So, both of them Eguchi was able to
discover in this particular material.
83
Now, followed by these there are significant discoveries by Kawai in 1969, and this is an
entirely new inorganic you may say not exactly you may say that synthetic piezoelectric
material. So, this is a synthetic one, because the last one that we had discussed was
derived from natural organism material. So, this is a synthetic organic material which is
known as polyvinylidene fluoride. In fact, because of the presence of fluorine, some it
has some inorganic contributions, but it is a synthetic essentially polymer cis
polyvinylidene fluoride and it has shown a large degree of piezoelectric activity.
Now, at a later stage, Fukada actually developed piezoelectric biopolymers on the basis
of this PVDF. Another important landmark was from the Katchalsky and his group in
Israel, who have developed actually responsive gels, but these gels were mostly chemo
mechanically activated. So, these are gel polymers and this shows a very high degree of
shrinkage or swelling; that means there is a bulk volume change in the presence of an
acidic or alkaline environment.
Now, joining these two concepts, in 1980, Hokkaido University, people have first
developed responsive gels that are based on electrochemical activation and following
that line subsequently, Osada and Kishi developed in 1989.
They have developed polymers that can demonstrate very large strain under relatively
low activation voltage, which was followed by Bar-Cohen in 2010, in terms of the
development of electro active polymers on a specific polymeric system called Nafion.
84
Now, the question is; where am I going to use this kind of an electroactive polymer?
Why are they where I am repetitively telling that it is so important in terms of robotic
development. Well, one of the very important applications of this type of polymer is in
terms of locomotive, locomotions or generation of locomotive forces. So, let us look into
that how in robotics you know gradually, this transfer from the motor to motorless
locomotion has taken place.
(Refer Slide Time: 16:49).
If we look into the animal locomotion based inspirations, you will see that starting from
the motor based rigid robots like this kind of a Quadruped Stair climber Titan-6 or a
snake-like ACM R5 or for that matter a Bi-pedal Walking Robot M2.
We have also seen Asimos, which are all basically motor-driven some of them are servo
motors or other Russell's DC motors, but these are all motor driven systems, but taking
inspiration from animal locomotions. Gradually, these locomotions have been achieved
in a different manner.
85
This different manner is if you study the animal locomotion, you will see that they
essentially use muscles for actual motion and we will call it actuation. Now, these
muscles are attached to bones by the tendons, and the muscles contract to move our
bones by pulling on them.
So, you have to keep in mind that these muscles can only contract and hints; they can
only pull; they cannot push. So, that is one of the important aspects to be kept in mind.
So, this is why in any joint, the muscles are used in pairs; one muscle of the pair
contracts to move the body part and the other muscle at that time actually relaxes and
later on the other muscle actually contracts to return the body part back to the original
position. So, the muscles that work like these are called antagonistic pairs. So, we have
basically two types of muscles called agonist muscle and antagonist's muscle.
If you look at your own elbow for example, what you will see is that in our own elbow, if
we actually move our arm, if let us say that we want to pull our arm, you would see that
this is the part which is pulling. So, this is the part which is actually, agonist you know
the muscle and the back part is actually the antagonist.
So, that is how the motion of this a joint is actually happening by constantly the pulling
of the two biceps and the triceps muscle. So, it is a very similar kind of a concept that we
will be actually applying in the case of the smart muscles also.
86
Let us see how we can apply this antagonistic actuation by using shape memory alloy.
Now, here is an example in which people have developed a shape memory alloy-based
agonist and antagonist motions.
87
So, the whichever is pulling is an agonist and the other one is relaxing. So, antagonist
and look at the trajectory that it is creating. So, the other side you have seen that agonist
and antagonist motion and you can see here that it is a nice rectangle that has been
drawn. Very similar things are done for almost all our muscles and by varying these pairs
motion, you can actually do it faster or slower.
There is another very interesting application of this agonist-antagonist motion. By using

actually electroactive polymers by a group in Switzerland EMPA group and they call this
88
to be BLIMP. So, this BLIMP is essentially a Pressurized Helium filled, it is a big fish
that is like eight meters long and it is actuated by electroactive polymer using this
agonist-antagonist configuration and it is designed to work at various frequencies voltage
and phase shift between the body and the tail actuation.
So, you can see that this is an 8 meter big and you can see these muscles, you know
which is it is a very flexible inflated structure.
89
So, first they are pressurizing helium gas into it and you can see that how these muscles
are working here and you can see that by these you know agonist-antagonist working of
the BLIMPs EAP electroactive polymers, how they are very nicely you know moving the
tail as well as the body in the space and is generating the motion.
If this is in the laboratory, you can see how this motion is happening, also they have
tested such a system in outdoor. So, you can say that this is possibly the fast organismoid
90
robot, which is a fish-like robot which is based on the muscles that are essentially, bio-
inspired in nature. You can see the electroactive polymers here.
So, this is where they are actually testing this entire system outside, and you can see how
nicely this is manual varying controlling its position. So, that is one of the examples of
the electroactive polymers.
Now, we will talk about some of the classifications of the electroactive polymers that
they themselves can be grouped into two types; Electronic EAP, which is driven by
91
Electric field or Coulomb force, and Ionic EAP, which changes the shape or by mobility
or diffusion of actually ions. So, as the name itself is suggesting and it is also any true
that electronic EAPs are much faster, the Coulomb force-based EAPs whereas ionic
EAPs are actually slower because they depend on the mobility or the diffusion of the
ions, which is a slower process, but of course, they can generate a much larger change in
the system.
Now, electronic EAPs, which is also known as EEAP, they are like dielectric EAP one of
the example, another is electrostrictive paper, ferroelectric polymer, and liquid crystal
elastomer. On the other hand, the IEAPs are like ionic polymer gels, ionic polymer-metal
composite like; IPMCs, Nafion, Duponts, Flemion, etcetera, then there are some
conductive polymers and even carbon nanotubes. So, these are the groups which show
the ionic electroactive polymer
We will now focus on the ionic electroactive polymers because, as I told you that they
are the best performing material as far as the large deformation goes and also as far as
the power consumption goes.
They consume very little amount of power, and so they have a real future in terms of
developing different types of motions in robotic, you know limbs or so-called arms and
legs of a robot. So, let us look into the ionic polymer matrix composites or this kind of
you know ionic electroactive polymers.
92
If I compare between the EEAPS and IEAPs, that is what I was telling you that EAPS
needs a high activation voltage whereas ionic EAPs need very low voltage and they have
this EAPs have a high energy density and rapid response time in comparison to these
IEAPs have relatively slow response times, but the amount of deformation is much more
and they perform better under wet condition.
So, if you look at one of these ionic electroactive polymers, which is IPMC that is ionic
polymer-metal composite, the example here is the Nafion. Now, this type of material
actually consists of a Polymer matrix that is sandwiched between two metallic layers.
Now, Nafion specifically is actually a Perfluorinated co-polymer of
PolyTetraFluoroEthylene or PTFE and a Perfluorinated Vinyl Ether Sulfonate. Now, you
have actually come across PTFEs in your day to day life when we talk about non-stick
pans, etcetera.
So, PTFE actually gives the strength to the material. So, PTFE is one part, as I told you,
there is a co-polymer. So, one co-polymers are made of two polymers. So, one part is
PTFE, which gives them strength and the other part is this perfluorinated Vinyl ether
sulfonate. So, this perfluorinated vinyl ether sulfonate, which is heavily fluorinated
essentially and it has a sulfonate group, that sulfonate group is actually the source of
polarity in the system and note down that these perfluorinated groups have OH- in it. So,
it means that it has a fixed OH- attached to it always ok.
93
So, this is. So, if you have a material which is these a combination of the two and then if
you put them in an electrode, then you can generate a fixed negative charge on one side.
So, you can see here that there is this electrode here, and there are these fixed negative
charges, which is generated with the help of the sulfonate groups. Now, these fixed
negative charges, the moment I attach these cathodes, what will happen is that the water
molecules that is present inside the IPMC they will start to go towards, massively
towards these negative, you know the direction the cathode.
So, as there is the driving of the water molecules, this side will bend; if there are more
water molecules, one side will bend know. So, probably this what you are getting a bent
thing and with respect to time, some water molecule will back diffuse, which actually
talks about a relaxation.
So, a small relaxation will happen, but essentially you can generate a large bending in
this kind of materials, but of course, one of the precondition is that it needs to be wet
because then only the water molecules movement towards these cathodes will create this
large hydrophilic expansion in the system.
94
Now, depending on the type of electrodes, you can actually generate the positive and
negative bending in IPMCs. So, as you can see that this is a kind of a typical IPMC that
is how it looks like under the normal condition and under the bent condition, and you can
get a large deformation out of this system and with a low actuation voltage and fast
response.
You can actually generate by changing two different types of electrodes on an IPMC,
you can generate two different types of curvatures in it, you can generate a downward
95
and you know this downward and an upward kind of curvature. So, you can generate
double curvature in the system by using two different types of electrodes in the system.
So, that is one of the advantages of the system.
Now, this kind of IPMC based actuators they are used for single link manipulators,
multi-link grippers', vibration generation control, and 4-bar manipulator. So, most of the
locomotion generation can be done with the help of this type of IPMC based actuation
system.
96
So, some of the applications you can see that like this is in terms of a dust wiper, you can
see that how this dust wiper actually works in terms of cleaning in fact, the last
reformation and then there are other applications also like these sample handling, in
robotics, this I already told you in different applications are possible or particularly for
the planetary applications people have already used this type of electroactive polymers.
Now, I will just conclude this particular session of the lecture by comparing, because we
now are in a process that we can summarize actually, that the different smart materials
via the biological system in terms of their performance. So, this is what we will be doing
in the final concluding slides.
If I compare mammalian skeleton muscles with initially, the field activated electronic
electroactive polymers, and then we will compare them with the other groups of ionic
electroactive polymers. Now, the mammalian skeletal muscles produce large strains.
How large? We always say large in the quantification here is about 20 percent and the
stress is; however, moderate about 350 kilo Pascal, stiffness can be varied, but the fuel
that is used in terms of the nutrition that is a high energy fuel, efficiency is near about 40
percent, and they have a good work density about 40 kilo Joule per kg. They have a high
cycle life because they can actually regenerate also.
The disadvantage is that; this type of material is not an engineering material, and they
have a narrow temperature range of operation in the high temperature, they cannot work,
97
and they expend energy to maintain force you know without moving. So, that is one of
the problems that they do not have any catch state.
Now; however, an interesting thing to learn is that they have a very interesting 3-
dimensional system that integrates both the sensor, energy delivery, waste heat removal,
local energy supply, and repair mechanisms, all in one pack. That is something that kind
of compactness we are still not able to achieve as far as the synthetic you know materials
are concerned.
Now, let us look into the field activated electronic groups fast among the synthetic
materials. The first group is the field activated electronic group like PVDF, high
mechanical energy density relatively large actuation force, in the mega Pascal range can
operate for a long time in room condition and rapid response about millisecond range
and can hold strain under DC activation. So, they are all electrically activated. The
disadvantage is that the product is mostly the monopolar type of actuation, and it requires
a high voltage. So, high voltage is the real disadvantage of the system.
In comparison to that, if I consider the dielectric elastomers like some of the examples I
told you in terms of Rontgen's initial experiment and the other elastomers, they can
actually show large strains. It can be as large as up to 380 percent, much larger than the
mammalian skeletal muscles. So, that is what is very large stress is moderate only
several MPa, but still, it is comparable with respect to mammalian muscle large work
density and moderate to high bandwidth, not very high bandwidth this has a very high
bandwidth the field activated groups. Low cost, low current, and good electromechanical
coupling.
Now, what is the disadvantage? You need to generate high voltage, locally, of course,
typically requires a DC-DC conversion, and it is quite soft, pre-stretching mechanisms
add substantial mass and volume because pre-stretching is important in terms of its
movement. You remember that in the initial Rontgen's experiment, there was a mass that
was used to prestretch the rubber band. Now, there is potential to lower fields using high
dielectric materials and based on readily available materials. So, this is a good one, but
still, it has certain disadvantages.
If I consider the ferroelectric polymers, then the strain level is less, only about 7 percent,
stress is high 45 MPa blocking stress relatively high, you know much-much higher in
98
comparison to these, which is in kPa range very high work density, stiff and strong
coupling. It also requires high voltage, and the cycle life is one of the problems because
there is some fatigue that is observed, and it has a limited temperature range. So, here the
lower voltage and fields people are trying to achieve now, you know in the ferroelectric
polymers and the energy density is in favored and small devices with high frequency in
something that is under focus.
Let us look into the other group, the ionic polymer group, which we have discussed at
the end. They have bidirectional actuation, and they have bistability, and they require
low voltage. Now, this bistability is a fascinating concept. It means that this kind of
structure can have two different stability configurations, and it can very fast switch
between these two different stability configurations.
The disadvantage is that they always require some electrolyte because there is some
movement of the ions that have to occur, voltage requirement is not very high, but this
electrolyte presence is significant under wet conditions. Generally, that is why it requires
protection from evaporation, low electromechanical coupling, and specifically for IPMC
does not hold strain under DC voltage.
For conductive polymers they can generate high stress, but the strain is moderate-low
voltage, and these are stiffer polymers, but they have low electromechanical coupling,
99
which means not much of electrical energy will be converted to mechanical energy. They
are promising for low voltage applications.
Also, there are CNTs which require which gives a high-stress low voltage, but the strain
is low, and IPMCs which actually takes low voltage, very low voltage, and large
displacement, but it has a low coupling in terms of efficiency, and usually it has no catch
state; that means, consumes energy in holding position and requires encapsulation, but it
has a great potential as bulk material you know properties you know they can be actually
improved due to individual nanotubes. So, this is a comparison between all the different
types of these electroactive polymers versus the muscles.
So, with this lecture, I will conclude the smart materials and their applications in terms of
the development of organismoid robots. In the next series of lectures, we will look into
the cognitive aspects of the system, and particularly we look into the living system and
the cognitive aspect of it, but we will start with the say, for example, the description of
the highest, you know you know in terms of the cognitive capabilities that is the brain
itself. So, we will discuss these parts and because that is also an essential element in our
organismoid robot development.
Thank you.
100
Lecture – 05
Good morning students, welcome to the course on Foundations of Cognitive Robotics.

We have reached week-2 of this course now. Now, before I start the lectures of week-2,
let us quickly summarize what we have learned in week-1.
Summary of our learning of week-1. The first point that we have learned is that cognition
requires embodiment we have told you, I have told you about the difference between
strong AI and weak AI system. And the cognitive robots belong to the weak AI class for
which embodiment is essential.
Now, embodied robots are classified into two categories; organismoid robot and
organismic robot. So, these are the two categories you would see organismoid and
oganismic robot. We will be focusing on organismoid robot. As I told you that
organismic robot is still far ahead in the future. So, we will first, we will start to think of
how to materialize an organismoid robot.
Now, organismoid robots, you know because they try to simulate like a biological
system. In order to do that, they would require sensors and actuators. These sensors and
101
actuators we have seen that in some of the traditional robots they are based on, like the
actuators are based on motors or servo motors, and the sensors can be different types of
off the shelf standalone sensors.
But if embodiment is unfocused, then one of the important building blocks of these
sensors and actuators could be based on smart materials; that is why I told you about four
different types of smart materials which can be used for sensors as well as actuators.
Piezoelectric smart materials, magnetostrictive smart materials, shape memory alloys and
electroactive polymers.
But if you remember that the last two shape memory alloys and electroactive polymers
can be actually developed in terms of developing artificial muscles which can have
agonist-antagonist structure. So, with this building block that we have discussed in the
summary of week-1, now we will go to week-2 of our lecture.
Let us now look into our learning goals of week-2. My plan for week-2 is that now
because the cognition is into the picture, I will first discuss about the architecture of the
brain, and the center for cognition for reptile and human by looking into the specific
areas of brain. And then from brain, I would go to even in a top-down approach to the
architecture of a neuron, and we will also talk about how to model the neural impulse.
102
Now, let us look into the cognitive perspective which we will be having with us, because
when will be talking about the biological systems like the brain and the other nervous
systems, then there can be a biological perspective towards it, but our focus will always
remain towards cognition. So, let us look into that what do we mean by cognition, and
from that point of view how we can look into a biological system.
Our current perception of artificial intelligence actually touches three broad goals; the
first one is the understanding of the biological system. For example, we will be trying to
understand the nervous system today, and the mechanisms that bring about this
intelligent behavior in humans or animals that is our first goal. The second goal is some
of the abstraction of general principles of intelligent behavior.
As we will be talking about the biological systems, we will try to make these
abstractions. And then further as the final goal is application of these principles to the
design of useful artifacts. Now, we would consider all intelligent systems both living and
non-living under one single envelop which is called agent. This is the direction that you
will see in Rolf Pfeifer and Josh Bongard’s book on how the body shapes the way we
think.
103
So, let us think into these agents. What we mean by this agent? So, what is an agent, and
is cognitive robot an agent? Well, an agent is anything that can be viewed as perceiving
its environment through sensors and acting upon that environment through effectors. So,
there are two important points here. One is perceiving the environment through sensors,
and another is acting upon the environment through the effectors. Both of them are
important from an agent’s point of view.
Now, when we talk about cognitive robot, we are talking about machines that have at
least some agent like characteristics, irrespective of whether they do any useful work for
humans or not. This would include humanoids, pet robots, entertainment robots, service
robots, rescue robots, etcetera. So, they should have some agent like characteristics in the
sense that they should have some sensors and they should have some effectors associated
with them.
Now, in this same spirit, we will talk about intelligent agents as you can see. These are
the agents which always comply with the physical and social rules of their environment.
So, it is not only essential to have some sensors and some effectors, but also, they should
comply with the physical and social rules of their environment and exploit those rules to
produce diverse behavior. Now, whenever they are doing these the intelligent agents, this
is generally manifested through cognition. So, this is what is important, that this is
104
manifested through cognition. So, let us then look into that how exactly this
manifestation happens in the system.
Now, whenever these cognition would come into our purview, we would always
remember these very interesting quotation by Hippocrates in the 5th century BCE, when
he said that from the brain and from the brain alone, arise our pleasures, joys, laughter
and jokes, as well as our sorrows, pains, griefs and tears. Through it, in particular, we
think, see, hear, and distinguish the ugly from the beautiful, the bad from the good, the
pleasant from the unpleasant. All the time the brain is quite, a man can think properly.
So, even in the 5th century BCE, people understood that when we talk about the
cognition, there is somewhere in the brain that the center for cognition lies. And they
have understood that when we talk about cognition, we are talking about emotions, it can
be both good emotions as well as bad emotions. It can be related to our thought process;
it can be related to our, you know, capacity of judgment, capacity of identifying what is
beautiful, differentiating bad from the good, and pleasant from the unpleasant. So, that is
essentially what we want to impart to a cognitive robot as well. And in the biological
system, by and large, this is controlled by nothing but the brain.
105
So, the question is how does brain activity relate to cognition? Well, that is a long story.
Let us just begin with the basic architecture of the human sensory and processing system.
We will first look into the human sensory and the processing system; through that you
will be able to appreciate what a complex architecture is required to get the level of
cognition that Homosapiens are having. Let us look into the human nervous system.
The human nervous system is essentially divided into two parts; central nervous system
and peripheral nervous system. Now, central nervous system would consist of the brain,
106
and the spinal cord, whereas the peripheral nervous system would have the cranial nerves
and the spinal nerves. Function wise, the central nervous system is having the cerebral
hemispheres, diencephalon, cerebellum, brainstem. So, this all the three or four
important parts, they are all inside the brain when we actually look into the brain, we
look into all these things.
In addition to that, the central nervous system also consists of the spinal cord. So, they
basically take part in the analysis, and integration of sensory and motor information.
Now, whenever you have some internal or external environmental change, your sensory
receptors at the surface and within the body if it is external, then the outside you know
thinks of the sensory receptors at the surface. And if it is inside the body, then it is a
visceral one. So, those sensory receptors, they actually collect the information of the
environment, and they add it up in the sensory ganglias. And from there, the sensory
components are actually accessed by the central nervous system.
On the other hand, that is the input for the central nervous system, then the central
nervous system what it gives back to the peripheral nervous system is in terms of
different motor components. We can divide it into two parts; one is the visceral motor
system; another is the somatic motor system. The visceral motor system which would
consist of sympathetic, parasympathetic and enteric divisions, they have autonomic
ganglia and the nerves, and they control things like for example, our heart rate, our
digestions and our all other internal organs that is controlled by the visceral motor
system.
On the other hand, the somatic motor system controls the motor nerves, and they control
the effectors like the skeletal muscles, so that is where the somatic motor system would
work. And the visceral motor system would work on cardiac muscles, glands and the
smooth muscles which occurs in the in large intestine regions etcetera. So, that is the
overall you know circuitries architecture of the nervous system in a human body.
107
Now, let us look into some of the sub elements of it. Well, if we look at the central
nervous system, you can see how it looks like, the top you have the brain. And the brain
itself you can see is having different parts what we can see from here is the cerebral
cortex; we can see also the thalamus here. And below the thalamus, we can see the
brainstem; and below that we can see the spinal cord. So, this is this entire thing is the
central nervous system.
How does it work? Well, look at one sensory receptor let us say we are considering one
finger. Now, here are the receptor ends. And you have two different kinds of this red and
blue denotes two different kinds of nerve fibers. One group in the red group is the
mechanosensory afferent fibers. Afferent means something which is conducting the
signal inward towards the spinal cord, and then towards the brain.
Now, mechanosensory afferent fiber is one which senses the changes in the mechanical
forces, stresses, deformations etcetera. And the other one is the pain and temperature
afferent fiber which senses the if there is some cutting bruise so the pain or if there is a
temperature, so this is what carries it and they need the blue ones also goes towards
through both of them goes through the dorsal root of the ganglion cells. And you can see
this dorsal root of the ganglion cells here.
Now, this is the part of the central nervous system. And here the spinal cord itself is
divided into four important parts, cervical, thoracic, lumbar and sacral. As you can see
108
that the cervical region, you can clearly see up to the shoulder level, then the thoracic
part, the body part, then the lumbar part and then, so it is the belly and the bottom part
and then the sacral the bottom most part. So, that is what is the region of the spinal cord.
Now, let us look into the other part that is the peripheral nervous system. As I told you
that it has two parts in it parasympathetic and the sympathetic divisions. Now, if you
look at these very carefully, sympathetic division which is with the red bar is what is
always required whenever it urgency, you need to do something immediately. For
example, you are exclaimed it dilates the pupil.
It will inhibit the salivation; it will constrict the blood vessels; it will relax the airways,
accelerate the heartbeat, and there will be stimulation of the sweet gland, sweat glands,
inhibition of the digestion, stimulation breakdown of glycogen and the other release of
the glucose – you need energy, inhibits activity of things like intestines, relaxes the
urinary bladder, stimulates sometimes orgasms etcetera. So, this is all in terms of the
sympathetic division.
Now, I told you that in the peripheral nervous system that these sympathetic jones, they
actually the nervous systems they have the role of taking care of all the emergencies of
the body. But our body is not always in an emergency mode it is also having the
relaxation modes. So, it is at that mode when we will be having the other group of nerves
109
of the peripheral nervous system that is the parasympathetic nerves, they will be playing
an important role.
Let us look into their role. And in fact, what you will notice is that just like when we
have discussed the muscles that you have agonists and antagonists, similarly there is
always contradictions in the architecture. There is one group which excites it and there is
another group which actually inhibits the process. So, during relaxation, it is the
parasympathetic nerve which will come into picture. Let us look into their actions.
If you look at the parasympathetic division, you would see that here it is constricting the
pupils so much so that in ultimate stage of relaxation you may simply sleep, it will
stimulate salivation. It will constrict the airways you do not need too much of oxygen
anymore; it will slow the heartbeat, because you are relaxing. It will stimulate the
digestion process now. There will be a slight stimulation of glucose uptake and glycogen
synthesis. It will inhibit the gluconeogenesis.
Stimulate the activity of the intestine, stimulate urinary bladder, stimulate peniles of
clitoral arousal. So, these are all under the parasympathetic division. So, the peripheral
nerve system always works between the two. Sometimes these is important when you
have to survive, and sometimes this is important when you have to maintain the
condition, the homeostatic condition.
Now, what you will see is that many of them are having you know the sympathy nerves,
for example, are having either they are directly coming to these you know the roots – the
ganglia roots, otherwise they may have some sort of you know they are ganglias closer to
the organs itself. So, that is also possible in some of these groups.
Now, in the other hand in these case for the parasympathetic ones, you would see that
here some of them are coming to the parasympathetic ganglions. And from there they are
actually going to the directly you know this group is going to the brain stem area.
Whereas, the other group here these last three of them, you would see that they are going
to the sacral area. And from the sacral area, the signal goes to if required it goes to the
brain stem. So, this is about the peripheral nervous system.
Why this cognitive architecture is important for us? Because we have to think of it when
we will be developing the robotic cognitive architecture, some sort of a similarity is
110
needed in terms of developing intelligent agents. So, that is why it is important for us to
study the biological system to make abstraction of these systems, and then try to
implement it.
We will now go into the central nervous system and the core of the central nervous
system that is the brain. Now, we will look into one of the most sophisticated central
nervous system, the core of the central nervous system that has been developed by
nature. We look into the human brain. Now, let us look into a model of the human brain
fast, and then I will describe the system to you one by one, sometimes going deep inside
the system.
Here is with us a model of the human brain. Now, as you can see here that this human
brain is divided into two parts; the left and the right. And there is a fissure between the
two, so that is what is called the longitudinal fissure. So, we have the left part of the
brain, and the right part of the brain and that is from the top when we are looking at. And
this very top part of the brain, the top part is actually is the frontal lobe of the brain; we
will talk about it when you will see the diagram.
Now, if I look at a side view of the brain, you see we are looking we are now taking it
towards the side, now let me just tell you that the eyes will be towards this direction the
frontal side ok, and the ears will be in this direction. Now, as you can see that there is a
111
deep fissure here, this deep fissure is known as the lateral fissure which actually
separates the frontal lobe from the temporal lobe. So, the temporal lobe is here.
We can also see in the top another fissure that is visible ok, and this fissure is known as
the sulky the central sulky, and that actually divides the frontal lobe and the parietal lobe.
So, this is where from the central sulky, the front part is the frontal lobe part and the back
part is the parietal lobe part.
Now, you can see that just like the frontal lobe divided between left and right. Similarly,
if I just turn it you will see that the parietal lobe is also divided between left and right.
And not only that below that we have two different parts also here, these are parts of the
cerebella. So, this is when we look into the brain from the top.
Now, I am going to even open it up from you in order to know that where this brain is
connected with the so called you know the spinal cord. So, let us look into that. Now, if I
actually try to open it up, this will be visible to you.
So, let us take keep these other parts away. Now, we are deep inside the brain and we are
looking at the brain from inside. As you can see that from the spinal cord, this is where
we are actually getting all the nerves inside the brain region that is the brainstem as you
can see. Now, this brainstem, this part is what is called the medulla oblongata, and then
you have the pons area, and then you have the midbrain.
112
And even deep inside you are going to have the other parts which we will discuss later
on the thalamus and hypothalamus and all these regions, but this is where the brain gets
connected with the these you know peripheral nervous system through the spinal cord.
So, this is what is the overall view. So, we have to keep in mind once again the basic
important part is the frontal lobe, the temporal lobe and the parietal lobe, this is the
cortex area of the brain. And also I have shown you the cerebellum.
Now, let us look into the details of this brain structure into the major landmarks of the
brain. If I take a medial view of the brain, then we will see the right hemisphere. So, if I
take the medial view, you will see the right hemisphere part of it ok. And just keep your
mind that this is what is the frontal lobe.
So, this is the frontal part, this is the frontal part ok. I will just write F here, so that you
understand that. This is the frontal part. And this is what is the parietal part, so I will just
write P here, so that you would understand. This is the parietal part. So, eyes are here in
this direction. And this is what is the temporal lobe, so that is the T, I will put here that is
the T part. And deep below the parietal part, we have the occipital part here. So, I will
just write it as O which will say that this is the occipital part here.
And if you remember that this is what I have shown you a few minutes before so that is
the brain stem. So, this is the stem part of it ok. So, I will just try to write here that this is
the stem part of the brain. So, you know you can get the bearing of it ok. And of course,
113
this is the cerebellum part, so that is what is the C I put. So, I will keep it as Ce, so that
you know this is the cerebellum. On the other hand, this top area these frontal and
parietal and all these things together that is the cortex part ok, so that is the cortex. So,
this is the cortex part of the brain. So, this is just to give you an overall bearing of when
you are looking at the brain. And in this case we are looking into the right hemisphere
ok.
Now, if I actually open up from the longitudinal fissure, if I open up the brain, then
inside you will see this region and that is what is known as corpus callosum. The corpus
callosum is actually white, because it is myelinated. Myelinated means here there is a
fatty layer over all the nerves, whereas this area is the gray area the top part is the gray
area.
Now, inside that I told you that in from the brainstem if you go up you can see the
cerebellum here that is what is the cerebellum part that I was telling you. And then if I
you can see also here the pons area and the you can also see the midbrain, and on the top
of it you can see the thalamus part of it. In fact, the thalamus is also having the right
thalamus and the left thalamus; here it is the right hub. So, you are going to see the right
thalamus here ok.
And if you look at the whole brain system again you see the frontal lobe the most
important part is the prefrontal lobe part of the frontal lobe. And there is a pole here this
is also known as the frontal pole. So, this is the frontal pole, I can call it shortly as FP
that means, this is the frontal pole ok. So, this is the frontal pole area. Then you also have
similarly one pole here that is the ultimate culmination point of the temporal pole, so I
can call it as Tp which is below the lateral fissure ok. And similarly you have the
occipital lobe here and this is where you have the occipital pole. So, this is where you
have the occipital pole ok. So, these are three important poles.
Now, another thing I want to show you from this diagram that you can see that some
region seems like they are actually going up bulging up the hill regions ok. So, these are
the regions which are like we will call them to be gyrus. So, this is what will be the gyrus
region, that means, here you are getting the bulging or the hill like situation. So, these are
gyrus region gyrus region ok, ok.
114
Now, you can see the other part that is the valleys you can see and valleys are called the
sulcus. And it is the central sulcus that I have shown you earlier, so that is the central
sulcus region ok. So, central sulcus essentially divide the cortex into two part that is the
frontal cortex the F the frontal cortex part and the parietal cortex part, this is where it is
dividing between the two. So, this is when you are actually getting a medial view of the
brain.
Now, let us look into just very similar to the right hemisphere. If I look into the left
hemisphere, you can again see the central sulcus here. And you can see that the central
sulcus is having two hills in between these central sulcus valley is there. The hill in the
front part is called precentral gyrus, and hill in the backside part is the post central gyrus.
This precentral gyrus and postcentral gyrus, they are very important you know
somatosensory actions; I will talk about it soon
Now, you can see also the other parts in the frontal lobe and this is where is the
prefrontal region of it. So, you can see the frontal lobe, you can see the parietal lobe as I
told you here that is the parietal lobe. Parietal lobe itself is having two parts this is the
parietal the frontal part and that is the somatosensory part, and this is what is the
occipital part of it that is where is the occipital lobe.
And in the side where the ear is there let us say we have the temporal lobe as you can see
the temporal lobe is here. And below the cerebellum is actually visible, and that is where
115
you see transverse cerebral fissure, below which the cerebellum is there. Also you can
see this slightly bulged pons and the medulla oblongata below which the spinal cord is
there.
Now, anything slice here you would see that the top part that is where is a gyrus part, and
below that up to certain level you have the cortex gray matter where you do not have any
myelin cover. And if you go deep inside the sulcus, you will get the white matter like the
corpus callosum I showed you earlier. So, this is what is the structure of the left
hemisphere which you can see that it is very much of a similar structure like the right
hemisphere.
Now, what are the functions of this left and the right hemisphere. Let us look into that.
The major functions of left and right hemisphere is in terms of that the left hemisphere
takes care of the verbal, linguistic description, mathematical and sequential analytical
activities here. The back part there is this stereognosis the depth sensing part, and even in
the back side you have the right visual field here.
So, you can see that whatever you are watching from the eyes they are getting criss-cross
connected. So, the right visual field is coming here in this region the left in the left
hemisphere, whereas the left visual field getting criss-crossed is coming here in this in
the right hemisphere in the right in the you will get the left visual field.
116
Now, in the right hemisphere, there are non-verbal things – musical, geometrical, spatial
comprehension, temporal synthesis, non-speaking hemisphere, these are the right
hemisphere part and behind that you have this left depth left side depth perception. The
depth perception of the left side is coming in the back side of it. Now, one thing is very
important here that these in today’s definition you cannot very distinctly always say that
left does this and right does this because there are lot of coordination between both of
them that happens.
Important thing here is that you can see in the middle the presence of this corpus
callosum. In fact, these two views are very intermittently mixed. So, the left and the right
there you know very intermittently mixed. So, if I actually cut the two parts of the brain,
you would see that there will be kind of a delay of the signals of something like 1
millisecond or so. But they are you know very highly connected that is why I told you
that many works of the left are also influenced by the right and vice versa through the
corpus callosum part of it.
Some of the important points of the cortical control we have to keep in our mind that
there are three important point here. One is that we have some actions which are locally
resolved these are called reflexes like you can see, the reflex here, the reflex on the wrist
area ok, the reflex here, the reflex on the knee is a very popular reflex area the reflex
117
here. So, these are local reflexes for which the central nervous system is not associated
directly this happens.
Now, the in terms of the sensation, all the somatosensory sensation of the left side goes
to the right side of the brain and vice versa that is the right side goes to the left side of the
brain. However, when we talk about coordination then right side has to work with the
left. So, then they actually crossed on both the sides together; otherwise left goes to the
right and right goes to the left that is the usual pattern of the cortical control.
So, the question as I told you that what joins the two hemisphere? So, the hemispheres
are completely separate individualistic divided by the longitudinal fissure that runs
between the two hemisphere from the anterior or the frontal part to the posterior or the
back part of the brain. The link between the hemisphere is provided by the corpus
callosum. So, this we have to keep in our mind this is a very important thing that this
corpus callosum which actually joins the left and the right.
Now, it is like a large arch of white matter. The number of actions that travel between the
two hemisphere is estimated to be more than 100 million. So, it is a huge number of
nerves which criss-crosses to the corpus callosum. It has fibers that project between the
hemispheres in an orderly way with regions in the anterior portion connecting similar
brain areas, the frontal areas, frontal lobes as well as in the posterior portion connecting
118
similar brain areas in the occipital lobe. So, corpus callosum connects both the frontal
parts as well as the occipital parietal parts of the left and right sub parts of the brain.
Here there is an important consideration I would like to tell you that is there that the
function of the brain is localized or is it that every area works on everything. Now, even
though there are associations between different areas, but it was found for 100s of years
through surgeries, neurosurgeons mostly and through other scientific endeavors that
there are indeed specific areas of the brain which are associated with specific activities.
Like the Broca’s area of the brain also called the convolution of broca; this is the region
of the brain that contains neurons involved in the speech function. So, this area is located
as you can see in the frontal lobe part of the left hemisphere of the brain, and it was
discovered in 1861 by a French surgeon named Paul Broca who found that it serves a
vital role in the generation of articulate speech. So, the Broca’s area actually is needed in
terms of speech generation.
On the other hand, how do you know that what you are talking about is making sense
that comes you know that was discovered after nearly about 50 years after Broca’s
discovery that is by the Wernicke’s area. So, Wernicke’s area is the region of the brain as
you can see that it is coming on the temporal lobe backside of the temporal lobe of the
left temporal lobe of course. And this is where it is important for the language
development, and it is responsible for the comprehension of the speech while broca area
119
is related to the production of the speech. So, language development or usage can be
seriously impaired if there is a damage to Wernicke’s area.
Now, sometimes there can be a good language development, but suddenly the
connectivity may get lost, in that case you will get a disconnection lack of conduction
and that is known as a typical disease called aphasia. Let us look into the case where this
Broca’s area loses contact with the Wernicke’s area. So, I have gathered two small
YouTube videos in order to you know you to in order for you to appreciate that what
happens if this connection gets lost.
My final intention is to tell you through these that indeed there are parts in the brain
which are specifically trained for specific activities. Let us look into it what kind of
aphasia happens if this Broca’s area and Wernicke’s area lost touch with they lose touch
with each other. Let us look into that.
120
What is so interesting about this lady, she clearly is very intelligent and she clearly
understands the instructions and she clean can express herself in a very good way, but
you could see how after a while she completely disconnects; the what she wants to do,
what she is heard she should do and what she actually does.
This is a conduction aphasia. So, in this case in this conduction aphasia what we will be
actually getting is that the person will not be actually able to get all the you know
numbers, that the person is asked. So, these ladies having conduction aphasia.
In this case, you will be getting another type of an aphasia and another type of an
conduction aphasia, where this person will be actually losing the talk you know the
capacities to actually comprehend what he is talking about ah.
121
Now, let us look into the brain from an input-output point of view; as I told you that the
central sulcus, play a very important role. Behind the central sulcus we have the parietal
lobe part of the brain, and on the parietal lobe I am showing here two regions; the
primary somatosensory part, and a part of somatosensory associations.
So, primary somatosensory part they actually picks up all the signals from the thalamus
region, which are related to somatosensory related senses. For example, you are touching
something if you remember the finger example, then it comes to this primary
somatosensory part. And this association part actually mixes these signal along with
other signals that we get from; let us say from the visual association, let us say from the
temporal association, all these things the association happens here, like you get the full
understanding of the situation.
In the backside, the other input is coming from the in the occipital lobe that is where is
the visual association is coming and here also the green part, this is where is the primary
visual input comes and this is where the visual association comes. And similarly, you
have the primary auditory input coming here and you can see the sylvian fissure of the
lateral fissure, below which you can see the temporal lobe here. So, the primary auditory
inputs are coming here, and then you have the brocas region somewhere here and the
Wernick’s area, they actually help in order to process the things and also to create the
effector, so that you can speak you know something meaningful.
122
So, this is how all the inputs are coming in the back side of the brain. If you look at the
mapping of it of the cortex on the left, you have the input coming on the back side of the
brain. So, this is from the central sulcus onward in the parietal lobe part as you can see
that it starts from the toes, hips and trance and all these regions, wriest and you can see
that the hand plays such an important zone in large amount of cortex areas devoted
towards understanding somatosensory activities from the hand. No wonder that you
know, human beings they have evolved in a way that they have learned how to
maximally use their hand in comparison to the leg.
And the next very important region is the face region; you can see that how each part of
the face like sensing from the brow, from the eye read, from the face itself, the lips, their
jaws, swallowing, everything comes in this big region. So, there is a large part of the
parietal region which is devoted towards the somatosensory homunculus. Now,
homunculus means a small homosapiens almost as if sitting here.
Now, just like that if you look at the frontal part, this is where I have shown the frontal
part. In the frontal part, if frontal lobe you have the motor homunculus part and it was
fast that Wilber Penfield actually denoted all these regions. Now, what you will see in
the motor homunculus part, just like these somatosensory part it will start with these
genitalia toes, foot, leg, hip, etcetera; then this is the effector part.
123
So, this is how you will be able to move your fingers, and you can see how large this
index finger is taking the place in the brain. And then you can see the eye, the nose, face,
upper lips, lower lips, teeth, tongue, pharynx, and the abdominal part; so there is a large
part once again here.
So, you can see that this is the region through which we will actually control the
movement of these lips, so that is the motor part of it and that is from the frontal cortex.
And from the parietal cortex, this is the region where we are taking the senses the
somatosensory senses of different parts of the body.
As I am giving you a lot of information, let us try to once again summarize them in terms
of the major lobes of the brain. There are four major lobes in any hemisphere of the
brain; these are the frontal lobe, the parietal lobe, the temporal lobe and the occipital
lobe. The role of the frontal lobe is in terms of the executive functions, social
stimulation. So, they are they call that the frontal lobe is what is the (Refer Time: 48:30)
sign of homosapiens.
So, all social stimulations, emotional expression, sexual behaviors, working memory,
language, decision making and the control panel of our personality, our ability to
communicate, everything comes under the domain of the frontal lobe.
124
Parietal lobe mostly senses, it integrates the sensory information from various parts of
the body, knowledge of numbers and their relations, in the manipulation of objects. Its
function also includes processing information relating to the sense of touch. Then we
have the temporal lobe which is having the primary auditory perception, such as hearing
which holds the primary auditory cortex region; and also it receives sensory information
from the ears, secondary areas process the information into meaningful units such as
speech and words. Also the temporal lobe takes place in terms of as I told you that the
speech recognition and stability, and also in some part the depth recognition.
Occipital lobe that is the very back side of these cortex area, it participates in vision
processing. It processes and interrupts everything we see that is in the back side of the
brain. The occipital lobe is also responsible for analyzing content, such as shapes, colors,
and movement, and also for interpreting and drawing conclusions about the images we
see.
Now, in the very back of course, this is not part of the you know the neomammalian part
that is not of the part of the cortex, but an older part of the brain is the cerebellum part,
but it is visible here. The role of the cerebellum is in terms of you know, receiving
information from the sensory system, from spinal cord, other parts of the brain, and then
regulate the motor movements.
We say sometimes that all these things that have to be learned first is learnt in the frontal
cortex, parietal cortex; and then it is passed on to the cerebellum, so that helps us in
terms of posture balance coordination, and balanced muscular activity that happens
through the cerebellum. So, we have to keep these four important lobes in our mind
allowing with the cerebellum and their functions.
125
Now, what happens if we if some of these lobes in the sensing region, we have said that
there are specific regions for each of the sensing; what if they get mixed that can create a
problem called synesthesia. It is a neurological condition in which stimulation of one
sensory or cognitive pathway may lead to automatic, involuntary experiences in a
secondary sensory or cognitive pathway.
For example, when you are reading you can get a vision, like you are reading the
numbers those who are having synesthesia will also see we call us, as they will be
reading the numbers; even if the numbers are not colored that is the you know that is an
interesting condition. So, this actually tells us once again that there are distinct regions
and what happens if the distinct regions actually get somehow cross connected.
Now, we have seen different parts of the brain from the top. Now, you may wonder that
where from the cognition the self identification, the perception, the thinking process,
where from it finally comes; all these areas that we have seen, they are mostly like
sensors and actuator controlling area. Deep inside the brain, deep inside the brain we
have the areas which actually generates the sense of the perception of the self.
126
You can see that you have the cortex area here, and deep inside you can identify this
region as the thalamus region. For example, and here you can see these ascending
reticular activating system or ARAS; this is where the nerve signals are all going up.
And this is where they are all you know, a there is a huge nerve junction that is
happening here that junction of nerves is what is the reticular system, so there is this
ascending reticular activating system. Now, it is a believed that this ARAS system is
thought to be required for the normal conscious waking state of a homosapiens.
127
Now, so far what we have seen in the human brain is from a top-down approach, it is like
a analogous to a step pyramid of the Machu Picchu, it is a top-down approach as you can
see that there are different divisions there. But the point is that even though it is a top-
down, because also to say a bottom-up that is the way the brain actually a grew with
respect to time in the evolutionary time rule, but they are all interconnected; in fact you
can see here that they are all laterally connected.
So, even if it is having a kind of a top-down kind of an architecture, but there are lot of
you know connections between different parts of the brain. So, it makes it as a
fascinatingly integrating complex system. In the next talk, we will look into some of the
deep brain structures and their roles.
Thank you.
128
Lecture – 06
Good morning students. Welcome to the Foundations of Cognitive Robotics. In my last

lecture, I told you about the top part of the human brain. Now, today I am going to take a
look inside what is there in a human brain. Some of the things that are hidden from the
top we are going to find out that what is their inside it.
To begin with I will first talk about how this entire brain system has evolved. There are
some parts of the brain which are quite primitive part, some parts which are relatively
more modern and some parts which are very modern. Generally, it is considered that this
part of the brain this is more or less the primitive part of the brain. So, this is the
primitive part, very primitive part you can even see it in some of the very primitive
organisms also. So, this is this is the primitive part, ok. So, this is the primitive part.
Whereas, there is some part which is somewhere here you may say that which is
relatively in the middle phase of development, so it is not very modern, but it is
something which has developed after the primitive part. So, this is relatively modern I
say and just I had just put it as relatively modern, relatively modern. And there is some
129
part which is very modern. So, this is the part which is very modern. So, we can just call
it to be a very modern part of the brain.
Now, when I have discussed about the brain structure in my last class, I have mostly
focus on this very modern part of the brain that is the telencephalon part of the brain.
Now, of course, if I look at it from the other way that is from the primitive side then
there is this important part which we will be calling as the brain stem part of it which
includes these you know these areas like the hind brain, the mid brain, so that is the brain
stem part of it this is what we will be focusing more today.
So, the area which is above the spinal cord, and then on top of that the brain stem part
there are also areas which are like diencephalons, and also there are some parts on top of
that which are you know some part of the telencephalons which also we will be
discussing today which are all relatively modern, ok. So, that also we will be discussing
today.
So, these are the three things, one by one we will start from the very primitive part. We
will talk about the brainstem part where we have this rhombencephalon part of the
hindbrain, and the mesencephalon part of the midbrain, then the diencephalon. And then
we will gradually go towards the top where we have the basal ganglia and we have
cingulate cortex and such areas. So, we will talk about them next.
130
Now, if I actually look into the different brains if I try to compare the brains of say
something like the very higher order animals, like say the mammals, and compare it with
reptiles, like say turtles. Now, you would see that these three important parts of the brain
that is this telencephalon midbrain, and the cerebellum part of it they are mostly present
in almost all of these, ok.
So, it is not that the mammals are only having telencephalon. The same is also present in
birds, is also present in turtles, is also present in amphibians and in fishes. And the same
thing is true for the midbrain and the cerebellum part. Only thing is that the proportion of
each one of them is quite different from species to species. So, that is something that we
have to be very careful about when we look into it.
If we look at how a turtle responds and also if you look at that how for example, a
mammal like a monkey is responding.
131
If you compare between the two you would see that in case of a turtle and in case of a
macaque the common part is that both takes let us say the input from the vision both take
it towards the hippocampus. So, that is common between both of them.
However, whenever we are talking about a monkey it goes through so many different
regions in the brain. So, the path of the signal is quite complicated. On the other hand, if
you look at the turtle that path is very simple and straightforward relatively in
comparison with the macaque.
So, that is when you may say that even if both of them are having the same elements like
say the telencephalon where the hippocampus is there or does that these you know
sensory organs. But in one case they actually go through a vast many regions of the brain
different ganglia’s or ganglia like you know nuclei’s are involved in it, whereas, in the
turtle’s case that is not involved. So, the visual association is much less that is what
possibly is the major difference between the way the signals are processed by the turtle
and the signals are processed by the macaque.
Now, all these things have very important consequences as far as the behavior of
different types of animals are concerned. They are cognitively different because of that
they have different ratios or proportions of the brain areas and also they have different
associations of networks in the brain areas.
132
Now, let us look into the brain structure and let us see that what are the hidden elements,
and then we will go back to the slides and we will see how we can correlate between the
two. If I look at the brain structure, this is what if you remember we have discussed in
the last class also.
Then you can see that if you try to remember that this was the prefrontal cortex and this
is the frontal lobe part and then you can see the back side we have the parietal lobe and at
the very back we have the cerebellum part of it, ok. This is all as far as if you take the
brain, if you look at it from the top this is what you are going to see.
Now, let us try to remove it and let us see what is there inside. So, let us try to do that
very carefully. And let us try to separate through the longitudinal fissure. You remember
this is the longitudinal fissure let us try to separate one brain out at least, so that is what I
would be trying to do and I will keep the rest out.
133
Now, if you look at this part which is you know you can still remember that this is where
was our frontal cortex areas, and you can still remember that in the back side we had this
temporal lobe part of it. And what is there inside which is hidden is actually the brain
stem part of it. So, you can see that in the brain stem. We have these areas which is
known as medulla oblongata, then we have this area which is known as the pons area,
then we have the midbrain here.
This you can see this, this is the mid brain area and then you can see this beak here. This
is the area which contains the hypothalamus that is the easiest way to note down the
hypothalamus. And, this is always indicating towards a gland here which is so very
important pituitary gland which actually controls the behavior you know in terms of the
hormonal behaviors.
Now, and on top of these hypothalamus we have the thalamus here and in this side we
have the pineal gland, which contains the melatonin’s. So, this is a very primitive
structure of the brain, ok. And beyond this we will be getting more and more relatively
modern structures. I will talk about them one by one. But this is something that we have
to keep in our mind that the brain stem and the cerebellum which was somewhere in this
side. This are some of the very early developments of the brain.
134
If you look at the brain then from bottom up then as I told you that this is the brain stem
part which starts with the pons you know pons is somewhere here pons means a bridge,
so that is somewhere here. So, below that you have this medulla oblongata. So, medulla
oblongata then pons and then the midbrain and then the thalamus; so, that is the very
primitive part.
Now, from there if you look at these slight something more we have added here. What
we have added here is the hippocampus and the amygdala’s. And then from here if you
go further up you can see the fluid ventricles, fluid ventricles are needed to supply the
nutrition to the brain that is added. Then, if we further go here what we can see that from
the fluid ventricles we are actually coming to the basal ganglia part of it, ok.
So, once we are in the basal ganglia, we are nearly reaching towards the modern part of
the brain and then from basal ganglia on top of it we have the white matters. As I told
you earlier that these are highly myelinated nerves and then above the white matter we
have the gray matter the cortex part of it and you can see the cerebellum in the bottom.
So, the best way to actually keep an idea about the brain structure is thus you can
practice it that how to grow it from the bottom till the top.
Now, once you practice that how this entire brain from the lower part which is connected
to the peripheral nervous system through the spinal cord, how from there at the brain
stem pons mid brains etcetera and then beyond that how it is connected once you know
135
you have got an overall idea about it. Then we can actually let us try to go one by one
inside this element. So, let us first focus on the brainstem itself on different parts of the
brainstem different components and what is their functionalities.
If I look at the brainstem we can see that this is the area that we are referring to and a
very important area here is pons and also the medulla oblongata is somewhere here and
the midbrain. So, that is the part that we are talking about which is basically posterior or
the you know the back-side part of the brain adjoining and structurally continuous with
the spinal cord.
Now, this part, the entire brainstem part has important relay you know roles in terms of
the regulation of cardiac and respiratory function, consciousness and the sleep cycle. So,
all these things you know whatever is the autonomous part of the nervous system, they
are definitely controlled from here.
So, it consists of the three regions as I told you medulla oblongata, pons, and the
midbrain. And this is indeed, even if it is very primitive this is indeed a very important
part of the brain as all the nerve connections from the motor and sensory systems of the
cortex from the top that also passes through it to communicate with the peripheral
nervous system.
136
Apart from the fact that it controls many of the voluntary movements of the nerves, but
also you have to, also make it a point to see that all the sensory nerves, the
somatosensory connections they are actually coming through this medulla oblongata.
And, then the pons and the midbrain, and then they are coming to the different regions
through the thalamus to the different regions of the cortex.
So, essentially you know everything that is happening in the top part of the brain they are
all getting connected through this area. So, some of the sensor signals this part itself
processes, and for some it works like the passage, the connectors from which this all the
signals will be going towards the top part of the brain. That is something we have to keep
in our mind.
The first important sub element of the brainstem is the medulla oblongata. It is the lower
half of the brainstem which controls the autonomic functions and connects the higher
levels of the brain to the spinal cord even though it makes up just 0.5 percent of the total
brain weight, but yet it has a very important role.
So, it regulates some of the very basic functions that is needed to leave the autonomic
nervous systems, which includes respiration, cardiac function, vasodilation and all
natural reflexes like vomiting, coughing, sneezing and swallowing. Say for example,
respiration is carried out by the you know these kind of with the help of chemoreceptors.
Cardiac centers are actually controlled through sympathetic and parasympathetic systems
137
which these medulla controls. Vasomotor through the pressure receptors, baroreceptors
the blood pressure and the reflex centers through the they control these all these reflex
actions that I was talking about.
On the top of the medulla oblongata, we have the pons region which is a relay station
between the forebrain and cerebellum that passes the sensory information from the
periphery to the thalamus. And it is a Latin word of bridge because it is joining between
the thalamus area midbrain thalamus area and the area below that is the medulla
oblongata and the spinal cord.
Now, it also has nuclei that regulate sleep for example, respiration, swallowing, bladder
control, hearing, equilibrium, taste, eye movement, facial expressions, facial sensation
and posture. So, here the pons control some of the non-autonomous sensory systems
also, not just the autonomous sensory systems. And it also contains the sleep paralysis
center of the brain and plays a role in generating the dreams. So, that is about the pons.
138
Now, we come to the midbrain region. So, we have talked about the medulla, we have
talked about the pons and now we are in the midbrain region. Midbrain or so called
mesencephalon is a portion of the central nervous system which is associated with things
like vision, hearing, motor control, sleep and wake system, arousal and of course, the
temperature regulation.
In fact, this a system is one of the most archipallian in origin, sharing its general
architecture not only with our other primate you know other groups of primate species,
but also with many of the vertebrates also. And a very important region in the midbrain
where we have the substantia nigra, they actually create dopamine and that plays a role
in motivation and habituation of species from humans to the most elementary animals
such as insects. So, that is one of the important source from midbrain onwards the
dopamine production. So, that is the importance of the midbrain.
We have talked about the midbrain part which is somewhere here. And now we will be
going to the thalamus part of the brain which plays a very important role, I will tell you
later on in terms of cognitive processes. So, it is in the top of the midbrain and it is below
the you know the cortex areas etcetera. So, somewhere in between which connects
between the lower part of the brain and the upper part of the brain that is where is the
thalamus is coming into the picture.
139
The thalamus as you can see that it has two parts in it, the right thalamus and the left
thalamus. And it is just in the middle part, it is just at the central part which connects
between the cortex part, and the cerebellum and the on the other hand the brain stem and
everything.
So, it is located above the brain stem and between the cerebral cortex and midbrain
which has extensive nerve connections to both that is what is thalamus. And it is the
largest structure which is in fact, derived from the embryonic diencephalon part of it.
You remember I have talked about that just below telencephalon, we have the
diencephalon. So, this is the diencephalon part from which actually these you know the
thalamus is actually derived.
Now, the main function of the thalamus is to relay motor and sensory signals to the
cerebral cortex of course. As I told you that it works like a junction station and except
the olfactory signals. Olfactory signals are directly connected to the brain or the cortex
part, but except that all other sensory signals actually travel through the, they have to
travel through the thalamus that is very important.
Now, this thalamus also has some additional parts outside that is the epithalamus region
and the most important thing is the Pinial body and the Habenula nuclei and Caudate
nucleus. Out of that I will just only talk about the pinial body which is a very important
one in terms of course, reptiles it plays a very important role, in terms of the color
140
change, for example, for the higher order animals also pinial body has some important
functions.
Now, you can see the position let us say in that entire picture you can see that the this
this particular gland the pineal gland is somewhere here. You can always see that it is
opposite to the hypothalamus that big bird’s beak and the opposite to it you can see the
small region here that is where it is the pineal gland.
Now, this is part of what you call epithalamus and it is a dorsal posterior segment of the
diencephalon. So, it is the posterior the back side of diencephalon and that includes the
habenula and other interconnecting fibers also.
So, the pineal gland which is a very small pinecone shaped endocrine gland and these
actually controls the you know hormone called melatonin, and this hormone related to
you know serotonin it is secreted by the pineal gland and it stimulates color change in the
skin of reptiles. In case of mammals, the same gland is actually involved in terms of the
sleep wake, and the reproductive cycles in mammals. They have a pineal gland has a
very important role to play. So, this is in one of the epithalamus systems that this pineal
gland comes into picture.
We have seen the thalamus part of the brain. Now, let us look into this bird’s beak part of
it which is known as hypothalamus. All our well-being our homeostasis, that is
141
controlled by the hypothalamus. It is also one of the very primitive parts of the brain. So,
let us look into the functions of the hypothalamus. And as I told you that in from close
very close to the beak we have this pituitary gland which is actually controls. So,
hypothalamus controls the pituitary glands, they actually secrete the hormones and that is
actually controls many of our organic behavior.
So, let us look into the hypothalamus. Hypothalamus is as I told you the main role is to
keep the body in homeostasis condition. So, this is a homeostasis condition. That means,
you know that your body temperature, for example, appetite, heart rate, blood pressure,
everything it is kept under control by the homeostasis.
So, it actually gives you a sort of healthful balanced bodily state. That is the role of the
hypothalamus. So, it has a very important role. And as you can see I told you that this is
the big region that is where you have the hypothalamus and it is so close to the pituitary
gland that is what it controls directly.
So, it is a part of the part where you know many essential functions of the body such as
the body temperature, the thirst, appetite, weight control, emotions, sleep cycles, sex
drive, childbirth, BP, heart rate, production of digestive juices and the balancing of body
fluids. These are all the role of the hypothalamus. So, it has a very important role to play.
142
Now, hypothalamus has actually it is not a single monolithic thing, it has many important
what you call nuclei’s inside the hypothalamus.
Let us look into these nuclei. So, some of the important nuclei for example: this nucleus
paraventricular nucleus. So, this is one of the important nucleus paraventricular nucleus.
Then also you have dorsomedial nucleus, mammillary body, then there is a ventromedial
nucleus. So, and here there is a supraoptic nucleus.
So, all these they play different important roles. For example, whatever you know
information, the contextual information that comes through your cerebral cortex,
amygdala and the hippocampal region that actually comes to the hypothalamus.
And these different regions that I have talked about, they actually also get input from the
sensory inputs. The sensory inputs are coming from the bottom side, ok. So, sensory
inputs are coming towards the hypothalamus and from the cortex you are getting the
input.
So, from both the sides you are getting the inputs. Based on it you are actually taking a
decision and that what is this state tells me that, what will be my set point. For example,
set point I will call it and so let us just try to keep this in our mind that this is the part
which will be part of the set point and this is the part which are going to give us the
inputs.
143
So, we compare between the set point and the input that is what the hypothalamus is
going to do. And then it takes the actions. And these actions go to the visceral motor,
somatic motors, neuroendocrine, behavioral responses or some actions goes towards the
motors, and towards the muscles, and some towards the secretions of different types of
hormones that further controls the organs. So, that is the role that is played by the
hypothalamus.
We have talked about thalamus and the hypothalamus. Now, we will gradually enter
towards the cortical region, the cerebral cortex region of the brain and we will be just at
the base of the cerebral cortex region of the brain. So, if you look at it what I am talking
about we have covered all these regions and now we will be talking about this region
where just at the base we are having some people call it as a limbic system, the border
system.
But a very important part some of the very important parts of the border systems will be
the three things, one is that we will talk about here the basal ganglia part. And we will
also talk about here the hippocampus part somewhere here, and we will talk about on top
of it just below corpus callosum we have actually what you call the cingulate,
particularly the anterior cingulate, ok.
So, the entire cingulate cortex that is the part beyond which the actually all other cortex
that I had already discussed the frontal, prefrontal, parietal cortex etcetera is there. So,
we have just then reached the cortex part of the brain. Let us look into their functions and
their you know characteristics.
144
One of the most important part here is the basal ganglia part which is essentially a group
of subcortical nuclei of varied origin and you will find generally in the brains of
vertebrates including the humans and this is situated at the base of the forebrain and at
the top, at the top of the midbrain that is where is the basal ganglia part of it.
Now, this is associated with a variety of functions. It controls for example, the voluntary
motor movements. You will see that if this is lost in many of our voluntary actions that
are lost procedural learning for example, ok, habit learning, eye movements, some
people say good bit of cognition and emotion.
These are the things, particularly there is an area which is adjacent to the basal ganglia is
the amygdala that controls many of our emotions. So, what happens if the basal ganglia
do not work? Then you get two different types of disorders like the Parkinson’s disease
and the Huntington’s disease.
Now, in the case of Parkinson’s disease you will be having a gradual loss of the ability to
initiate the movement, whereas in case of Huntington’s disease you will be unable to
prevent parts of the body from moving unintentionally. So, that is what is the
Huntington’s disease. This is in fact, a very fatal disease.
Now, in most regions of the brain the predominant classes of neurons actually use
glutamate as the neurotransmitter and it actually have an excitatory effect. On the other
145
hand, in basal ganglia great majority of neurons in basal ganglia actually use something
called GABA, gamma aminobutyric acid and that works like a neurotransmitter that have
inhibitory effects on their target. So, by nature this is somewhat different because it
works on the GABA.
So, if you just see some of these basic disease for example, what happens in the
Parkinson’s disease. So, you can see it in a very small video. Parkinson’s disease or PD
is the second most common neurodegenerative disorder after Alzheimer’s disease and
affects 1 percent of individuals over the age of 65.
146
It is classified as a hypokinetic movement disorder and occurs due to degeneration of

dopaminergic neurons that project from the substantia nigra in the midbrain to the basal
ganglia. The initial precipitant of neuronal death is unknown and by the time symptoms
emerge between 50 to 80 percent of dopaminergic neurons have already died.
The cardinal features of PDR, tremor, rigidity, akinesia or bradykinesia and postural
instability easily remembered with the mnemonic trap.
147
The most essential feature for diagnosis of PD is bradykinesia which is characterized by

a reduction in both speed and amplitude of voluntary movement. This is most striking
during repetitive movements like finger tapping, where tabs are slowed and become
progressively smaller in amplitude or writing where letter size decreases across the page
also known as micrographia.
148
Bradykinesia is also manifested as a loss of facial expression or hypomimia, reduced
vocal volume or hypophonia and slowing with a variety of other activities including
walking, dressing and turning in bed.
The tremor in PD is typically a unilateral, rhythmic, resting tremor between 4 to 6 Hertz

and can classically start as a pill rolling tremor of the thumb and forefinger.
Rigidity is an involuntary increase in muscle tone present throughout the range of

movement and can be tested by passively moving the arm, neck or leg. If rigidity is
149
smooth and consistent throughout it is called lead pipe rigidity. If however, there is a
ratchet like quality it is called cogwheeling, which is thought to be due to rigidity
superimposed on an underlying tremor.
Parkinsonism also causes characteristic gait abnormalities like stooped posture, small
shuffling steps, reduced arm swing and freezing. So, make sure to ask your patients and
watch out for falls.
150
PD is treated with levodopa which is a precursor of dopamine. The neurotransmitter that
is deficient in PD. An excellent response to levodopa in a patient with asymmetric tremor
predominant Parkinsonism is supportive of the diagnosis of PD.
It is important to keep in mind that PD is the most common form of Parkinsonism, but
patients may develop Parkinsonism from other causes including medications, such as
antipsychotics. Other degenerative disorders such as multiple system atrophy or
progressive supranuclear palsy, and Wilson’s disease which causes copper deposition in
the basal ganglia, alright.
So, we have reached up to the basal ganglia which is at the base of the cortex. Now, in
the cortex particularly a near the you know temporal lobe we have a very important
region called hippocampus which is very important in terms of the memory of the brain
system. So, let us look into the hippocampus region.
151
Now, look into the hippocampus region which has a major role in learning and memory,
and though even though it is situated in the temporal lobe it is essentially a part of the
border part of the brain in the limbic lobe.
And it has also roles in terms of spatial navigation, emotional behavior, so not just the
memory and regulation of some of the hypothalamic functions. And it is the biggest
source of the theta rhythm. This is important. You have to keep in your mind; at a later
stage we would see in terms of the origin of the theta rhythm and its use, hippocampus is
one of the biggest source in terms of the theta rhythm.
You can see the position of the hippocampus here. So, it is you know that is the
hypothalamus region and then on top of it you can see the hippocampus. So, that is the
way. So, it is in this is it has connections with thalamus and also with the hypothalamus
this entire region, together you have the hippocampus.
152
Well, the next important one is actually the cingulate cortex without which we cannot in
fact; develop the cognitive model of brain. So, cingulate cortex plays a very important
role in developing the cognitive model of brain.
Now, this is on the medial side or inner side of the cerebral cortex, as I told you that it is
in inner side. And it consists of cingulate gyrus that is the topper upper part of the peaks
and its continuation into the cingulate sulcus. So, you remember I hope that the gyrus
and the sulcus, the peak and the valleys, the peak and the valleys. So, both of them you
know you will see the cingulate cortex.
Now, this is just above the corpus callosum on the medial side of the brain and although
it does not fall into any of these typically frontal parietal occipital part, because it is just
above the corpus callosum. Corpus callosum, if you remember actually joins the left and
the right side of the brain etcetera, but it is an essential part of this limbic system. It is
used in for example, things like error detection, ok.
Also, it has a very important role in social evaluation; that means, you know what is our
role in the society, it continuously testing. Increased level of consciousness that is what
without singular cingulate cortex you can think of it. And, it has been shown that this
area is greatly active in people who are emotionally more aware than other people. This
part is also responsible for things like pain, memory and learning which are controlled in
this region.
153
We have talked about some of the hidden brain elements; we have also earlier talked
about the cortex, now it is the combination of all of these things together that actually
makes the whole brain function. Towards this direction different people have tried to
develop different models.
There are some very early models in 1960 where people try to actually you know call it
is the MacLean model, where it is developed in terms of the brain is developed as a
triune model of three parts, that is a reptile part of the brain. Then the paleo mammal part
of the brain and that is the very early mammal part of the brain and then the primate part
of the neo mammal part of the brain.
However, today we know that these are not exactly the same. But we can still say as I
told you earlier that like the brain stem part and the midbrain part this is the part which is
like you can consider it to be the primitive part of the brain, ok. Then, you have the more
and more as the evolution progress, you get more and more developed part on the top
side of it. But still this model gives us some important behavioral matching, ok.
So, even historically and it is a useful concept to begin with let us look into this model,
and then we will see what are the some of the more advanced model towards this
direction.
154
If you think of the gradation of human brain as I told you this MacLean’s model and that
talks about this a triune brain for example, ok, the triune brain part in evaluation. Now, it
as I told you that they try to explain the brain in terms of three parts, the reptilian brain
composed of the basal ganglia and the brainstem part which has the medulla oblongata,
pons and midbrain, etcetera. And that is involved with all the primitive drives related to
thirst, hunger, sexuality and territoriality as well as habits and procedural memory, ok.
So, that is the one-part reptilian brain. Then the paleo mammal brain which really
includes the hypothalamus, hippocampus and amygdala, cingulate cortex etcetera which
is the center of our motivation like you know emotions, memory, ok, dopamine’s for
example is motivation and behavior such as parenting that was one of the suggestion. Of
course, today we know that it is not parenting is not only for the mammals, but also for
the reptile birds for example, there are these parenting behaviors.
And, then this model said that there are new mammalian parts consists of the neocortex
which enables higher order things, like language, abstraction, reasoning and planning.
This is somewhat true that really the cortex for the top part that is involved with many of
these important parts of the work. So, this was the first attempt to describe that how
different parts of the brain are involved with different nature of works.
There are in fact, some commonly studied conscious and unconscious brain events on the
basis of these. For example, you can see the whole list, I and you can just you read it for
155
your reference. Some of the conscious brain events could be something like immediate
memory, explicit cognition and things like you know strategic control, ok, things like the
rehearsed items in working memory, wakefulness and dreams, ok, intentional learning,
normal vision. So, these are all part of the conscious you know brain events.
Now, if you look at the unconscious brain events then longer term memory that is one of
the unconscious brain. You do not know, but it is happening continuously. Procedural
memory; the brain is keeping the procedural knowledge continuously in some areas even
without discussing with you apparently, automatic control for example. For all the
automatic nervous systems you know the parasima the sympathetic nervous systems for
example.
Then the deep sleep coma sedation, ok. So, these are the some of the things which are
controlled by the unconscious brain elements, so brain events. So, this is what you know
we can keep in our mind and you can try to find out that what areas of the brain actually
control the conscious versus the unconscious brain events as a sort of an exercise.
We will now talk about three of the greatest minds in brain science who have actually
who have very important contributions in terms of developing the cognitive architecture
of the brain. As I always told you that if you have to develop a cognitive robot, you have
to understand that what is the cognitive architecture of the brain itself. So, this three
people have you know phenomenal contribution in the field. Let us look into their
proposed models and let us look into some of the modern views towards that.
156
We will first begin our journey with Francis Crick. Many of you have heard of Francis
Crick as the discoverer of DNA, he in fact got his Nobel Prize for this. But between 1979
to 2003 Crick championed something else. He worked with his collaborator Christof
Koch and he developed a framework for thinking about the conscious mind, where from
the consciousness rises.
His aim was to map all the concepts associated with the concepts associated with
consciousness to properties of synapses to properties of action potentials, action
potentials and neurons. So, in fact this is what is called NCC, neural correlates of
consciousness in the brain.
So, he is the first person who started focusing on the fact that it is there is some
biological connections between the consciousness and the biological elements. And in
order to know that you have to start with the neuron itself, the neurons, action potentials,
synapses and how the neural’s network that was the first you know initiation towards this
field.
157
The next is also by another very famous Nobel laureate in neuroscience that is Gerald
Edelman. Now, Edelman expressed consciousness in terms of the morphology of the
brain. His theory is known as Neuronal Darwinism is based on three basic tenets that is
you know it is based on the development of selection, experiential selection, and the
reentry part of it.
Now, when a you know for example, a child is first developing, at that time the child is
having billions of neurons in him, but at the first phase this is the first phase of
developmental selection, it involves the somatic selection of neurons, ok, so during the
growth and development. So, you can see that this is where the fast neurons are coming
and at the developmental phase some of these neurons are selected and the networks are
getting built and that is what is the somatic selection phase, that is what is the
developmental phase.
After the developmental phase then there comes a continuous process of synaptic
selection; that means, the joint selections and the reinforcement. Like you can see that
this particular region, this particular region, they are more reinforced, whereas this region
is not reinforced. So, this reinforcement starts and that actually and develop a new spatial
temporal event this new spatiotemporal map in terms of different sections or groups of
neurons. That is the second phase. So, this is the phase where this experiential growth is
happening.
158
And, finally there is a very interesting process of reentrant mapping that happens. And
what it tells is that consciousness is essentially a dynamic process where this reentry has
to happen. Reentry means a group of neurons which will continuously crosstalk between
each other irrespective of what input is coming to one and or to the other.
It will continuously you know kind of in touch with each other and that actually this
reentry would actually help it in terms of predicting what will be the output of the
system. So, there are several reentric region. And what Edelman said is one of the
important reentry region is possibly the thalamus region of the brain.
Beyond Edelman there is another person who had enormous contribution; his name is
Herbert Simon, another Nobel Laureate. Of course, in a somewhat different area in
economics in 1978, but he is considered to be a pioneer in artificial intelligence, in
decision making, cognitive psychology and computer science.
So, together with Allen Newell he built actually computer programs which can simulate
the human cognition, especially ones that would solve problems with a trial and error
like a generate and test system. These are called general problem solver. He has
proposed a concept of blackboard architecture, in which various items could be written
to blackboard by different let us say group of neurons nuclei and it was visible to other
systems that could read and write on the blackboard. As a model for human
159
consciousness, this blackboard resembles the theater of consciousness. So, that was one
of the important contribution.
Herbert Simon has also contributed towards developing a very interesting algorithm
suitable for AI, and this algorithm is also known as crest algorithm that is chunk
hierarchy and retrieval structures.
Now, this is a symbolic cognitive architecture. For the first time somebody has talked
about cognitive architecture, which is based on the concepts of limited attention, limited
short term memories and chunking; so, sort of learning. It is essential in the architecture
and it is a model as the development of network of nodes which are connected in various
ways.
Let us say you consider this graph, that first of all you are getting inputs from all the
different sensory organs, eye, ear, hands etcetera. Now, what in the CHREST model he
proposed is that, there will be some feature which will be extracted from these
mechanisms and then a very interesting thing happens. So, that is the first part a stimulus
is perceived and converted into a set of features. Then these features are actually sorted
by the tests of the discrimination.
So, there is a discrimination network which sorts these you know features and send them
to pointers to a node within the, you know these limited term memories. So, then these
160
are the long term memories. So, from this long term memory they actually send the
indicators of this long term memories to the short term memory. So, that is what you
know it actually goes here, to the short term memory. So, discrimination network thinks
of what are the points where you have to connect it then from the address of those points
comes to the short term memory.
And then finally, an action might be taken by the system you know if it is you know a
new learning process or else you start a new you know feature to extract. And many of
this process can be actually done parallel like eye movement and learning and many of
the other process you can do it parallel.
So, that is what was the CHREST algorithm. This for the first time an algorithmic model
had come up where you know you do first collect the features from the input, then you
take them through a discrimination network, check whether you have already associated
points, nodal points, with this experience and then those things where the associations
are that is the only thing you declare to the short term memories. And, if it is new some
new learning takes place and again you go ahead for the new inputs. That is the way you
know he tried to simulate the brain.
Now, there are counter examples of Simons this kind of a AI vision because there is no
place of body in his vision. Now it seems, all right that when we move, we walk, we talk,
we write, we dance; all these things are controlled by the brain. So, that is fine that the
161
brain quite obviously, controls the body, but the question is how could the body
influence our thinking. And it is something which is very important, because perhaps our
thoughts are not as free and independent as we would like them to be they are indeed
highly constrained shaped by our bodies.
So, body not only constrains, but also enables the thoughts in the way towards the which
you know the body gets advantage out of it. So, the crucial notion is to explain this
process of embodiment that is the counter to the Simon’s AI logic.
In fact, so there are these two logics, one group is these AI group. So, this is the AI group
where the body is just a peripheral extension and the central part is the brain itself.
Whereas, the other group for example, the Rolf Pfeifer and John Bongard’s group, they
talk about how our body, how our different somatosensory experiences as you can see all
around us they actually shape our thought process in the brain. So, both the approaches
we will actually try to cover while designing the cognitive robot.
162
Lecture – 07
Good morning students, welcome to the class of Foundations of Cognitive Robotics. We

are looking for the neural networks inside the brain in the 2nd week lecture. And now we
will look into the basic system that actually builds this network. That means, we will
look into the configurations the characteristics of the building block of the brain and the
central nervous system which is the neuron concept. So, today our focus will be on
neuron.
When we have discussed earlier we also told you that there are two approaches to the
brain body problem. In one approach we say that the brain is the fundamental of all the
processing’s and the cognitive behavior; hence, the focus has to be on the brain. On the
other hand, we have also talked about another group of thought where it is the periphery
it is the dynamic systems that constitutes the body and that actually influence us
influences the brain that was the thought process.
Now, one thing that is common between both of them is that in both the cases the answer
is in terms of the neurons and their formations of networks. So, before concentrating on
163
any one type of these systems we have to understand about these fundamental building
block or the neurons of living system.
Now, when we talk about neurons; if you remember that we have talked about central
nervous system and peripheral nervous system. We have talked about that how the brain
and the you know spinal cord itself so how these are the regions that are the part of the
central nervous system the brain and the spinal cord. And how the other parts are like the
part of the peripheral nervous system for example, the spinal nerves the cranial nerves
etcetera.
Now, in one thing that is common in both the central nervous system and peripheral
nervous system is the neurons. Of course, the neurons are having different characteristics
in these different systems. So, that is what we are going to have a look into it.
164
So, when we try to subdivide neurons in terms of the nerve cells. There are actually three
major functional categories under them; the sensory neurons, the motor neurons and the
inter-neuron’s. Now, sensory neurons actually carry information from the body’s
peripheral sensors into the nervous system for the purpose of perception motor
coordination’s etcetera.
So, if you just go back to the slide you would see that these are the things from which the
sensory neurons are bringing these sensory feedbacks and they are putting those things
into the nervous system into the central nervous system so that is the sensory neurons.
Sometimes we also call them as the afferent neurons. Afferent means something which is
carrying towards the central nervous system then it applies to all information that is
reaching the central nervous system from the periphery whether or not this information
leads to sensation.
Now, the other one that is from the brain to the muscles these are actually you know
functionally these are something which the motor neurons take care of and these are the
efferent neurons. So, afferent neurons and efferent neurons afferent means going inward
and efferent is coming back outward. So, they carry the comments from either the brain
or the spinal cord. So, if you look back again that either the brain or the spinal cord sends
back the comments in some of these nerves and these are the efferent nerves which are
the motor neurons.
165
Now, there are some which are neither at the receptors or nor connected to the muscles
they are in between. And they actually make connections sometimes between the sensor
to the motor neuron sensor neuron to motor neuron or may be sensor neuron to another
sensor neuron or motor to motor so these are called interneurons. And they are in fact,
the most numerous and they are subdivided into two classes the relay type and the local
type.
The relay or the projection interneurons they have long axons and convey signals over
considerable distance from say one brain region to another. The local interneurons they
have short axons because they form connections with nearby neurons in local circuits.
So, these are the four sensory neurons, motor neurons and interneurons.
So, now that we know functionally that there are these sensory neurons, motor neurons
and interneurons; we would like to see that what is common in all these neurons. And we
would like to see that in all these neuronal structures; what are the issues that one has to
look at?
As neurons are the building block our brain actually organizes all the perceptions be it in
terms of storing in the memory or be it in terms of immediate behavioral responses and
these are accomplished by the neurons itself. And there are very large number of these
neurons that are presented in our brain which is in the range of about 10 to the power 11
neurons. So, you know these are these are huge mind boggling numbers of this neurons
166
and they are of different types at least a thousand different types of neurons you will find
in the brain itself.
These neurons as I have just now told you that there are a huge number of neurons that
are present in the human brain itself. And if you look at it in different types of species
starting from the lower order species to the higher order; you would see that one of the
thing that distinctly distinguishes between the species or as we are actually upgrading the
species. And that is in terms of the total number of neurons that is one and the second
point is in terms of the number of connections or the synaptic joints that these neurons
make.
So, let us look into it how this number of neurons or the synaptic connections they
actually change as we take our journey from the lower order systems to higher order
system.
If we look into the brain neurons from lower to higher order let us say we start with the
roundworms; it has about 302 neurons and about 7500 synapses. If you think of jelly fish
it is you know sub order of magnitudes higher about 5,600, for snails it will be about
11,000 neurons, for fruit fly it is huge 250,000.
For frogs now we will be talking about in terms of millions about 16,000,000 and for a
house mouse it will be about 71,000,000, when you come to a bird for example, flinch
167
this is about 131,000,000 so this will start to go towards the billions. And the same thing
is true for the parakeets also that this will be also having in terms of about a billion or so.
So, you can see that how as we are going from the you know lower order to the higher
order you can see that how these you know as the order is actually becoming higher and
higher order the number of neuron is actually increasing number of neuron has a
increasing trend.
And that is also true for the number of synapses because you can see that for roundworm
there is only about 7500 synapse whereas, for fruit fly you know already it is so high and
then for a simple thing like a mouse it is even higher. So, the synapses also actually
increased as the number of neurons increase the synapses also increased.
If you look at even higher animals let us say if you look at elephant for example, you
would see that this will be having about 251 billion brain neurons. If you look at say for
example, marmosets it has much less total number of neurons, monkey is about 6 billion,
gorilla is about 33 billion, chimpanzees about 22 billion and for humans they are having
86 billion.
Now, one interesting thing to note is that; the total number of neurons for a human is
much less in comparison to the elephant. This is you can see that this is 86 billion for
168
human the total number of neurons whereas, for elephant it is 251. Of course, you have
to consider that the size of an elephant is much bigger than the man.
So, it is not just the total number of neuron that would actually signify the level of
intelligence for that you have to measure the number of neurons in the cerebral cortex
itself.
In the cerebral cortex elephant is having only 5.6 billion neuron. Gorilla is having about
9.1 billion neuron and human beings are having about 16.3 billion neuron. So, it is this
high presence of the neurons, you can see it is very very high it is the highest. So, it is a
high presence of these neurons in the cerebral cortex that actually signifies the
intelligence in the system.
So far, we have discussed about the total number of neurons, their presence in terms of
different types of living systems. And now we will focus into each single neuron in a
system. When we look into these we will see that there are certain important aspects of a
neuron that we need to look into.
Now, not all the issues we will be covering in this lecture, but I will first make you aware
of the issues that are related to these neuronal single neurons you know what are the
issues that are important point.
169
There are five basic issues of a neural network; the structural component of an individual
cells, then the mechanisms by which the neurons produce signals within and between the
nerve cells, the patterns of connections between nerve cells and between nerve cells and
their targets, muscles and its gland effectors, relationship of different patterns of
interconnection to different types of behavior, and finally, how neurons and their
connections are modified by the experience, different experiences that will gather how it
is.
Now this is a vast subject of it is own. So what we will be covering basically right now is
the structural components of individual nerve cells and also in the subsequent lecture the
mechanisms by which neurons produce signals within and between the nerve cells.
170
If you look at a typical neuron; you would see that it consists of four important parts.
What are these four important parts? So, I will first of all there is a cell body and
secondly, there are these regions which are the dendrites. There are two types of dendrite
some dendrites actually go out from a process and then these are the apical dendrites.
Some dendrites are directly from the cell body or the soma then they are called the basal
dendrites.
Now, then we have the most important thing that is the axon so this is where is the axon
region that is the axon region. And finally, the axons aimed at certain interesting points
which are called synaptic terminals. And there can be many synaptic terminals as we
have seen in earlier that in different you know number of neurons can have ten times
thousand times million times you know connections in terms of the synaptic you know
synapses of the neurons.
Now, basically in this whole system the center of attraction will become the axon
because it can convey electric signals over distances ranging from 0.1 mm to 2 meter.
These signals which are also known as action potentials are initiated at a specialized
trigger region near the origin around these point these signals are triggered.
So, this is what of course, this is the configuration of a typical neuronal system. Now,
there are actually many differences between different neuronal systems. There are some
171
of the other points also which we will later on talk about as you can see that if this the
blue regions there these are giving you actually the myelin covers.
And between these covers you have the node of ranvier and you can see the myelin
sheath in the blue regions ok. So, between that you can see the nodes of ranvier, they
actually together form a very important part in terms of the mechanism by which the
action potentials actually travel ok.
So, the signals are gathered from different you know dendrites essentially. And they are
integrated in the soma once they cross a particular value then only the triggering occurs
and the signal starts to travel and then the signal reaches to each one of these synapses.
Again these synapses beyond a certain region they actually carry the signal to the next
neuron and that is how this whole thing; and this next neuron these are the this is the
presynaptic neuron the top part and the next parts after the synapse is the postsynaptic
part. So, this is the typical neuronal system from one neuron to the other neuron.
Now, if you look at the types of neurons; you will see that the common description that I
have so far told you there are certain finer variations of it. For example, the first variation
is in terms of the unipolar neurons. Now, unipolar means that it is having only one
process in it; that means, only one connections from the cell body and part of it is related
172
to axon and the other part is related to the dendrites. So, that is what is the unipolar
neuronal cells.
Now this is one of the simplest cells and this is something that you will usually see in the
invertebrates. For vertebrate’s unipolar systems you will see mostly in the autonomic
nervous system. If you remember that I have talked about sympathetic and
parasympathetic nerves. And there are these sympathetic nerves which directly controls
the breathing the heart rates etcetera. So, this is where the autonomic nervous system
where you will this is one of the primitive part of the human body where you will see the
unipolar neurons that are present in the system.
Sometimes you will see that not like you know from a single soma you are having only
one process that is coming out in terms of excellent dendrons, but there are more than
one processes that are coming out. So, if there are two of them that comes out then it will
become bipolar system and if there are many then it will become multi polar system. Let
us look into these different types of systems as well.
Now, we will look into what we call the bipolar neurons so these are the bipolar neurons.
For the bipolar neurons it has an oval soma and as you can see that there are two
different parts two different processes. So, the dendrites coming and in one part their
joining in the cell body.
173
And the other part is from where the axon is triggering and taking the signal out. So,
because of these two polar part this is the bipolar configuration many sensory cells are
actually bipolar. For example, sensory cells from retina, from olfactory epithelium of the
nose they are actually bipolar in nature.
Now, if you look into more complicated neurons then; you would see that there are some
neurons which are not exactly bipolar they initially start with a bipolar configuration, but
this is what we call them to be pseudo unipolar system. So, for pseudo unipolar system
there is these dendrites, there is this peripheral axon to you know and then the signal is
coming here.
But instead of having a different pathway you can see that this is where is these outer
axon central axons going to the axon terminals. And both of them are joined to the cell
body through a single process that is why it is called a pseudo unipolar. That is other
than these they are actually of two types.
Now, these cells develop initially as bipolar cells, but the two cell process actually fuse
into a single continuous structure; that emerges from a single point in the cell body.
174
We also have systems like multipolar neurons. So, as the name suggests that it will be
having many different processes you can see that here, here particularly you know
whenever you will be having basal dendrites you will see many of such things so here as
well as here. And this is where is the axon part and there are many inputs that are coming
in to it so these are actually multipolar many inputs taken by it.
Now, this is a very common type of a neuron in the case of vertebrates. And they vary
greatly in shape especially the length of the axon part that actually greatly varies
depending on the size of the living beings etcetera and also in terms of the extent
dimensions, intricacy of the dendritic branching. Usually the extent of branching
correlates with the number of synaptic contacts that they will be having.
For example, if we are talking about a spinal motor neuron; then there is a relatively
lower number of dendrites. Like something like 10,000 contacts and out of which about
1000 could be on the cell body and 9000 could be with the dendrites. But if you look at
the dendritic tree of a Purkinje cell then which is in the cerebellum part then it is much
larger and bushier and it can receive as many as million contacts.
So, depending on which region of the brain there you will see the number of synopsis for
a multipolar neuron will actually change.
175
These are some of the variation so for example; this is for the hippocampus we have
earlier talked about hippocampus. The neurons in the hippocampus are something like
these pyramidal cells and you can see that they are having so many basal dendrites
directly from the soma itself also they have a few apical dendrites and then it has the
axon.
And this in fact, phenomenally increases even more for the Purkinje cell of cerebellum.
As you can see that there is a huge number of synapses, you have to keep in your mind
that cerebellum is something which is used for many kinds of learning, you know
procedural memory related to it the motions, movements related to that you know
something that we for example, learn in terms of particular behavior motion or any
dynamic behavior.
So, you can easily imagine that how many neuron connections are required for such a
system. Now this is what is that is the Purkinje cell of a typical cerebellum system. So,
these are the variants of multipolar systems
It is very interesting to note that it is not just the neurons that actually develop a nervous
system like the brain and other central nervous systems etcetera. There will be neurons
which will be accompanied with something which is you know it is very extensively
abundant in the system. When we talk about the brain system for example, it will be
about 80 percent of it will be this type of cells and these are called glial cells. The role of
176
these glial cells is that they actually complement and they try to supplement the neurons
in terms of supplying.
Say for example, nutrition’s to the neurons, in terms of absorbing the additional
potassium and also sometimes the neurotransmitters excess neurotransmitters. So, they
ah they try to actually keep the neurons in a healthy level immune it from the diseases
and also help them in terms of you know facilitate them in terms of the neural
transmission.
So, that we will look into this has a very important role in terms of developing the neural
network. So, we will look into the glial cells now.
To look into the glial cells as I told you that their basic role is actually to support the
nerve cells and they greatly outnumber the neurons. They are 2 to 10 times more glia
than neurons in the vertebrate central nervous system. And this name glia actually in
Greek it means glue they do not glue as such, but they actually glue themselves with
each one of the individual nerve cell. And they surround the cell body the axon
sometimes the dendrites and they also greatly differ from the neurons morphologically
because they do not have things like dendrites or axons.
Now, glia also differs functionally because even though they arise from the same
embryonic precursor cells they do not have the same membrane properties like neurons.
177
For example, they are not electrically excitable and they are not directly involved in
electrical signaling, they are indirectly involved and there are many interesting variations
of that axons that we will see.
There are different variations of glial cells of course, for example, you have
oligodendrocytes; that is one type. We have oligodendrocytes and then Schwann cell and
then the astrocytes. Now, both oligo dendrites and the Schwann cell their role is in terms
of actually giving these myelin covers. And they are something you know which are part
of the micro glials.
In fact, there are some macro glials macro glials these are part of the macro glials all
three of them. Where there are some micro glials also I have not shown them here they
actually help the neural network system in terms of immunities etcetera.
Now, as you can see that the main role of both of these two types of macro glials is to
give this layer of myelin cover. So, you can see that they actually cover the axon. So, that
is very important because wherever the cover is not there you will have the nodes of
ranvier.
Now, this cover actually is you know it helps in terms of the action potential the
traveling of the action potential in terms of the speed it actually increases the speed you
know ah. Because it actually gives a coverage insulating cover of the axon layer so it
178
actually helps in terms of the action potential we will discuss about it in the mechanism
part. So, it helps the traveling speed in terms of increasing the traveling speed of the
action potential.
Now, there are also another group of glial cells called astrocytes they both actually keep
these things with the capillaries and as well as with the nodes of ranviers as you can see
from a typical neuron you know they keep contacts of both of them. And because they
have this star shapes that is why they have these astrocytes that is the name of this. So,
they also work in terms of giving supplying nutrition and also absorbing some of the
things from the neuron itself.
So, in terms of the role of this astrocytes glial cells, they of course, they help in terms of
separating the neural cells that is one. And thereby they can insulate the neuronal groups
and the synaptic connections, but also astrocytes are highly permeable to potassium ions.
So, they help to regulate the potassium concentration in the space between the neurons.
As you know that during the action potential travel potassium flow is controlled very
heavy, the potassium goes out of the axon, and astrocytes basically reserve this
potassium flow and then supply the potassium.
So, repetitive firing of these firing of this system they actually may create excess
extracellular potassium you know firing of the axons. And that is something which could
179
otherwise interfere with the signaling between cells and that is where the glial cells come
in terms of a rescuer because it absorbs this additional excess potassium that is one
important role.
They also help in important housekeeping chores that promote efficient signaling
between neurons. For example, they take up the neurotransmitters also from the synaptic
zones after release and sometimes they also release the growth factors. So, thus glial
cells they actually, they have a very much complementary role in terms of the neural
transmission.
We have seen the basic morphology of a neuron. Now we will try to understand that how
this neuron works in a complete system. Let us say the easiest of it is in terms of a reflex
system. So, we will try to see that how in a reflex system the afferent neurons. Let us say
you know there are pressure receptors in our hand so if I press it the afferent neurons are
going to actually take the signals towards the motor.
And in some cases you are going to see that there is a reflex action that is happening to
the system. So, we will try to explain the neural circuit in terms of what happens during
the reflex action. Let us look into that.
180
Let us take an example of a reflex cell. So, the sensory information you know if you just
let us say if you take this region just below that knee if you heat it if you give a stimulus
the sensory information is conveyed to the central nervous system ok.
So, that is something that will happen from the muscle spindle there is this muscle
spindle. And there is this particular muscle which is these quadriceps or the extensor
muscle that is the top part the extensor muscle. So, from here the signal will travel and it
will come to the spinal cord.
Now, once the signal is coming to the spinal cord there are two things that are going to
happen. One is that because of the reflex action there will be another, neural you know,
neuron which will actually it is the motor neuron so this is the flexor motor neuron.
So, as the sensory neurons are actually bringing the signal the flexor motor neuron will
actually, sorry, in this case the extensor motor neuron the flexor motor will come at just a
little later stage. The extensor motor neuron will take this signal back to the muscle
spindle so that it will be able to pull the system. I told you that it actually pulls and as the
pulls the leg goes outward so the leg moves outward.
Now, as this is happening then you have to also make sure that the hamstring is ready
about it otherwise the hamstring itself may actually oppose the motion then it would not.
So, as you are pulling the hamstring has to be kept ready that it should not you know
respond to that. So, that is where the you know the extensor motor they would actually
once they get this sensory neuron signal they would actually come into action. And the
flexor motors the flexor motors will actually keep this part reduced in order to assist this
motion.
So, once again we have to keep in mind that first of all the heating will generate signals
in the muscle spindle which will be captured by the sensory neuron and that signal will
come to the spinal cord. Once the signal comes to the spinal cord then two things will
happen one is that the signal will travel through motor neurons in this case the extensor
motor neuron back here which will actually pull this system. So, that pulling is going to
occur here as I have shown you earlier that it will be pulling.
And simultaneously a part of this sensing is going to a interneuron this is the inhibitory
interneuron. And this inhibitory interneuron is going to alert the flexor motor neuron
181
system which is which in turn will be working on the flexor itself to keep it ready in
order not to oppose this motion. So, this is the example of the reflex action that happens
in the knee and that is what we call a knee jerk reaction in a system.
Now, let us look into a small video in which you would be able to see that how this knee
jerking actually takes place.
Hyperreflexia can be demonstrated in the patellar reflex in this case the patient is
suffering a stroke affecting the left side of the brain ah. The patellar tendon extends from
below the patella which is right here it is a broad banded tissue it is easily palpable. If
you are not sure where it is having the patient extend their leg which will cause the
tendon to shorten and you can strike directly on the tendon.
And that is reflexes are very brisk I mean fact there are few extra beats of movement
which referred to as clonus, again classic for an upper motor neuron syndrome. In cases
like this reflux can be elicited by simply tapping on the tendon often. We see the same
reflex requires very little stimulus.
Here again with the normal side patellar tendon again extending from below the patella
strike on the tendon. So, that would be a normal reflex and certainly diminished prepared
with the certain hyperreflexia.
182
Now, in this entire you know reflex system we have seen also two interesting thing. One
is called a convergence and another is called a divergence of neurons. When the sensor
sensory neurons are actually, so the sensory neurons so when they are actually bringing
the signal. So, let us just you know call just denote them as sensory neurons.
When they are bringing the signals they actually put it into many neurons so this is
actually diverging. On the other hand, when the signal is after divergence when it is
coming through the motor neuron so; at that time there is a convergence of the signal that
is happening. So, this is the motor neuronal system. So, you will see that in any reflex
action there is both convergence and divergence of neurons that is taking place.
Now, if you look at it to even more in a minutely you will see that there are two different
types of circuits that are present here feedback and a feed forward circuit. Now, what is
the role of the feedback circuit? Well let us say for example, I told you that the extensor
muscles the afferent neurons are actually bringing the signal to the soma of the extensor
motor neuron that now you have to pull this extension muscle.
Now, what if it continuously starts to pull that it may actually harm itself. So, as it is
pulling the feedback is going and there is an inhibitory neuron which actually nullifies it
beyond a certain level it says enough this is to be crossed. So, this is what is the feedback
inhibition part which; actually judges up to what extent the excitation will be tolerated.
183
On the other hand, there are quite a few feed forward inhibitions also. For example, you
consider the same afferent neuron which is actually innervating the extensor muscles the
signal is coming. I told you that one part of the signal goes to the extensor motor neuron
the other part goes to the inhibitory interneuron. So, this is where it is going and this
actually gives an inhibitory signal. And that inhibitory signal now works on the flexor
motor.
So, the flexor motor this negative sign actually goes and it actually informs the flexor
that you have your role is secondary you have to help the extensor in terms of the reflex
action. So, this is a feed forward that is directly going to the system. Whereas, this is the
loop so this is a feedback system. So, you will see both this type of feedback and feed
forward circuits that are present in a simple neuronal circuit like reflex itself.
Now, there are actually many different types of neuronal systems. As you can see that
you have these model neurons that we have worked on, but there are the sensory neurons
which essentially have this it is like the bipolar neurons I talked about.
Then the motor neurons are there which directly works on the muscles. And then there
are interneuron which works between the neurons and then there are other types of
projection type of interneurons local interneurons. And also there are some
neuroendocrine cells also which actually works between the capillaries and the next level
of neurons.
184
So, for all of them one basic problem that you can simulate through this model neuron is
that; every one of them will be having some input, that input is coming like say some
sensory input it can come like that or it can come from the synaptic joint like that. So,
that is the input part then there is a integrative action that is happening inside the soma.
And then based on that there is a trigger and there is a conductive part that is what that
conduction is taking part in this axonal part.
And then finally there is an output, that output is in terms of certain chemicals that will
be transmitted and the new receptor is going to take it. So, a new receptor is going to
take this signal ok. So, there are there can be many such receptors there.
So, you can see that the signal is essentially electrical here, the signal is electrical in this
region also, but whenever it is coming here then it becomes actually a chemical signal.
And that is are these are the four the stages in which it happens; input, integration,
conduction and output.
Well, in this lecture I have given you an exposure to a neuron and how a neuron behaves.
Now, we will go much deeper in the next lecture in terms of how the neuron transmits
the signal. That is what we will be covering in the next lecture.
Thank you.
185
Lecture - 08
Good morning students. Welcome to the class of introduction to Cognitive Robotics. In

this lecture we will focus on the dynamics of a single neuron itself. So far, I have given
you the morphological description of a neuron that, what does a neuron consist of and what
are the different parts and what are the different types of neuron.
But now, the time has come for us to focus on the neuronal dynamics. Because only if we
understand the neuronal dynamics, we will be able to understand how a group of neurons
will work together. And then we will be able to understand that how these manifestations
of networking of neuron would actually affect the cognitive aspects of let us say, homo
sapiens or any such other you know living beings with group of neurons associated in
related in relation to their brain action.
So, it is extremely important for us, A. to understand the neuronal dynamic, so that we can
understand finally, its manifestation in terms of different cognitive aspects. It is also
important for us when we will be developing a cognitive robot in the other side in terms
of the development of the robot, we have to understand that similar to these neuronal
networks we have to develop some artificial neuronal network of kind of similar
complexity.
So, hence it is absolutely essential for us to understand the neuronal dynamics which is
what we will be trying in today’s lecture.
186
In this lecture, I will mostly focus on Hodgkin and Huxley model which is a classical
model of a dynamic model of a neuron. And based on the 1962 Nobel Prize of Hodgkin
and Huxley so, that we will focus the governing equations, the conductance versus gate
relationship, the gate variables. Then we will talk a little bit about neuron as oscillator and
the Winfree model, so that will be the flow of this lecture.
Now, before we talk about the H-H model itself. Let us first see that how the action
potential gets generated inside a neuron, because ultimately the dynamics of the neuron
will be governed by the dynamics of the action potential. How the action potential moves
with respect to different opening of the gates, that is what we will be ultimately matter. So,
let us look into that.
187
If you look at the excitation at the neuronal level and then what you would see is that. Let
us say, there are many many inputs that a neuron will be gathering from its dendrites. Now,
there are two types of dendrites there are some called basal dendrite and some called apical
dendrite.
Basal dendrites are directly related to the soma and apical dendrites basically come from
far away and through a channel that actually comes and sends the signal to the soma. But
right now, let us consider that there are n number of inputs that are coming to this dendrites
into the soma which is the cell body.
So, if we look at these procedure. So, you have these inputs are coming from all over ok.
So, signals are connected by neurons from the dendrites. That is the first part. And then
the soma basically it will work like an integrator. So, the cell body soma sums the incoming
signals both specially as well as temporally.
So, let us say that if there is one signal coming from this part, another signal coming from
this part then, inside the soma this will become a you know something which will be an
integration of these two signals.
So, it will be somewhat more than what was there. Of course, the signals will be little bit
of decaying as it will be coming from these sides towards this distance ok.
188
However, if the two signals are actually they are at the same time then there is a good
chance that they will be integrated these two dendrite signals they will be integrated here
and you will be getting.
So, in terms of special integration like that if there are n number of channels which are
contributing say one more channel is also giving. So, you will be getting an even higher
signal ok. So, that is how the signal integrations will happen at the soma now, when
sufficient input is received. In fact, a threshold of course, there is. Nowadays, people have
discovered that there is not a fixed threshold, but there is a threshold band.
But whenever these threshold around that region it is exceeded, then only the neuron will
generate an action potential or spike. This is also called firing of the neuron. So, this will
happen around this area ok. This is where the firing of the neuron will start and then that
action potential will start to travel. So, the action potential then is transmitted along the
axon that is how it is going and finally, it will go to other neuron.
So, this x 1’s will generate the action potential. It will come here and then from this
synaptic junctions it will go to all other neurals y 1 y 2’s are different outputs, that are
going to go to you know different n number of neurons. And if however, sufficient input
is not received, then the inputs will quickly decay and no action potential will be generated.
So, the timing is very important. The more input signals will arrive together, the higher the
chance that there will be a strong input which will come.
And the more number of the inputs will actually be integrating, there also is a higher
chance that this input is going to exceed the threshold level and hence the action firing is
going to take place.
189
Let us look into it with respect to a practical case. So, we have talked about if you
remember the reflex action. So, let us look into it in terms of the reflex action. If you
remember that in case of a reflex action, the nerves or the afferent nerves were actually
linked with the muscle spindle.
So, this is where is the muscle spindle. So, whenever we were actually giving a little bit of
heating, there is this stretching that is happening and this tweaking of the muscle spindle
is actually getting the you know the triggering in terms of the neuron and that triggering is
coming through the myelinated action to the cell body and from there further it is going to
the synaptic terminals.
Now, if you look at it that the initial stimulus of the signal the duration of heat is important
ok. So, if this is the initial signal, there will be a little bit of degradation of the signal and
finally, that initial signal will create if it crosses the threshold. So, because this is above
the threshold level the usual resting potential is about -60 milli volt and about -50 milli
volt you will get a threshold level. So, this will be around -50, -50 or so.
So, this varies, but roughly you know it is about -50. If it is above this then you will see
that there is a spike that is forming. And as many number of times this is happening, that
many number of spikes will be forming.
190
So, these spikes are formed then the spikes are traveling and at the output level this is
actually creating the neuro-transmitters. Now, if the signal continues with a higher value
and for a you know some kind of a high intensity values.
Then what can happen is that, again once again it will check that whether it is above the
threshold level or not. So, if it is above the threshold level then this spike intensity may
actually increase. So, the frequency of the spike not the intensity the frequency of the spike
generally the spikes actually the amplitude will not change ok. You can see here also it is
about 20 here also it is about 20. So, this is not going to change.
But between this signal and this signal, because this signal has input signal has a higher
intensity, what you may see that it will get reflected in terms of more number of action
potential spike. So, there will be more spikes there will be greater spikes in this case ok.
So, the spikes will be higher in this case and that would mean that the neuronal firing will
be actually higher in this case. So, that is what we have to greater spikes greater spikes
will be there in this case.
So, that is what we can see. In fact, the other case. Suppose, if the duration is longer in this
case the duration is longer. So, it is above the threshold level for a longer period of time.
And in that case, what will happen is that, now the number of spikes, the density of the
number of spikes will be something like you know that it will be not affected, but it will
be more in terms of the duration of the spike that total duration of the spike initially was
smaller here, it will be bigger.
And another thing also you can see that as the graded potential is degrading, you can see
that here also the density of the spike that is actually getting lighter and lighter. And
accordingly, you would see that the generation of you know the neurotransmission in
different cases will be different. So, finally, that will be the output of the system. So, this
is how typically a neuronal dynamic will actually happen. In this case it is for a reflex
action.
191
Now, there is one point that we have to always keep in your mind that in the beginning,
particular neuron will be always at its resting potential level. So, it will be at its resting
potential ok. So, wherever is this resting potential it will be at that level in the beginning.
In this case it is showing that this is the resting potential level usually it is up to -60
millivolt. So, that is where it will be. So, that is the resting potential.
Now, once the you know charges or the inputs are accumulating then what you will be
having is called the graded potential. So, basically this happens when lot of sodium’s are
actually you know rushing inside. So, there is a movement of sodium that happen. So, this
part in this part there is lot of sodium which will be actually sodium+ which will be actually
going inside the neuron ok.
So, basically what happens is that in a neuronal cell the inside part of the neuronal always
maintains itself with a negative potential. So, as the sodium will be going inside then, you
see that from the resting potential it will be actually going up. So, that is what it is
happening it is going towards the positive side.
Now, this is something that we actually call there is a technical term for it is called
depolarization. So, that is what is going to happen to the system.
So, as more and more you know sodium is going to go inside the cell there will be more
and more depolarization of the cells ok. So, that is what is the first result. That is why it is
192
crossing the threshold level and it is going to a high level. And as soon as this happens,
immediately the action potential will be released; that means, immediately there will be a
spike that will come up.
Now, once there is sufficient sodium then the sodium’s which are + right. So, they will
start to repel each other and hence they will spread inside the system and they will also
further reduce the inflow of the sodium. So, that will come down that is why you can see
that the potential is coming down. Simultaneously, there is a potassium movement that
will start to happen.
So, this k+ is going to leave the system this happens slowly ok. So, this will create the
hyperpolarization and that will create actually the further negative part of the signal. So,
depolarization will actually inspire the action potential generation whereas, hyper potential
will actually create you know negative of action potential.
It will not allow the generation of action potential we have to keep this in mind. That is
why in any common signal it will be the firing action potential will be something like a
spike followed by a slow negative side until it comes back to the resting potential.
So, that is the very common pattern of a signal that you will find in any neuronal signal.
Now, the other point we have to keep in your mind is that the more the receiving signals
are like 4 synapses 8 synapses 16 synapses you are going to see that these graded potential
is going to increase and that will create finally, the generation of the action potential.
However, as I told you earlier also that the magnitude of the action potential by and large
remains constant that does not change.
Only if the intensity is very high it may result on many high density action potentials ok.
And if the intensity is low it may result on few action potentials which are separated with
respect to time in a much lighter density ok. So, these are the common things that we have
to keep in our mind.
The other point is that during the hyperpolarization phase the cell generally would not, the
neural cell generally would not, initiate one more action spiking. Particularly there is one
region up to this region which is also called absolute you know part of the
hyperpolarization. So, this is the absolute part ok. So, it is absolute hyperpolarization.
There you know it will not allow any firing of the action potential at all.
193
But it is the later part this part which is known as the relative part of the hyperpolarization
this is where it will allow a firing if at all required it will allow the action potential to fire.
So, these are the points we have to keep in our mind in terms of the neuron action potential.
Now, that you know about that how an action potential gets generated in a neuron, let us
look into that what you know how we can mathematically define this action potential
generation in the system.
So, and to do that we will take ourselves to the Hodgkin’s-Huxley theorem which is the
classical you know theorem of explaining this neural dynamics. Later on, we will see that
this theory further gets modified, but let us start with the very popular fundamental theorem
which governs this.
So, we will now talk about the Discovery of the Nerve Impulse Flow from a Squid Giant
Axon that was done by the Hodgkin and Huxley in 1952. And as you can see that you
know they both Hodgkin and Huxley working on a giant squid.
So, these become like a legendary photograph that was the first time that the record of a
typical action potential of a neuron that has been done actually. Now, the reason why they
had chosen the giant squid action the action of a giant squid is basically, because this is
quite thick actually.
194
So, it was easier in those days to carry out the experiments and to see the action potential
ok. Now so, you can see here that typically how this action potential signal is. But you
have to of course, cross the threshold level in order to see this firing that is happening into
the system.
And you can see that there is a you know squid here and you can see that the nerves which
are associated, the nerves you can see here which are associated, with the giant you know
squid giant actions they call it. Now, this is used by the squids by the by for their escapes
for all emergency purpose.
So, let us look into it. Let us look into that, how generally the squids actually propane? As
you can see that this is how squid is generally propelling and at that time it does not use
much that giant action you know that is there for the squid.
195
So, this is how they are tracking. But you will see whenever emergency comes how fast
this squid is actually going to work.
So, if you continuously follow that, how is happening? And you can see that you know
this is the squid and suddenly you will see that this is the point where it has actually given
this kind of a signal and then it is happening at a very fast rate.
196
Now, let us look into the Hodgkin-Huxley model that has been developed based on the
experiment on the Giant Squid action. Now, what is the nerve speed that was there in this
particular case? Well this is about 30 meter per second. So, that is the kind of a nerve
speed.
And that means, it will be about 108 kilometer per hour. So, something like a very high
speed car or a very high speed train that you see that will be having this kind of a nerve
speed. Now, for your information, for human: muscle fiber control speed is something like
197
425 kilometers per hour almost like the air-crafts taking off speed. And the touch is about
274 kilometer per hour almost like a bullet train speed.
However, the pain signal that goes at a very slow speed something like 2.2 kilometer per
hour. So, this is just a reference that where we are in terms of the giant squid speed which
is not very high, not very slow.
Now, by fixing electrodes into the nerve cells they call it voltage clamp then, they have
actually you know they have been able to excite and they can see the spike which occurs
generally in the millisecond level thousandths of a second in the millisecond level this
reaction happens. You know it is corresponding to 30 meter per second.
So, the cell membranes are electrically polarized and difference of electric potential exists
as I told you in general and that is in the negative side. And that happens and the this
change also. So, these are all these all happens because of the presence of the ion channels.
So, there are three different types of ion channels: the sodium ion channel, potassium ion
channel and leakage channel and also there is a calcium channel.
So, these all about four, but in our model we consider three. Sodium, potassium and
leakage channel. The calcium we will not consider in our model now. So, what happens
as I told you that there is these sodium’s which will be rushing inside ok, and then it will
be followed by the potassium which will going out of the nerve cell. So, that would create
basically this kind of a positive and then followed by a negative signal.
So, there will be you know part A which is the rise of the potassium conductance, it should
be actually not potassium sorry it should be sodium conductance. So, that is one part. And
followed by B in which there will be the potassium conductance part of it ok.
198
So, let us look into the Hodgkin’s-Huxley model now. So, as we will be actually applying
the capacitive current Ic which is defined by the rate of change of charge q at the membrane
surface.
So, Ic is the rate of change that is dq/dt. Now, the charge q(t) is related to the instantaneous
membrane voltage Vm(t) and membrane capacitance Cm. So, you know that there is this
relationship that exists which is q= CmVm.
Now, the ionic current Iion, if I try to actually represent what is happening inside an action
potential in terms of a simple electrical circuit, then you can consider it to be something
which is a combination of you can say resistors of variable resistance. So, these are the
resistors and that is basically shown in terms of conductance. And also these are the
capacitors.
So, a combination of them. And as you can see that there are actually three of them that is
used in this model, that is the sodium, the potassium and the leakage. Generally, these
three that is used for this case.
Now, the ionic current Iion is subdivided into three distinct components, that is the sodium
current part corresponding to the sodium part of it. And then a potassium current part IK
and a small leakage current IL and which is primarily carried by the chloride ions. So, these
199
are the three currents that are actually streaming into the system. So, and that is the full
system along with the capacitance corresponding to each one them.
So, you can write a mathematical equation. And which is like the Cm. Cm is the membrane
capacitance. So, Cm is the membrane capacitance ok, and Vm is the voltage that you are
actually either you are applying the voltage or you are measuring the voltage membrane
voltage. And Iion that is the current that is actually passing through the ion circuit. So, that
is the Iion part of it. And then Iexternal, that is the externally applied current.
𝑑𝑉𝑚
𝐶𝑚 + 𝐼𝑖𝑜𝑛 = 𝐼𝑒𝑥𝑡
𝑑𝑡
So, in this model we are applying the current from outside. As a result of it, this will
actually generate Iion and also it will generate these Cm . dVm/d t and that is related to the
capacitance ok. So, these are the two things that this Iext is going to develop. Now, Iion itself
are actually can consist of three or four part of it. So, it depends on the conductance of
each one.
𝐼𝑖𝑜𝑛 = ∑ 𝐼𝑘 = ∑ 𝐺𝑘 (𝑉𝑚 − 𝐸𝑘 )
𝑘 𝑘
As I told you sodium, potassium, leakage like that. And then the difference so that
conductance time the difference between the voltage Vm of the membrane potential - the
200
equilibrium potential Ek. So, that will give us the ionic current each ionic current. So, that
is what we will be actually this equation will tell us.
Now, in the macroscopic conductance of the H-H model, this will be actually a combined
effect of many many you know microscopic ion channels ok.
So, it is not just one single ion channel. There are many ion channels which are coming
into it. Each individual ion channel can be thought of as containing one or more physical
gates that regulate the flow of ions through the channel. An individual gate can be in one
of the two steps, that is permissive or non-permissive. So, this is either it will allow the
thing the you know ion to move inside the neuron or it will not allow.
So, at some point in time, let us say pi(t) is the fraction of gates that are in the permissive
state; that means, it will allow the let us say the sodium to come inside. Then 1 − 𝑝𝑖 must
be in the non-permissive state. So, the probabilistic variation in terms of rate constant.
𝑑𝑝𝑖
= 𝛼𝑖 (𝑉)(1 − 𝑝𝑖 ) − 𝛽𝑖 (𝑉)𝑝𝑖
𝑑𝑡
Let us say 𝛼𝑖 (𝑉) is fraction which will go from non-permissive to permissive state. And
𝛽𝑖 (𝑉) is corresponding to that which will take it from non permissive to non-permissive
state. So, this is the dynamics that will continuously happen.
201
So, at any point of time; that means, the fraction of gates that are actually in the permissive
state: d pi/d t. It will depend on two things, one is that this weightage of 𝛼𝑖 (𝑉) how many
are actually contributing to it? And in a 1-pi stage ok. So, 𝛼𝑖 (𝑉)(1 − 𝑝𝑖 ). And then how
many are in the other way transmission, permissive to non-permissive. So, that is 𝛽𝑖 (𝑉)𝑝𝑖 .
So, this ratio of 𝛼 and 𝛽 is going to actually give me the d pi/d t, which is going to tell me
that what is the fraction of gates that are in the permissive state? So, a steady state value
of this equation will be something like pi(t) tends to infinity and that will be coming up if
𝑖 𝛼 (𝑉)
I actually put this as 0, you would see that it will be 𝛼 (𝑉)+𝛽 (𝑉)
.
𝑖 𝑖
𝛼𝑖 (𝑉)
𝑝𝑖,𝑡→∞ =
𝛼𝑖 (𝑉) + 𝛽𝑖 (𝑉)
1
𝜏𝑖 (𝑉) =
𝛼𝑖 (𝑉) + 𝛽𝑖 (𝑉)
𝑖 𝛼 (𝑉)
So, if you solve this you will see that pi would become 𝛼 (𝑉)+𝛽 (𝑉)
. And the corresponding
𝑖 𝑖
1
time constant is 𝛼𝑖 (𝑉)+𝛽𝑖 (𝑉)
.
Now, let us say the conductance of the gates. Let us call each one of the conductance of
the gates in terms of Gk. So, the conductance of the gates as Gk. So, it will be each
individual conductance and that product of this probability of the opening of the gates.
202
𝐺𝑘 = 𝑔̅𝑘 ∏ 𝑝𝑖
𝑖
So, it is gk times product of that pis ok. So, let us say we are talking about the sodium
channel then there is a gain there, GNa times pm3 times ph ok. We will call it as m3h ok.
And similarly, for the potassium channel this is gK pn4. We will call it g a potassium GK it
should be gK n4 and these are different type.
3
𝐺𝑁𝑎 = 𝑔̅𝑁𝑎 𝑝𝑚 𝑝ℎ = 𝑔̅𝑁𝑎 𝑚3 ℎ
𝐺𝐾 = 𝑔̅𝐾 𝑝𝑛4 = 𝑔̅𝐾 𝑛4
So, now I putting these conductance into the ionic equation, I can write that the sodium
part is gNa m3h ok. So, gNa m3h times Vm - ENa then, I can write the potassium part as gK n4
(Vm – EK) and then, the leakage part that is gL x (Vm - EL). So, that is how the conductance
are related to the gate relations.
̅ 𝑵𝒂 𝒎𝟑 𝒉(𝑽𝒎 − 𝑬𝑵𝒂 ) + 𝒈
𝑰𝒊𝒐𝒏 = 𝒈 ̅ 𝑲 𝒏𝟒 (𝑽𝒎 − 𝑬𝒌 ) + 𝒈𝑳 (𝑽𝒎 − 𝑬𝑳 )
Now, this m, n, h these are all gating variables. Let us look into it how this gating variables
itself changes, because they are the things that actually control the ionic current flow in
the neuron.
203
If you look into the gating variables, you would see that there are three gating variables
that we have considered in this model m, h and n and their values will be always between
0 and 1.
So, if m cube h is equal to 1, then the sodium channel is open and that conducts with the
characteristic conductance gNa. And if n4 equals to 1 then, the potassium channel is open
and that conducts with a characteristic conductance of gK. So, each one of them it is will
happen only when m3h is 1 then, the sodium channel will be opening or n4 is 1.
Then the potassium channel is going to open. And as you can see with respect to time. You
can see that for each of the cases like, this is the n4 and you can see that how this n4 with
respect to time how this will be changing and how this gK will be changing. So, you can.
In fact, draw a best-fit curve for each one of these conductance versus the time data.
𝐺𝑘 = 𝑔̅𝑘 𝑛𝑗 (𝑗 = 1 − 4)
So, that you would know how with respect to time the conductance will be changing or
the gate variables will be changing. This you can do for all the gate variables.
Now, based on all these things together you can actually write the final governing equation
which is corresponding to the H-H model. So, the core of it is this first order differential
204
equation ok, that is the capacitance times the d Vm/d t + Iion equals to the external current
that is actually used for excitation of the neuron.
𝑑𝑉𝑚
𝐶𝑚 + 𝐼𝑖𝑜𝑛 = 𝐼𝑒𝑥𝑡
𝑑𝑡
𝐼𝑖𝑜𝑛 = 𝑔̅𝑁𝑎 𝑚3 ℎ(𝑉𝑚 − 𝐸𝑁𝑎 ) + 𝑔̅𝐾 𝑛4 (𝑉𝑚 − 𝐸𝑘 ) + 𝑔𝐿 (𝑉𝑚 − 𝐸𝐿 )
𝑑𝑚
= 𝛼𝑚 (𝑉)(1 − 𝑚) − 𝛽𝑚 (𝑉)𝑚
𝑑𝑡
𝑑ℎ
= 𝛼ℎ (𝑉)(1 − ℎ) − 𝛽ℎ (𝑉)ℎ
𝑑𝑡
𝑑𝑛
= 𝛼𝑛 (𝑉)(1 − 𝑛) − 𝛽𝑛 (𝑉)𝑛
𝑑𝑡
Now, the Iion itself as I told you it will be having three parts in it. One part that the sodium
part which is governed by the m3h, which if it is equals to 1, there is a current the sodium
current, then there is a potassium current which is governed by n4, and then there is a
leakage current gL.
And each one of these how many part of fraction of actually the sodium gates will be
participating that is governed by this simple first order relationship that, d m/d t is
𝛼𝑚 (𝑉)(1 − 𝑚) − 𝛽𝑚 (𝑉)𝑚.
𝑑ℎ 𝑑𝑛
Similarly, = 𝛼ℎ (𝑉)(1 − ℎ) − 𝛽ℎ (𝑉)ℎ and = 𝛼𝑛 (𝑉)(1 − 𝑛) − 𝛽𝑛 (𝑉)𝑛. So, a series of first
𝑑𝑡 𝑑𝑡
order equations are to be solved along with the ionic current equation which is a non-linear
equation. In order to obtain that what will be let us say, the voltage of the membrane how
that will be changing with respect to time.
So far, I have talked of neuron as a dynamic model and how the voltage spikes the origin
of the voltage spike. How the voltage will be changing with respect to time as you are
applying an excitation current into the system and how you can explain it in terms of
different gating variables?
Now, we will see that instead of a time domain representation of a single firing, if a neuron
starts to fire continuously can I model that in terms of what you call an oscillator? Now, it
is known from the non-linear dynamics representation of a system that, whenever a system
actually starts from a quiescent level it is completely placid the neuron before the firing of
the action potential to kind of up you know oscillator level where the signals are
oscillating.
205
So, initially it is under a quiescent state then it starts to oscillate. This kind of a behavior
is called a bifurcation behavior. And there are different bifurcations model that can actually
explain.
So, let us look into the neuron as it fires action potentials with respect to time, how we can
model the neuron as an oscillator? Let us look into that. Then we will see how we can feed
different bifurcation models into it.
Now, here we are talking about a neuron as an oscillator. And as you can see here that this
particular signal this is actually the action potential the firing of the action or the action
spikes that is happening. So, that is the action potential with respect to time and you can
clearly see that is the depolarization phase. You can see and how the sodium currents
actually come down.
So, like you can write it as a depolarization phase ok. So, write it as depolarization slightly
for a part this is the depolarization. And then you can see that this is what is the
hyperpolarization phase, this part below the threshold. So, that is what is the
hyperpolarization and you can see how the spiking is happening.
Now, let us say that I would not look into it in terms of this is of course, with respect to
time. I am not looking to it with respect to time, but I look into it in terms of this m, h or n
206
which actually represents the different ionic currents. So, if you look at it that for example,
you know how the ionic currents are varying as the spiking is happening.
So, let us look into it from this point of view that here we have three axes, we have h, n
and we have the voltage. So, this is the voltage axis end. So, to begin with let us say, that
we have a low voltage and then suddenly the voltage is increasing once this threshold is
crossing. And as the threshold is crossing and then we will see that there is you know
initially there is a value of h that we are having ok.
And as the threshold has crossed, then we can see that the value of h is actually decreasing
and it has decreased to a low level. And you can see that the n has started to increase. Now,
if you remember it is the n which is related to the potassium part of it right. So, that is you
know once again increasing. So, this is what it is continuously happening in the system
that h decreasing, n increasing further h changing.
So, that is what continuously happen and this actually can be modeled in terms of an
oscillator that the states. If the if I consider the h and n to be two states then, the states
themselves are actually forming a circle in a phase diagram which can be depicted in terms
of an oscillator.
That is the basis of a neuronal oscillator model just for two channel two activation states.
Now, if you consider m, h and n then it will become a three dimensional space.
207
So, as I told you that this oscillator this site of oscillators can be actually modeled
mathematically in terms of three different types of bifurcations: A Saddle-node, Hopf
bifurcation and Homoclinic bifurcation. But out of them, mostly our discussions will be
for the first two cases that is the saddle-node and the Hopf bifurcation case.
Now, these bifurcations how do they look like the bifurcation diagram if you look at it.
Then you would be able to see that for the saddle node there is one node which is stable
and one which is unstable.
So, corresponding to the stable node. So, whenever there is this kind of you know the
quiescent mode is actually getting you know countered by the generation of the action
potential then suddenly, this frequency will start and in the stable node this will create a
spiking of the actional signals.
Now, there is also a possibility of two neurons joining together. In such cases what can
happen is that signal of one affecting the other may actually create a kind of a bifurcation,
in which you know you can see that it is initially at a stage and then it actually goes out
and in create a new limit cycle.
So, it is a much bigger limit cycle. So, this is where the oscillation will start. So, this is a
smaller limit cycle it is a small you know kind of oscillation, this is the bigger one as the
neurons are actually getting synchronized.
208
So, that is the Hopf bifurcation case. So, these are the two the saddle node and the Hopf
bifurcation which we will mostly see the for the neurons as an oscillator. The third one
which is a homoclinic bifurcation as you can see the bifurcation diagram here, that is
something that is not very natural for the neuronal oscillation system.
Let us now look into a very simple model of two neurons joining their hands and how a
neural oscillation pattern can form out of it.
209
Let us look into the network of the oscillators. Now, to understand that you have to
consider two neurons like the two neurons here. They can take inputs from the apical
dendrites as I told you that is far away from the soma or they can also take input directly
from the dendrites which are directly on the soma, these are called basal dendrites. And
based on the integrations they let us say the action will be fired and then you will be getting
a spike like this.
Now, there is a synaptic junction here, where the spike is coming and the neurotransmitters
are getting released. And that let us say it is taking by the basal dendrites of the second
neuron and also some other signals are coming from the apical dendrites based on that the
firing happens. So, then the two neurons are actually these are the two pyramidal neurons
actually find in found brain.
So, they start to correlate with each other; that means, as one is firing the other also starts
to fire.
Now, what happen is that. Initially these different neurons can have different frequencies.
But Norbert Wiener actually hypothesized that even if initially there are different neurons
which are in a cluster and that are operating at different frequency levels, but with respect
to time you will see that there is a pulling effect that is happening. You can see that say
two neurons here you know one is at a higher one is at a lower level and one neuron which
is around this level.
210
And you can see that these two are also pulled together and somewhere close to 10 Hertz
they are going to merge together. Now, whenever this en masse is happening; that means,
these oscillators are pulled together to come into a single frequency. This is what we will
say that the neurons are getting too are getting synchronized fully synchronized. Now, if a
neuron is actually having a frequency of firing too slow, it will try to speed it up or if it is
too fast, it will try to slow it down.
Such that, you know it will finally hit into a common platform of a particular frequency.
In this case Wiener have shown this with respect to the human alpha waves which is of
this kind of a low order between 8 to 10 Hertz.
Now, Winfree had proposed a probabilistic model based on this. And he said that this is
not only for the neurons, but also it is similar phenomena is found for lots of things like
chorusing crickets, flashing of fireflies and of course, the circadian pacemaker cells.
𝑁
𝜅
𝜃𝑖̇ = 𝜔𝑖 + ∑ 𝑃(𝜃𝑗 ) 𝑅(𝜃𝑖 )
𝑁
𝑗=1
So, these similar events happen. And to explain these, one can actually do it with respect
to the phases of each one of the individual oscillator. So, the first order relationship of the
phase, the rate of change of phase will depend on one frequency, which is the kind of the
211
frequency of that system along with you can see that there are some constants that
influence functions.
For example, P and the total number of oscillators, the coupling strength between the two
of these neurons and also the sensitivity function. So, using all these things there is a
coupling part that is going to dominate ok. So, and these two things, omega i and this part
together, they are going to define the rate of change of phase in a system. So, that is the
Winfree’s mean field model.
Now, Kuramoto the famous scientist from Kyoto University of Japan has actually
expanded this. So, Kuramoto put this Winfree’s intuition about phase models on a much
firmer foundation.
And he used the perturbative method of averaging to show that for any system of weakly
coupled, nearly identical limit cycle oscillators, there is long term dynamics which can be
given by the phase equation itself. And that is something along with the Chimera state
which you know actually make this pretty famous discovery.
212
Lecture - 09
Good morning students, welcome to the course on Foundations of Cognitive Robotics. In

the last class, I have told you about the mathematical modeling of a neuron, I told you that
how with the help of the classical Hodgkin-Huxley model we can actually mathematically
model the action potential of a neuron.
Now, today we will go beyond this, today we will talk about how we can have some of the
most recent theories to explain these kind of you know propagation of action potential. So,
let us look into that what will be the outline of today’s lecture.
Today, we are first going to talk about the Hodgkin-Huxley model summary. So, that you
can summarize and then we will talk about that, what are the properties which is still
unexplained by the Hodgkin-Huxley model and then we will talk about the development
of a new model.
We will then talk about the wave equations the development of solitons and the
significance of the bilipid layer membrane, piezoelectric modeling and a co-propagation
213
model. So, these are the things that we will discuss today. So, first of all let us try to
summarize the Huxley-Hodgkin’s model.
The HH model in a nutshell it starts with some of the key equations, the first equation that
I told you is that is the membrane current equation. So, here the neurons are considered as
you know kind of an assembly of capacitors and resistors. So, in that neuronal circuit and
also the cell itself as a kind of a source of potential or a battery.
So, the neuronal current which is actually the membrane current is 𝐼𝑚 (𝑡) and Cm is the
membrane capacitance and 𝑉𝑚 is the membrane voltage. Now, 𝑔𝑁𝑎 is the conductance with
respect to sodium channel, 𝑔𝐾 is the conductance with respect to potassium channel and
𝑔𝐿 takes care of the leakage current which is mainly with chloride and other ion channels.
𝑑𝑉𝑚
𝐼𝑚 (𝑡) = 𝐶𝑚 + 𝑔𝑁𝑎 (𝑉𝑚 − 𝐸𝑁𝑎 ) + 𝑔𝐾 (𝑉𝑚 − 𝐸𝐾 ) + 𝑔𝐿 (𝑉𝑚 − 𝐸𝐿 )
𝑑𝑡
𝑔𝑁𝑎 = 𝑔𝑁𝑎0 𝑚3 ℎ,
𝑔𝐾 = 𝑔𝐾0 𝑛4
So, one thing we have to keep in our mind here that we are representing the resistance in
terms of conductance g which is actually 1 over the R. So, it is the reciprocal of the
resistance that we have to keep in our mind, ok.
214
So, just like we write say for example, when in terms of current if we have to write then it
will help us, because we write usually current as what? Current we write as V/R and in
this case we will write it the current as what we will write it as gxV conductance times the
voltage. So, that is the difference that we have to keep in our mind.
And of course, the other part of the current that is related to the capacitor that is remaining
the same, because it is in relation to the capacitance and that is Cm dV/dt. Now, the other
important thing that I had discussed in the last class is that, this conductance like the
conductance of sodium, conductance of potassium, it is found experimentally by Huxley-
Hodgkin’s is that they actually vary with respect to certain gate opening and they found
out that there are essentially three types of gates, for sodium it is three types of m gate and
one type of h gate.
So, three m type of three type of m and one h gate; this actually controls the conductance
and for potassium this is actually four types of n gates which actually control the
conductance of the potassium.
Now, if there are three of the m gates and if each one of them is having a probability of m,
then it will be (m x m x m) which is the m3h, that is the probability of opening up of the
gates permitting the current to flow.
And similarly, in since there are four types of n gates. So, it is n into n into n into n. So, it
is n4. So, that is what is the gNa and gK. Now, m, h, n these are having their own rate
constant.
215
So, they have a simple kinetics which will be covered in the second equation, that is the
gate kinetics equation in which we will be talking about dm dt, that is the rate of change
of this gate type m, then dn/dt and dh/dt.
𝑑𝑚
= 𝛼𝑚 (𝑉)(1 − 𝑚) − 𝛽𝑚 (𝑉)𝑚,
𝑑𝑡
Now, I told you also earlier that this depends on two things, one is what one for let us say
for the first gate kinetics for the opening up of the m type of gates it will depend on
𝛼𝑚 which is of course, intrinsically a function of the voltage itself.
So, it is 𝛼𝑚 , where 𝛼𝑚 will be the typical rate constant corresponding to m and (1 − 𝑚).
So, (1 − 𝑚) is the probability that this n gates will be opened and then the opposite
probability that it will be closed will be related to 𝛽𝑚 (𝑉)𝑚, where 𝛽𝑚 also is intrinsically
a function of Vm.
𝑑ℎ
= 𝛼ℎ (𝑉)(1 − ℎ) − 𝛽ℎ (𝑉)ℎ,
𝑑𝑡
𝑑𝑛
= 𝛼𝑛 (𝑉)(1 − 𝑛) − 𝛽𝑛 (𝑉)𝑛
𝑑𝑡
216
So, similarly there for n gate you need 𝛼𝑛 , 𝛽𝑛 and similarly for the h gate you need 𝛼ℎ , 𝛽ℎ ,
which means in order to solve these you need to have the six sets of coefficients. So, these
are three pairs.
So, you need these six constants and these six constants are actually experimentally
determined. So, this is something we have to keep in our mind.
So, once we know these six constants, then we can go back to this equation and we can
find out that what is gNa, because then we know gNa is 0; m h we can find out and similarly,
we can find out gK and we can put it back to this equation in order to get the membrane
current. So, this is first of all the electrical part of the equation.
217
If we now look into the HH model cable equation, then you can see here that in the cable
equation we can actually keep the special variation of the membrane voltage. So, we can
equate the special rate of change of the special variation of the membrane voltage with the
membrane current by using the cable equation.
𝜕 2 𝑉𝑚 2𝑅𝑖
= 𝐼
𝜕𝑥 2 𝑎 𝑚
So, essentially what this cable equation does is that, let us say we have a soma and which
has some dendrites let us say and that is the soma and from that soma let us say we have
the action. So, let us say this is the action going and which will be ending to some synapse.
Now, this action is essentially modeled like a cable and the radius of this cable. So,
essentially this part it is this part which is modeled like a cable of some radius a. So, this
is the cable. So, let us see this is the cable and the radius of the cable is modeled as a, that
is one thing and the other thing here is this internal resistance. So, this internal resistance
is the resistance inside this cytoplasm of these axon.
𝜕 2 𝑉𝑚 2𝑅𝑖 𝐶𝑚 𝑑𝑉𝑚
= ( + 𝑔𝑁𝑎 (𝑉𝑚 − 𝐸𝑁𝑎 ) + 𝑔𝑘 (𝑉𝑚 − 𝐸𝑘 ))
𝜕𝑥 2 𝑎 𝑑𝑡
So, this internal resistance let us say that is denoted as Ri. So, if the internal resistance is
Ri and these cable radius is a, then the cable equation which essentially considers that there
218
is this continuous membrane which is continuously between different points of nodes of
ranvier, this can be actually segmented in terms of what you call a kind of a electronic
systems.
So, in terms of let us say some voltage and this resistance and the capacitance of the ith
cell. So, then if any one of this this system can be modeled separately and then you know
if you actually get the differential of it, then you will be coming to this particular equation
which will equate between the membrane current and the voltage of the membrane.
Now, membrane current already we know that the membrane current itself can be written
in terms of the current across the capacitance and then the across the sodium channels and
across the potassium channels.
The leakage part is not considered it is just neglected here, but you can improve that. So,
this is what you know if you substitute the Im from the earlier relationship; that means, that
we have done in equation 1.
So, if you substitute it, here you are going to get this complete relationship which is giving
you the relationship between the special rate of change of the voltage membrane voltage
with respect to the temporal rate of change of the membrane voltage.
219
So, the special versus temporal, but yet you can see here that, we still have the spatial
variable and the temporal variable here. Ideally speaking, we would actually like to get
equation which will be completely in terms of the temporal variable.
Now, that can be done of course, so that is what we will be talking about in the next
equation where we talk about the unified action for the action potential and here what we
do is that, we acknowledge the fact that the pulse is the action pulse can be actually
modeled in terms of a wave equation.
So, essentially if you remember that the action pulse if we try to model it is something like
this that there is this time t and then there is this membrane voltage which will be basically
starting say for example, from a negative state across the threshold go to a peak and then
it will again come down come down and then it goes back.
So, that is something like this and this is the part which you remember is the depolarization
part that is the depolarization part of it.
And, this is the part which is the repolarization part; that is the repolarization I am just
abbreviating it as RP and this is the part which is the hyperpolarization part of it right. So,
this kind of a signal now if you look at the action, so let us say this is what is my action
which is model like a cable and then, this action at different points if you look at it you
220
will see that this signal is actually the nature of the signal and also the amplitude is
remaining the same.
So, there are these loads of engaging between, which actually contain set of the intensity
losses, but this is how it happens and this is the way the signal will propagate and the
velocity of this wave velocity is what is theta here, ok.
So, then you know you can actually correlate between the voltage variation with respect
to type and the voltage variation V is special variation of the voltage.
So, once we have this relationship all we need to do is to replace the left side as where it
𝜕2 𝑉𝑚 𝜕2 𝑉𝑚
was by 1/ 𝜃 2 . So, this part of it and also from the left hand side of the earlier
𝜕𝑥 2 𝜕𝑡 2
equation we have just taken that 2 Ri by a, and we brought it in the left hand side.
𝜕 2 𝑉𝑚 2
𝜕 2 𝑉𝑚
=𝜃
𝜕𝑡 2 𝜕𝑥 2
𝑎 𝜕 2 𝑉𝑚 𝑑𝑉𝑚
( ) = 𝐶𝑚 ( ) + 𝑔𝑁𝑎 (𝑉𝑚 − 𝐸𝑁𝑎 ) + 𝑔𝐾 (𝑉𝑚 − 𝐸𝐾 )
2𝑅𝑖 𝜃 2 𝜕𝑡 2 𝑑𝑡
𝑎 𝜕2 𝑉𝑚 𝑑𝑉𝑚
So, it becomes 2𝑅 𝜃2 ( ) which equals to these current part of it that is 𝐶𝑚 ( ) then
𝑖 𝜕𝑡 2 𝑑𝑡
the conductance related that sodium and the potassium current.
So, this is what you will get the final cable equation, by solving this as you can see now
that this equation, it has been possible to bring everything with respect to the time domain
variation.
Of course, it is a non-linear equation because the gNa itself depends on m and m itself has
a rate dependence etcetera, but you can still get the entire equation with respect to the
temporal variation and that is a good point. So, it is easier to solve this kind of equation in
order to get the action potential of the system.
221
We now we will talk about the properties, which are actually unexplained by the HH
model. For example, we know that as the action potential actually crosses the action, there
is a change in the thickness and in the change in the length variation; in the length variation,
under the influence of the action potential.
Also, we know that the mechanical stimulus can actually actuate the action potential, how
whereas, the electrical model cannot explain that.
Also, during the first phase of the nerve pulse, heat is released from the membrane and it
is reabsorbed during the second phase. Now, if these entire you know neuronal system is
like a resistor then it will only dissipate the energy it will not you know release the heat
and reabsorb it just like an adiabatic and reversible phenomena. So, that is something that
also is not possible.
So, these are the some of the important things, which are not explained by the HH model
and hence this drives us to think of a better model towards explaining these facts as well.
222
Now, when we will talk about the development of a new model consider first of all that it
is based on the wave equation, because we already know that the wave equation is there
in place.
However, in this case the wave velocity is c that is fine, and the x that we had earlier
discussed is actually z here. So, z or x they are similar. So, this is the longitudinal direction
of the action that we are talking about; so that is what, is the z direction. Please keep in
mind just the symbols vary, but that part is the same.
Whereas, now these variation actually depends on something called ∆𝜌 𝐴 and what is this
∆𝜌 𝐴 ? That is the change in the lateral density of the membrane. This is something that is
experimentally observed, that the membrane density that is if you consider that the neuron
itself is like a membrane there are this that is the up some thickness, ok.
And, then I also told you that this membrane changes in the thickness direction as the
action wave propagate. Now, let us say that that change is creating a change in terms of
the variation of the area density and that variation of area density is ∆𝜌 𝐴 , then the wave
𝜕2
equation I can write in a different way here now as 𝜕𝜏2 ∆𝜌 𝐴 and that is if provided if this
c2 does not vary, of course it varies along the length because of some other chemical
𝜕
changes inside the system, but it is essentially 𝑐 2 𝜕𝑧 ∆𝜌 𝐴 .
223
𝜕2 𝐴
𝜕 2 𝜕
2
∆𝜌 = (𝑐 ∆𝜌 𝐴 )
𝜕𝜏 𝜕𝑧 𝜕𝑧
So, you look at it carefully that in the last an expression, what was it? It was with respect
to the membrane voltage right. So, the last equation was the wave equation was with
respect to the membrane voltage.
Now, we say no, we are going to go to something more fundamental and we are going to
look into it in terms of the membrane you know change of the area density of the membrane
and in terms of that we are defining the wave equation that is the change a we have to keep
in our mind.
𝑐 2 = 𝑐02 + 𝑝∆𝜌 𝐴 + 𝑞(∆𝜌 𝐴 )2 + ⋯
𝜕2 𝐴
𝜕 2 𝐴 𝐴 )2
𝜕 𝜕4
∆𝜌 = ((𝑐0 + 𝑝∆𝜌 + 𝑞(∆𝜌 + ⋯ ) ∆𝜌 ) − ℎ 4 ∆𝜌 𝐴
𝐴
𝜕𝜏 2 𝜕𝑧 𝜕𝑧 𝜕𝑧
Secondly, in the last equation we said that 𝜃 is not varying with respect to the length, but
now we are telling no, this c is actually varying with respect to z, it is not constant.
In fact, the c itself has a kind of a dispersion relationship; that means, c2 is something like
𝑐02 + it so, it is a; it varies with the density and some constant of it and it is a non-linear
variation with respect to the area density.
So, if I put our model like this, then I can actually put this expression of c in this entire
thing. So, we get this final model and what we will see is that, in this final model because
of the dispersion we will get an additional dispersion term here.
This is very very important ok, that is we get in addition to that the c is variation because
of the dispersion we get an extra variation here and this extra variation will be very
important we will look into it so.
224
Now, certain non-dimensionalization let us do for the model; which means, that for the
you know ∆𝜌 𝐴 we can actually with respect to the equilibrium lateral density, let us have
a variable u.
∆𝜌 𝐴 𝑐0 𝑐02 𝜌0 𝜌02
𝑢= , 𝑥 = 𝑧, 𝑡 = 𝜏, 𝐵1 = 𝑝, 𝐵2 = 𝑞
𝜌𝑜𝐴 ℎ √ℎ 𝑐02 𝑐02
Then, let us have x as a non-dimensional variable now, with respect to this thickness and
then let us have the time itself as another non-dimensional variable and we have these two
constants B 1 and B 2. So, this is the thing that will require to non-dimensionalize the last
equation.
So, if I do that and here, the h is a parameter which describes the frequency dependence of
the speed of sound ok. So, if I do that we are going to get that last equation in a much more
neat and clean form, which is that the non-dimensional variation of area density is now u.
𝜕2 𝑢 𝜕
So, that is will be 𝜕𝑥 of these velocity itself is with respect to a new non-dimensional
𝜕𝑡 2
quantity B.
𝜕 2𝑢 𝜕 𝜕𝑢 𝜕 4 𝑢
= (𝐵(𝑢)) −
𝜕𝑡 2 𝜕𝑥 𝜕𝑥 𝜕𝑥 4
𝐵(𝑢) = 1 + 𝐵1 𝑢 + 𝐵2 𝑢2
225
𝜕 𝜕𝑢
So, it is like (𝐵(𝑢)) and then that h factor is accommodated here in we get this
𝜕𝑥 𝜕𝑥
𝜕4 𝑢
relationship the dispersion part of is as − 𝜕𝑥 4 and the B(u) you can take up to two terms as
1 + 𝐵1𝑢 + 𝐵2 𝑢2. So, this is how we can actually get this whole expression in a non-
dimensionalized mode.
Now, if I get this whole expression in a non-dimensionalized mode the only thing now is
that, we can actually you know apply another parameterization in terms of the density
pulse we can use a coordinate transformation now and we can put a new coordinate 𝜉 =
𝑥 − 𝛽𝑡, where 𝛽 will be the dimensionless propagation of velocity.
𝜕2 𝑢
So, we can get the new equation in terms of not in terms of x, but in terms of 𝛽 2 𝜕𝜉2 =
𝜕 𝜕𝑢 𝜕4 𝑢
(𝐵(𝑢) ) − 𝜕𝜉4 , which is what is the dispersion part again.
𝜕𝜉 𝜕𝜉
We have to keep in your mind; that this dispersion if it is not there, then the system will
be a conservative system and it will not give you the relationship which is very important
for us in terms of the propagation of the wave.
2𝑎+ 𝑎−
𝑢(𝜉) =
(𝑎+ + 𝑎− ) + (𝑎+ − 𝑎− )cosh(𝜉√1 − 𝛽 2 )
226
So, the analytical solution of this density propagation if you solve this equation you will
get it, in this 𝑢(𝜉) in terms of two constants-a, one 𝑎+ 𝑎𝑛𝑑 𝑎− where 𝑎+ 𝑎𝑛𝑑 𝑎− can be
expressed in terms of the B 1 B 2’s 𝛽, 𝛽0 𝑠.
𝐵1 𝛽 2 − 𝛽02
𝑎± = − (1 ± √ ) , 𝛽0 < |𝛽| < 1
𝐵2 1 − 𝛽02
So, we will get it and the 𝛽 itself please keep in mind that it is a non-dimensional velocity,
which generally varies between 0.6 to 1.
Now, we will see something very interesting in terms of the behavior of the 𝛽, we will see
that this solution actually takes us to a very well-known wave solution which is in terms
of solitons.
Well, the solution in that last equation will actually give you what I was just telling you
that is soliton, ok. And, what is the soliton? It is a self-reinforcing solitary wave packet
and which will maintain its shape, remember the action potential maintaining same. While
it propagates at a constant velocity imagine that, this is what is the direction of the, action
potential.
227
So, there is a soliton that is propagating if that wave equation if we solve that will give us
a soliton and these are caused how by a cancellation of non-linear and dispersive effects
in the medium.
That is why I said that the dispersive effect is important because it is going to cancel out
the non-linear part. Now, solitons are the solutions of a wide spread class of weekly non-
linear dispersive PDEs describing the physical system.
As far as in 1834, John Scott Russell observed first a solitary wave in a canal in Scotland.
And, it is only late and you know the researchers like Heimburgs, Jacksons they have
found out that you know in terms of the neuronal voltage such a solution actually matters.
Now, let us look into that, how does a soliton actually behaves?
I will take you through a small animation from University of Tasmania and of its Scott
Forrest, I will show you the video which will explain you beautifully that, how does a
soliton propagates? Let us look into the soliton wave. .
228
229
230
231
232
233
234
235
Now, of course, there is one point that, what happens when two solitons will collide from
the opposite direction? Then, in such a case it actually any you know this there is a
blocking of heat that you can see that as this is what is the before the collision these two
waves are approaching you can see.
This is something that is still is to be actually checked with respect to the normal pulse and
you can see what is happening you know if post collision and that is what is the post
collision part of it. So, you cannot see either of the solitons here.
Now, if you use on the other hand a simpler form of collision which is a Hodgkin-Huxley
equation, which is Fitzhugh-Nagumo equation. Then, two pulses traveling in opposite
directions before and after you can see that you may see that the pulses are going to
annihilate after the collision.
Whereas, for these kind of a system this kind of annihilation does not actually happen,
they can actually pass through each other.
236
So, that is something that is still you know not explainable through the soliton model. So,
you can see that this is what is happening for a soliton that two solitons are approaching
and you can see that after they are collision; they are actually with a lesser much lesser
amplitude, but they are just going back to both the directions.
So, this is something that does not happen in this particular case whereas, in the case of
the you know simpler version of the HH model that is the Hodgkin that is the Fitzhugh-
Nagumo model you can see that there is a complete annihilation that is happening. So, the
point is that these Hodgkin-Huxley model predicts a complete annihilation of two waves
whereas, for soliton collision this does not happening.
Now, in reality; that means, the action you know a potentials in terms of waves if it reflects
back from the synapse, then it will once again pass through without this you know this
proceeding directions the propagation direction of the action wave we and both of them
will come down, but there will be no annihilation.
On the other hand, in the last model we have seen that there will be an annihilation. So,
that disparity is still existing in the system.
237
Does soliton model support anesthesia? Well, that is another point that, it is known that
well known that general anesthetics actually lowers the melting point of the lipid
membrane and you know that the neuronal membranes are actually of lipid membrane.
So, reduction of melting point because of these drugs or anesthetics; that means, that there
has to be a increase of the required pressure to generate the density wave. And, that means
that, you know these density waves will not be able to get generated so easily. So, in the
presence of anesthesia, the free energy requirement will increase which will inhibit the
soliton formation and that explains that why an anesthesia would work.
In fact, there is an experiment that Heimburg’s group have carried out, where the ambient
pressure level of a of tadpole which is under anesthesia that is increased to about 50 bar
pressure and they found that the effect of anesthesia is overcome and these tadpole is able
to move again. So, thus the soliton model provides a mechanism for general anesthesia
that is something that happens in terms of the soliton model.
Let us now carry out a brief summary between the Hodgkin-Huxley model and the soliton
model and let us try to see that where we are with respect to both the models, as I told you
that both the models are partially successful in explaining something and in not explaining
something. So, which may actually generate further scope of research in this field.
238
In the HH model, we can see that the action potential is based on the electrical cable theory
in which the pulse is the consequence of voltage and time dependent changes of the
conductance of sodium and potassium.
So, it is purely electrical in nature. The model is consistent with quantized ion currents that
is proved. It is consistent with the channel-blocking effects also of several poisons, such
as tetrodotoxin.
This model is based on ion currents through resistors that is a problem because it is
therefore dissipative in nature. The reversible changes in heat and mechanical changes are
not explicitly addressed, but heat generation would be expected, but why the heat is
reabsorbed that is not explained. And, the model generates a refractory period.
On the other hand for the soliton model, which is a newer model the nerve impulse is
considered to be an outcome of electromechanical soliton wave that is coupled to the lipid
transition in the membrane.
And, the solitary character is considered to be a consequence of the non-linearity of the

elastic constants. It of course, does not contain an explicit role of poison’s and protein ion
channel. So, that is something that is to be checked state.
239
And, the theory is consistent with channel-like pore formation in lipid membranes that part
is consistent. And, the temperature part and it is it does not dissipate heat, but it actually
shows the whole model as an adiabatic process.
So, that is something that happens of course, it still does not explain, the change in terms
of the thickness for which possibly we need a more refined model.
In order to look into a more refined model, we have to first know that the bilayer lipid
membrane that is you know basically they are in the neuronal walls. So, if you consider
the neuronal wall each wall of this neuronal wall is actually consists of bilayer lipid
membranes.
Now, these bilayer lipid membranes they are known to be piezoelectric in nature. Now,
the ionic motion whether the piezoelectricity is happening due to electro-active polymer
or due to ionic motion that is something we need to still focus on. I have already told you
about the characteristics of electronic EAPs and ionic EAPs when we have discussed about
this kind of things in terms of active muscles.
So, in electronic EAP like dielectric EAPs, electrostrictive papers, ferroelectric polymers
or liquid crystal elastomers these are all electronic EAPs and they are capable of actually
these kind of motions, but it is slightly slower.
240
On the other hand ionic EAPs like ionic polymer gels, IPMCs, conductive polymers or
CNTs they are actually faster. So, they are faster and these are slightly slower. So, because
this nerve impulse is happening in the millisecond region so, this is millisecond region.
So, that kind of tells that it could be because of these thing.
So, well, this I need to change. So, this is ionic EAP. So, this part is slower. So, ionic EAPs
are slower and this is faster the electronic EAPs this is faster. So, because of these you
know millisecond region this actually sort of tells us that maybe this phenomenon is
because of the electronic EAPs, which are actually faster in nature. It is also possible that
this kind of a thing is actually happening because of an electro-thermal mechanical system.
If you consider the bi lipid layer people, have shown through experiments that the lipid
layer can be actually modeled in terms of piezoelectric system. So, then it will be you
know electro thermo mechanical system because any piezoelectric system consists of three
things together. So, there is this piezoelectric effect and there is this pyroelectric effect and
there is the mechanical effect.
So, that is why you know there is the temperature comes in the pyroelectric effect and the
voltage comes in the piezoelectric effect and the mechanical force occur. So, all the three
things together can be actually observed in this type of phospholipid layers. So, that may
actually verify the piezoelectric nature of the liquid crystal.
241
Now, this kind of piezoelectric nature will also show that there is a frequency dependent
and a temperature dependent variation. As you can see here that, the electric polarization
P is actually vary with respect to the frequency of propagation and with respect to
temperature both.
So, this is something like a bilayer system lipid system with a potassium channel protein
and you can see the lipid layer and you can see the potassium channel proteins. So, this is
something that, these is still to be explained that how this entire dynamics of the movement
242
of the potassium through the bilayer you know lipid layer happens with the help of the
piezoelectricity.
Now, as I told you about the piezoelectricity that the electronic polarization in
piezoelectricity is very fast in terms of the frequency 1012 hertz, ionic is about 109 hertz,
dipolar is 106 and space charge is even slower 103hertz or 1000 hertz or in the kilo hertz
level.
So, accordingly the response time is going to be varying the response time will be more
here. So, this will be more here the response time and the response time will be very very
less here.
Now, depending on what type of a polarization is happening in the lipid layer we will be
able to actually say that, what type of polarization is happening in the lipid layer.
243
So, if we consider an action to have this kind of you know kind of a patches of piezoelectric
patch then, what will that motion predict? So, here is an equation of motion as you can see
which is different from the earlier two equations that we have seen because, now we have
the longitudinal displacement 𝑈(𝑥) the spatial variation of it with respect to the wave
number and we can get the equation and this longitudinal displacement and similarly you
can also get for thickness wise displacements, but these changes are happening because of
the piezoelectricity that is there in the piezoelectric patches.
𝜕 2 𝑈(𝑥)
+ 𝑘𝑖2 𝑈(𝑥) = 0 ; … . 𝑖 = 𝐴, 𝐵
𝜕𝑥 2
So, in every piezoelectric patch you will see this kind of a change that is happening in the
system and that can also generate a kind of a pulse in the system.
244
𝑇
𝑌𝑘𝑟 = [𝑈𝑘𝑟 ⋮ 𝑁𝑘𝑟 ]
(𝐵) (𝐴)
𝑇𝑘 = 𝑇𝑘 𝑇𝑘
sin(𝑘𝑖 𝐿𝑖 )
(𝑖) cos(𝑘𝑖 𝐿𝑖 )
𝑇𝑘 [ 𝑧𝑖 𝜔 ] … 𝑖 = 𝐴, 𝐵
−𝑧𝑖 𝜔 sin(𝑘𝑖 𝐿𝑖 ) cos(𝑘𝑖 𝐿𝑖 )
𝑧𝐵 = √(𝜌𝐴 𝐴𝐴 + 𝜌𝑃 𝐴𝑃 )(𝐸𝐴 𝐴𝐴 + 𝐸̅𝑗𝑗𝑆𝑈 𝐴𝑃 )
𝑘𝐵 = √(𝜌𝐴 𝐴𝐴 + 𝜌𝑃 𝐴𝑃 )/(𝐸𝐴 𝐴𝐴 + 𝐸̅𝑗𝑗𝑆𝑈 𝐴𝑃 )
So, in order to solve this wave equation, I can actually use a traditional transfer function
matrix. So, where you actually put all the longitudinal displacements in terms of a state
vector and you can develop the transfer function and you can find out the impedance and
the wave constants and you can actually solve this in terms of finding out the wave
propagation of the system.
245
The interesting thing that you will note from these is that, there will be in such a system
invariably certain bands where the wave constants will be actually real in nature whereas,
certain bands where it will be imaginary in nature. So, these are the plot of the real and the
imaginary parts.
Now, real part of the propagation constant actually is telling that there is a kind of a pass
band. So, this is a pass band and this is a pass band. So, there are two pass bands, there is
a very small pass band here.
Whereas, wherever the imaginary constant will come into picture then you may say that
there are these stop bands which will be there. So, the creation of the pass band and the
stop band is actually telling us that, only certain frequencies are allowed through these
channels and certain frequencies are not allowed through these channels.
Now, based on this piezoelectric model and our earlier knowledge there is a new model
which is coming up which is called a co-propagation model.
246
I will just very briefly for your reference I will mention this model to you. In the case of a
co-propagation model which is a something like you know an advanced version where the
piezoelectricity is considered, you can see that it is considering a minimal mechanical
model of the action, which has an elastic and a dielectric cube filled with viscous fluid.
And, the action potential as it is passing, it changes the charge separation across the
dielectric membrane, because the thickness change and that is altering the membranes
geometry and that will create the electrical voltage.
So, there is a co-propagation of displacement, because that voltage will again change the
thickness. So, there is a co-propagation of displacement wave along with the electrical
pulse. So, this theory is telling that there will be let us say with respect to time, there will
be a voltage change, there will be a voltage pulse.
So, there will be a voltage pulse and also with respect to time as well as with respect to
you know the direction of propagation there will be something like a displacement change.
So, that also will be happening and they will be actually co-propagating both the models.
So, this is checked with the Garfish Olfactory Nerve, that there is this voltage propagation
and the change in the membrane displacement, both are happening and you can also see
that there is a mechanical heat that is happening and there is a change in the temperature
increase and decrease.
247
Similarly, the same thing is obtained from Squid Giant Axon, Hippocampal Neuron in all
the cases they see that there is a membrane displacement and that is happening and there
is a lateral displacement that is happening in the system.
I will now show you a typical simulation where you can see that how this wave can
propagate, imagine this is what, is your action.
So, you can see first that how this radial propagation of wave is happening it is starting
from one point how it is happening.
248
And, similarly you can see that how in axial direction also the wave is propagating, this is
where is axially it is propagating. So, thus it is actually possible to propagate the wave
both in the radial and in the axial direction under two different frequency bands.
Now, this is where we will put an end. In the next lecture, we will talk about different types
of you know the ways experimentally how we can actually obtain the electrical signals
from the brain, which is very important from the human robot interaction point of view.
Thank you.
249
Lecture - 10
Good morning, students. Welcome to the course on Fundamentals of Cognitive Robotics

and today, we are going to talk about a very important part of cognitive robotics
experiment. We are going to talk about EEG experiments, because in order to you know
what is there in a human brain from a robotics perspective, in order to develop a brain
computer interfacing it is absolutely important that we understand how to know what is
happening in the human brain.
That is what we will be discussing today that how to carry out EEG analysis and through
that how to know about the state of a human brain. Let us look into that.
I have talked about some of the reference books which will be useful now particularly the
first and the third book that is the Neuroscience book which is edited by Dale Purves and
the Control Systems: Classical, Modern and AI-Based Approaches these are the things
which will be important at this stage.
So, you can catch hold of these books and you can also try to get reference materials related
to the lecture.
250
Now, in today’s lecture we will talk about the history of electroencephalography or EEG
a little bit of it. We will also give a kind of a introduction to the EEG analysis, various
components or elements of the EEG analysis, then we will talk about the EEG waveform,
EEG results – analysis and interpretation, signal filters that are used in EEGs, how to use
you know some of these filters specially for artifacts like eye blink removal.
And, one of the most important thing against external stimulus how to use the event related
potential information and finally, I would also talk about the advantages and disadvantages
of the risks of the EEG.
There are various ways in which we can find out the state of a brain. So, EEG is one of the
way in which we insert electrodes over the scalp of a brain. So, it does not penetrate the
body, it remains over the scalp of the brain that is one way and then there are other ways.
For example, there are ways which are related to MRI and specifically called FMRI which
actually finds out that as the blood flows inside the brain and as this blood is getting
oxygenated, you can say that somewhat that those are the regions where more oxygen
supply is happening you can say indirectly that nerves are more active in those areas.
So, if there are more presence of blood flow then naturally because blood contains a good
amount of iron particle in the form of haemoglobin. So, the magnetic resonance you know
251
would be able to catch it and then from that we will be able to get a kind of a pattern that
is happening inside.
However, these techniques of MRIs and FMRIs they have a very good spatial resolution
because many parts of the brain what is happening that they can capture, but they do not
have a good temporal resolution; that means, over time they are actually very slow, on the
other hand EEG is very fast. It can very quickly say what is happening in the brain.
So, in fact, today a mixing of the two technologies it seems to be a much better way of
actually analyzing what is happening in the brain. So, before looking into the EEG let us
first very quickly look into all the other available technologies.
Now, there are various methods as I told you for observing the living brain and in terms
of time those which are slow will be in the right hand side of it and in terms of time those
which are fast they will be in the left hand side. Say for example, EEG and MEG is there
in the left hand side because they are very fast phenomena whereas, FMRI is relatively
slow and the PET technology or other studies of lesion etcetera these are very slow.
Now, in terms of the size that it covers because the brain is a large area. So, there are
actually tests like patch clamp tests which are in the dendrite level and then a single
neuronal unit also can be tested once again by placing the electrodes directly over the
neuron and at a higher level there are these optical dyes and of course, when the entire
252
brain is concerned then you can actually make a network of EEGs or MEGs and through
that you can actually check it.
There are other technologies which are just coming up which is like TMS which is
somewhere in between the EEG and FMRI technology. So, this actually gives us a global
picture that where these EEGs are and we can say that EEGs are a process of analysis
which is very fast millisecond level, but it works in a global scale on the full brain. So, that
is something that we have to keep in our mind.
Now, I will talk about a little bit about the history of EEGs. In 1875, Sir Richard Caton of
Liverpool is credited that is the first person to actually apply the EEG technique the
electrical phenomena of the exposed optical lobes of dogs and monkeys. So, essentially he
had literally opened up the dogs or monkeys brains and then on the optical lobes he has
actually inserted the electrode and then he has found out that yes, there is the electrical
pulse that are associated with it.
Later on, Sir Adolf Beck he has actually find out that in rabbits and dogs there is this
spontaneous electrical activity of the brain that is happening. And, even later in 1920 Dr.
Hans Berger in fact, we have the one of the waves of brainwaves alpha wave is named
after Berger’s wave and he is credited to be the first person to actually talked about
something like an EEG because so far in the first two cases these two cases it was actually
open brain surgery,
253
during that time it was tested, but during doctor Hans Berger experiments he was first able
to record EEG tracings from the human scalp itself and he discovered waves at a low
frequency like 10 Hertz in fact, alpha waves is what he has looked at. So, this is something
that actually shows the original wave that he has in fact, looked into and you can see this
is an alpha wave which is a very slow paced wave.
In fact, he carried out this experiment on his own son showing the alpha activity. So, that
was the first time you know the EEG is kind of it was proved that EEG can actually pick
up some interesting signals from the brain and later on in 1935 forester and Alteberger
they have used these for intra operative EEG which means during surgery they have used
this similar way of taking the signal so that they can monitor the patient state during
surgery.
So, that is a brief history about the EEG.
Now, let us introduce that what is this EEG test all about. Well, EEG test is all about a
medical device development for analyzing the electrical activity of the brain. The
frequency range is between 0.1 to 100 Hertz and the amplitude of the voltage is quite small
it is about 2 to 200 micron volt.
The general mechanism by which it works is that it picks up the summative charges of the
electric potentials using either passive or active electrodes. Now, the difference between
254
the passive and the active electrode is that passive electro electrodes are generally not fitted
with any amplifier. But, because the signals are very low sometimes it actually microvolt
range it make sense to have an amplifier on the electrode itself. So, then it becomes actually
active electrodes.
Now, both passive and active electrodes can be actually divided to further sub classes
which are known as dry electrode. This will be dry electrode and wet electrode. Well, the
difference is in terms of that in case of a wet electrode we actually apply some kind of a
silver chloride gel on the brain and then apply these transducers.
But, for dry we do not do it and generally dry electrodes are used for kids who are very
you know dynamic. So, wet actually creates disturbance for them, so that you do not get
the desired brain picture. So, that is why dry is generally preferred for such cases, but it is
not so accurate as the wet electrode.
Now, you may remember that I said earlier that the neurons are negative when they are at
rest, is something like -50 to -60 millivolt potential, but the moment a neuron is firing it
becomes positive and if there are group of neurons firing then there is a lot of positive
charges developing.
And, the EEG record this summation of charges by the electrodes and these signals are
transferred from the electrodes to the amplifiers because the signals as I told you is very
very small and then their process signals are then shown on the screen, sometimes it is
recorded also. So, this is what is the basic way of carrying out the EEG.
255
Now, the source of the EEG if you look at it very carefully you see that there are so many
neurons here and thus is deep inside the brain and as they are getting charged as you can
see here that the active synapses are here then these charge are actually coming and of
course, on their way they gets reduced a lot, but they are coming towards the scalp and
electrodes are actually going to pick them up.
So, you can see that in between there are so many areas that they have to cover actually.
And, naturally it becomes more and more feeble because here at this level it may be in the
milli volt range that we have earlier seen about the neuronal impulse conduction, but by
the time it reaches here it will be in the micro volt range. So, so much of charge dissipation
takes place.
So, whenever the wave of ions reaches the electrodes on the scalp then they can push or
pull electrons on the metal of the electrodes and that is where we get the signal. So, this
push or pull difference is actually measured at voltage across time and that is referred as
the EEG signal. So, that is what is the source of the EEG signal.
256
Now, what are the basic elements of EEG? Well, one of the most important thing about
doing EEG is to know that where to place these you know electrodes ok. So, the electrodes
how to place and how to make actually pairs of these then that actually will develop or
will be defined in terms of the montages. So, this actually refers to the placement and
mixing of this electrodes that how we are pairing it up.
Now, the EEG can be monitored with either a bipolar montage or a referential one. So, if
it is a referential one, generally we take the reference near the earlobes or if it is a bipolar
one then each one of them will be having its own references. Now, bipolar implies two
electrodes for one channel. So, you have a reference for that channel available for each
channel itself, otherwise you can have a common reference electrode as I told you which
is usually close to the other.
And, small metal discs usually they are made of stainless steel or tin or gold or silver, the
more expensive you would go for and they are covered with a silver chloride coating, they
are placed on the scalp in special position and these positions I will show you that this
positions are referred as 10 – 20 system.
Now, each of these electrode positions are levelled with a letter and a number. For
example, you can see here that Fp1, AF7 ok. So, F and AF that is anteriofrontal and then
only frontals ok, then frontocentral that is FCs, then only central that is C; some people
call this C as Z also.
257
So, in some things because in German central is zentral, so, some people refer them as Z
also. So, C, C5, C3, C1 and centroparietal, parietal, parietooccipital and temporal – all of
them you can see the abbreviations here.
Now, the other abbreviation is in terms of this numbers you might have noted. Well, the
odd numbers would mean that it is in the left side of the brain. So, you can see that this
side this is where it is in the left side of the brain, and the even numbers would say that
these are in the right side of the brain. So, all these numbers will be right side of the brain.
Now, the other important point that I will tell you is that keep these points in mind that
this is where is the nose and this is where is the occipital region, there is a tip at the occipital
region and we call it as inion. So, these are the two with respect to which we actually
spread it. Spreading of neuron is, spreading of electrode on the brain is an art on its own.
So, there is a rule that is called 10 20 rule I am going to explain it to you now and this rule
is made so that at different places of the world if this EEG test is happening, then there
will be a commonality people will not get confused between the positions. So, these are
the universal positions which is common all over the world.
So, whenever wherever a human scalp is getting tested, they do it with respect to these
positions and then they report the data and others can easily understand because these are
all universal locations. So, it is very important for us to understand these universal
locations.
Let us look into them.
258
As I was telling you earlier that the first thing before placing the electrodes you have to
mark is just above the nose is our nasion part and behind the brain there is a little lump if
you check your occipital area behind the brain that is called inion part; so, nasion and
inion. So, and the line that actually joins them this is the line that is what is the central line.
So, this is the central line.
So, you have to keep the central line in mind and everything is either there are places which
are left to the central line, this is left and this is the right part. And, you remember that left
all numbers are odd 1, 3, 7 etcetera and right all numbers are even that is one way to know
which one is left, which one is right.
Now, the other thing is that so, with respect to this central line if I go up to 10 percent from
the and this line is also known to us because this goes to the earlobes. So, these two lines
are actually known to us beforehand. So, that means, I can actually measure this you know
and 10 percent of the distance that is where inside also 10 percent we actually place our
first series of the electrodes ok. So, Fpz, Fp1, Fp2s, etcetera.
After that next placings are all at 20 percent, 20 percent. So, whatever is the total length
20 percent of it that is the distance where you place it everywhere else and again for the
inion 10 percent by the two sides of the inion left and right and all other areas are actually
all other areas here these are all 20 percent. So, this is the 10 percent, 20 percent rule that
we follow and based on that we actually place the electrodes.
259
Now, how many electrodes can we place? Well, we can place as you can see here that very
easily we can place in this manner you can see that there are central you know 5 electrodes
and then here you can see that there are 5 and 3, 8. So, 8 and 8, 16 electrodes are there and
center there are 5 electrodes and there are others which can go up to 32 electrodes EEG
machine. There are even other EEG machines which can go up to 64 electrodes.
So, you know there are various ones, but these are the basics that you have to cover and
these are to be placed when I say Fp1 the position is fixed that is it is 10 percent from this
side and 10 percent from this side, that is where the Fp1 has to be ok. And, then in between
you can add many mores that is not a problem and then you can accordingly you know put
this left odd and right even numbering system and the locations also you can number. So,
that is the way we use the 10 20 system of electrode placement.
We will now look into different components of the EEG. Now, as I told you that first and
the foremost important component of an EEG is actually the amplifier. Why the amplifier
is required? Because you need to actually amplify the signal which is initially at micro
voltage level to something like a milli volt level, so that we can easily analyze it.
So, you need an amplifier, but when you will be amplifying you will be amplifying the
noise also. So, that is where the role of amplifier is so important that it not only should
amplify the signal.
260
Now, let us say that this is the original signal and after amplification. Let us say this
amplifies, but also it should removes let us say if there are noises here it should remove
those noises ok. So, it should be a noise free. So, that if there are noises in between those
noises should not appear here. So, it has to amplify also keeping in mind the filter of it.
Now, it does this by taking energy from a power supply that is why you need a power
supply controlling the output to match the input signal shape, but with a larger amplitude.
And, human brainwave activity is too subtle to read unless the signal is amplified this units
are available now they can usually connect through a USB port and transmit signals to any
therapist computer and older units have a serial interface, but now there are much better
interfaces available.
So, in a typical amplifier there can be 32 channels the sampling rate can be 1024 Hertz
bandwidth can be from 0.1 Hertz to 400 Hertz. Of course, up to 400 Hertz you do not
require as I told you that may be up to 200 Hertz is good enough and also head stamps
with impedance of LEDs. So, that is what is the first important component of the EEG.
The next important component of the EEG is in terms of the filters. As you can see here
that these filters they use these filters, so that the original signal noises you can actually
attenuate depending on where you want to do it.
261
Like in 0 to 55 Hertz, after filtering you see this attenuated knowledge signal and then you
can also apply smoothing filters. There you can get an even better one and so, that is how
you can actually apply the filters. If you do not do it, then the data will be almost
unreadable.
So, that is why you need to clean up the raw data. So, this is also the raw data the first part
you need to clean it up by actually applying the filters.
So, we will talk about the filters. Now, the next point of course, is after filtering you need
to put that output. So, earlier days the output used to be using a writing unit and in a pen-
ink-paper system, but today you know it can.
So, the at that time it was important that what was the speed of the paper mechanism like
30 millimetre per second with at least the additional speeds of 50 and 60 millimetre per
second, so that you can record the data. If the speed is low then you will be missing some
of the high frequency components. So, that is where the speed is important.
And, today of course, the entire data comes in a digital screen which you can actually
record digitally and you can then analyze the data.
262
Now, whenever we talk about the EEG data, we have to know about the EEG waveforms.
And, starting from the low frequency to the high frequency, the waveforms that would be
coming up will be like delta band that is what is your delta band here and that is up to 4
Hertz. And, then you have alpha band that is 8 to 13 Hertz which is somewhere here; this
is what is your alpha band. And, you can have a theta band in between which is 4 to 8
Hertz; so, this is what is your theta band.
And, then in the same 8 to 13 Hertz region or 8 to 12 Hertz you can have a mu-rhythm
also. So, you can have a mu-rhythm. Now, you can see that both alpha and mu they share
the same bandwidth almost 8 to 13 and 8 to 12, but the pattern of mu and the pattern of
alpha these two are not the same, that is why it is little different. And, next then you have
the beta band that is about 13 to 30 Hertz. So, this is your beta band and finally, you have
the gamma band that is 30 to 50 Hertz. So, this is about your gamma band.
So, if you look at it very carefully the way it happens is that when you are in a very deep
sleep, then this is the you know this is the delta wave range. When you are dreaming or
day dreaming you are having episodic memories, then it is the theta; alpha is when you are
simply taking rest, you close your eyes alpha waves will be automatically coming up
within few you know minutes the brain signal will start to show the alpha.
Mu although is similar to alpha, but mu is different in terms of that it has it is related to

certain activities processes etcetera; beta is when you are fully conscious and in fact,
263
thinking of some activity movement then it is beta and gamma is even higher when you
are deeply concentrating on some task that is where the gamma will be coming into picture.
So, these are the various gradations of the EEG waveform.
So, for example, here I have also mentioned what I have told you few minutes before that
sensory motor mu-rhythms 8 to 12 Hertz even though they are similar, but they are they
look different. As I told you the mu from the alpha and they are used to recognize
intentions or preparations of movement and also some imaginary motor, this is where the
mu-rhythm comes into picture.
Now, beta are associated with actually alertness, arousal, concentration and attentions
etcetera. So, that is where the beta wave comes into the picture. And, then we have the
gamma band that is the last one that is where is the gamma and this is the beta. So, gamma
is for mental activities such as perception, problem solving, creativity etcetera and as I told
you that even though alpha band and mu-rhythms cover the same frequency range.
However, the waves are different and you can clearly see that the waves are these two
waves are different in nature. So, you can clearly see the differences.
Now, let us say you want to carry out yourself some EEG test you have a 16 channel or a
32 channel EEG setup available with you. So, the first there is a step by step process to
264
prepare the subject, we have to know it in order to get the good quality data. So, that is
what I will be discussing.
So, now, we will talk about the preparation for EEG and as you can see that first of all you
have to take care that the subject must wash his or her hair with shampoo. Conditioner is
not needed.
There should be as little oil as possible. So, that the results will not be tampered and the
subject must tell his, her health care provider of all medicines prescriptions and
supplements that they are taking. Important thing is that the subject should not have sugar
and if there is sugar etcetera then there has to be some precautions that are to be taken.
The subject must discontinue using medicines particularly the subject is taking some neuro
medicines, then they are to be discontinued because otherwise that may interfere with the
test. So, if the health care provider has directed then this is to be done. And, then also
another important thing, the subject must avoid consuming any food or drinks containing
caffeine for 8 to 12 hours before the test because caffeine has some neurotransmitters that
can affect the normal neuronal activity.
If the subject is having a sleep EEG sometimes we take the sleep EEGs, he or she may be
asked to stay awake the night before, so that you can get a deep sleep and through that you
can capture maybe the delta weights. And, of course, the subject must avoid fasting the
265
night before or the day of the procedure. Why because then there will be low blood sugar
and that may also influence the results. So, you need to avoid the low blood sugar as well.
So, these are the basic preparations that you have to take.
Next is what we will be doing during the EEG procedure. So, as you can see here that there
is a cap and this cap is having it has actually taken care of the 10 percent, 20 percents more
or less by these places where the electrodes are placed. So, a standard such EEG
experiment will take about an hour you have to keep in mind and the subject will be
positioned or a padded bed or chair you can do either way.
To measure the electrical activity in various parts of the brain there has to be a nurse or
EEG technician will attach 16 or 20 electrodes, that is why is these electrodes are. So, 16
or 20 electrodes to the scalp.
The brain generates electrical impulses that these electrodes will pick up and to improve
the conduction of these impulses to the electrodes a gel usually we use a gel of argentum
chloride AgCl gel, that is what is used generally between this electrode and the scalps each
electrode and the scalp.
And, the electrodes gather the impulses given up by the brain and it do not transmit any
stimulus to the brain that is to be understood, that is to be told many times for the kids, that
266
is to be told to their parents that EEG is only a sensor. It is not going to put any signal to
the brain. It only picks up the signals whatever is available outside the brain.
And, the technician may tell the subject to breathe slowly or quickly and may use some
visual stimuli such as flashing light to see what happens in the brain when the patient sees
these things and the brains electrical activity is all throughout recorded through the EEG
testing. So, that is the way the test is carried out.
Now, after the EEG procedure after the test is complete, then of course, you are removing
these electrodes. So, you need to remove, nothing is remaining on the body. The subject is
generally instructed to resume any medication by the medical caregiver and the subject
will be ready to go home immediately generally following the test and no recovery time is
required for this. So, that is what is after the EEG procedure, very simple.
267
Now, here I am talking about some of the typical EEG results that you will get and you
can see that when you are directly getting through from the channel how complicated
looking they are.
And, in fact, in some cases if there is a seizure kind of a thing if then you may see this kind
of you know activities that will say that there is something like a seizure happening.
Otherwise it will be like a very complex noise you know train or a signal with a low signal
to noise ratio.
Now, each area of the brain will produce a different brain wave strip as you can see here
every channel has a different strip in it and when examining the recordings neurologist
looks for certain patterns that represent the problems in particular areas of the brain
because these are all channels and you know which channel corresponds to which area of
the brain. And, the results are compared against a normal brain which has a specific
brainwave pattern.
So, that is the way we actually work on the EEG results.
268
Now, at this stage it is absolutely important to use the filters in order to make the results
of any sense. So, without filters many segments of EEG would be actually unreadable.
Now, EEG filters are not only used to remove noise, but also involuntary body movements
during taking EEG with blink many a times.
So, this blinkings are to be also filtered out. So, this is also needed to be done and the three
common filters that we are using in such cases are low-frequency filter or also known as
high-pass filter; high-frequency filter also known as low-pass filter and notch filters. So,
these are the three that we will be using.
We will now talk about that what are these three different sensors, what is the constitutive
relationships and how they are useful in each of the areas of EEG analysis let us look into
that.
269
So, the first filter that we have talked about is a low-pass filter; that means, a filter which
actually passes signals with a frequency that is lower than a selected cut-off. It allows those
low things and the higher frequencies are actually filtered. Now, the way it is done is by
using a RC circuit as you can see that there is an RC circuit that is used and the output
voltage with respect to the input voltage here can be simply written by Kirchhoff’s law as:
𝑑𝑣𝑜𝑢𝑡
𝑣𝑜𝑢𝑡 (𝑡) = 𝑣𝑖𝑛 (𝑡) − 𝑅𝐶
𝑑𝑡
Where, 𝑣𝑖𝑛 (𝑡) is a step function of magnitude V in that is the input voltage and output is
coming out of the you know signal filter or condition. Now, this 𝑣𝑜𝑢𝑡 (𝑡) can be written in
terms of the input as something likes 1 − 𝑒 −𝜔𝑜𝑡 you know that is common for a RC circuit
and where this 𝜔𝑜 is actually these RC circuit.
𝑣𝑜𝑢𝑡 (𝑡) = 𝑉𝑖 (1 − 𝑒 −𝜔𝑜 𝑡 ),
And, the cut-off frequency if we try to use it in terms of the corner frequency or cut off
frequency, then this 𝜔𝑜 is which is in rad per second you have to divide it by 2 pi, then
you will be getting that fc.
Now, you can actually apply Laplace transformation to this signal and then what you will
get as a transfer function is an algebraic relationship between 𝑽𝒐𝒖𝒕 and 𝑽𝒊𝒏 and that will
𝝎
be a first order filter like 𝒔+𝝎𝒐 . Now, because this polynomial is a first order s1, so, that is
𝒐
270
why this is also called as a first order filter. And, the response of the first order filter is
usually in terms of this kind of a direct cut off as you can see.
𝑽𝒐𝒖𝒕 (𝒔) 𝝎𝒐
𝑯(𝒔) = =
𝑽𝒊𝒏 (𝒔) 𝒔 + 𝝎𝒐
So, it is these are the pass bands by varying 𝝎 or the varying RC you can actually define
the bandwidth up to which signal you will be passing. And, the same thing is true for the
phase also, that up to which point the phases are to be retained. So, that is the fast low pass
filter.
We will then talk about what we call a high pass filter and that means, it will be you know
for the low frequency part will be actually filtered. So, it will be allowing the frequencies
higher than a certain cut off frequency. Now, why this is required? As I told you that
suppose I want to see a low wave something like a delta wave or an alpha wave then what
I will do? I will go for a low pass filter.
But, suppose I want to see something like a beta wave or something like a gamma wave,
then what I can do is that I can cut-off this alpha wave or delta waves and that is where the
high pass filter will come into picture. And, as you can see here that this is where the C
and R are in actually series unlike the last case and here the transfer function will then
271
become there will be a 0 in the transfer function in the numerator and denominator is the
same.
𝑽𝒐𝒖𝒕 (𝒔) 𝒔𝑹𝑪

𝑯(𝒔) = =
𝑽𝒊𝒏 (𝒔) 𝟏 + 𝒔𝑹𝑪
So, as a result of this the way in which it is filtering is just the other way around; that
means, up till this point the gain is very low. So, it will not allow the signal to come and
beyond this point then the signals are actually coming that is where is the pass band of the
1
signal. And as I told you that this cut off frequency f c is actually 2𝜋𝑅𝐶 that is 1 over RC is
𝝎𝒐
𝝎𝒐 . So, it is .
2𝜋
1
𝑓𝑐 =
2𝜋𝑅𝐶
So, that is what is an high pass filter. The phase is just the reverse in this case as you can
see that initially there is a phase and as the cut off frequency is crossing, then the phase is
actually getting filtered. So, that is the way the high pass filter is going to work.
Now, you can have a combination of a high-pass and a low-pass. In that case what we will
need is something which is called a band-pass filter. So, that means, some part of the signal
initially is cut and some part of the signal beyond a point is stopped and the pass band is
272
somewhere in between. So, that can be also a frequency response and this is known as a
band-pass filter.
So, essentially it will be as you can see this it will be a combination of some sort of a low-
pass and a high-pass filter. So, that is what is the band-pass filter. As you can see that the
phase here is also during the pass band it is actually changing the nature from positive to
negative, +90 to -90 the phase changes its nature. So, that is what is a band-pass filter.
Well, low-pass filter, high-pass filter or band-pass filter they are good when you are
actually chopping out some part of the EEG response and you are only looking at a
particular band of the EEG response.
But, as I told you that let us say at this moment itself you can see that my eyes are blinking
and if there is an electrode it is going to get that signal immediately and there will be a
sharp peak there and many times that actually camouflages the actual signal. So, we need
to filter that, in such cases we will be using actually something called a notch filter. So, let
us look into a notch filter.
So, here we are talking about a notch filter. Now, what does this notch to as you can see
the name itself is telling that it is actually notching under a certain band, it is not allowing
any wave to pass and this notch filters are actually second order. So, far we had first order
273
filters, but here these are second order filters and you can see that they are having two 0s
and also they are going to have two poles.
So, usually in you know if we draw the real versus the imaginary of x for such filters. So,
let us say real s versus the imaginary of s then, it is going to have there are 0s there which
will be in the high frequency range and followed by some poles which will be far away
from the real axis.
𝑽𝒐 (𝒔) 𝒔 + 𝝎𝟐𝒐
𝑯(𝒔) = =
𝑽𝒊 (𝒔) 𝒔 + 𝝎𝒄 𝒔 + 𝝎𝟐𝒐
So, this is what these are the poles P 1, P 1 and P 2. So, these poles are nothing, but the
roots of the denominator. So, this is this denominator is going to produce these poles. And,
the numerator which is also a quadratic polynomial is going to produce these 0s.
So, this is where you know is typically a notch filter. So, this is what is the amplitude and
this is what is the phase that you can see that the phase shifts from 0 to you know -90 to
+90 and then it comes back to 0 that is what is the notch filters phase characteristics. So,
these are the amplitude and the phase characteristics of a notch filter.
Now, the type of artifacts that I was just mentioning eye blink is just one of the artifacts.
There are many different types of artifacts which can be originated due to environment
noise, experimental error or some physiological artifact like the eye blink itself.
274
Now, environment artifacts and experimental errors and environment artifacts, they
actually come from external factors they are extrinsic; whereas, the physiological form
will be from eye blink, from muscle activity, from heartbeat, they are actually intrinsic
artifacts.
So, the environment artifacts can be eliminated by filters, if you know what is the
environmental characteristics, but physiological artifacts are generally more difficult to be
removed. Of course, you can use as I told you the notch filters, but they are generally
difficult because notch filters are generally for a specific filter frequency range.
So, if you have a specific peak at a particular frequency a notch here can work, but what
if this frequency changes its position? What if the heartbeat changes? Then this is not
effective. So, in such a case you need a little more involved type of an artifact removal
algorithm.
So, one of the algorithms that we use in such a case is called K-SVD algorithm which is
particularly used for eye blink artifact removal. As you can see here that all the data in the
14 channels of a subject and even you can clearly see the eye blinks, you can clearly see
and if you use a K-SVD which essentially works on a dictionary learning. It first learns
this frequency nature of frequency of this and then you know it creates a dictionary for
sparse representation using a singular value decomposition approach.
275
And, then this K-SVD actually is based on a k-means of a clustering method, that works
iteratively on each one of these signals and make a best fit and then it gives the filter data.
So, this is what becomes the filter data where this entire artifacts are absent.
So, this needs some algorithms to work now luckily most of the EEG comes with software
and these software’s will be definitely having K-SVD as one of the popular option which
can take care of this. But, you need to know that this kind of a filtering needs to be applied
for things like eye blink removal.
So far we have discussed about the ambient signal that is anytime if you place an EEG on
the brain what are the signals that will be coming out and how you can take out the artifacts
noises etcetera from it.
Now, we will look into that if you give a specific stimulus to the brain, let us say you give
me a task to actually play a game or to count a number or semi such any such tasks then
actually the external stimulus comes into picture and then there is something called event
related potential that we can measure from this signal that is what we will be discussing
now.
So, now, we will be talking about event related potentials or in short ERPs. These are small
voltages that is generated in the brain structures when you are subjecting it to specific
events or stimuli. Now, EEG changes that are time locked to sensory motor or cognitive
276
events that provide non-invasive approach to study these psychophysiological correlates,
this is where the ERP is actually very important.
So, it essentially reflects the summed activity of postsynaptic potentials produced when
large number of similarly oriented cortical pyramidal neurons in the order of some
thousands or millions of neurons will fire in synchrony while processing the information
and this ERPs can be divided into two categories.
Early waves – components peaking roughly within the first 100 milliseconds and they are
termed as sensory or exogenous as they depend largely on the physical parameters of the
stimulus. And, then the cognitive or endogenous ones which is after 100 millisecond or
200 millisecond as the subject evaluates the stimulus, then the reflection of that comes in
the form of cognitive or endogenous signals.
So, you see both of them are important for us from a cognitive robotics point of view and
this waveforms are described according to their latency that when they are appearing and
their amplitude.
So, these are some of the common ERPs as you can see here that Ps are the positive part
of the ERP in the signal and then there are negative parts also. So, early part within 100
millisecond you will get you will expect P1 N1s and beyond that you are getting P2 N2
and P3s. So, this is what is in the later part that you will be getting into the system.
277
So, P1 sensory peak which is elucidated by visual stimulus. P3 depends on task and there
are still some debates that what is the exact reason some people think that you know it
comes because of a specific type of a response particularly error related responses etcetera.
N1 is influenced generally by spatial attention and N1 is generally obtained from lateral

occipital regions. N2 is the error related negativity which is observed when the subject
makes an error or the no go. So, this is where N2 comes into picture and to some extent
P3. So, this is what is the ERP or the Event Related Potential.
Now, what is the advantage of this EEG? Well, first of all it is a non contact way of judging
the level of the brain. It is also useful during disorders not just a general brain condition
like seizure disorder, head injury, encephalitis, brain tumour, encephalopathy, memory
problems and sleep disorders. All these cases the EEG becomes a very very useful tool.
Also, when someone is in a coma, an EEG may be performed to determine the level of the
brain activity. And, the test can also be used to monitor activity during brain surgery. So,
there are many usage of the EEG.
278
What are the precautions? Well, first of all EEG in very rare circumstances can cause
seizures in a person who has already a seizure disorder. So, that is something that you
know is generally it is very rare.
Now, certain factors or conditions that may interfere with the reading of an EEG are low
blood sugar, body or eye movement, lights, drinks containing caffeine I already told and
oily hair that is why they say do not apply shampoo on the hairs. So, this is all that we
have to keep in our mind for a successful EEG test.
So, these are the points that we have to keep in our mind for an EEG test which is very
important for brain computer interfaces.
Thank you.
279
Lecture - 11
Good morning students, welcome to the course of Foundations of Cognitive Robotics.

We have now reached the final week of the course. And in this week we are going to
focus specially on intelligence. So, we will discuss about various theories of intelligence.
And in the first lecture of this week, I would like to begin with the theories of
intelligence from the psychologist’s point of view.
I have chosen four psychologists - four very famous psychologist’s point of view from
this aspect. I will talk about Spearman; I will talk about Thurstone, Howard Gardner, and
Robert Sternberg. So, let us look into that how they define intelligence.
280
So, as we are discussing about the theories of intelligence, my outline of this talk will be
that first I will talk about various concepts and types of intelligence. And then I will of
course start with the Spearman and his g-factor. Then Thurstone’s view of this g-factor,
and Gardener’s theory of multiple intelligences. And finally, I will talk about Sternberg’s
Triarchic theory, so that is the way in which we discuss.
Now, our story of intelligence starts with first the Spearman’s point of view. And the
reason why we are talking about the psychologist’s point of view intelligence is that so
281
far there is no standard definition in terms of that what exactly constitutes intelligence
that is absent. Some researchers believe that it is a single general ability, while others as
you will see that they would say that it contains of a range of aptitudes, skills, and
talents. So, you know there is a difference in terms of this.
But in general this is agreed that intelligent (Refer Time: 02:47) is a concept; it is not a
concrete thing. And what is this concept? These concepts will definitely have elements
like the ability to learn, the ability to recognize that there is a problem there, and the
ability to solve the problems. So, under these three corners, the fundamental tenets, the
theories of intelligence will be actually defined by different psychologists in different
way.
Now, let us first look into this Spearman’s way of the theory of intelligence. Now, before
we do that we have to keep in our mind that intelligence is generally described under two
parts; one is called crystallized intelligence, and another is called fluid intelligence.
So, what is the crystallized form of intelligence? Well, crystallized intelligence consists
of the accumulation of knowledge, facts and skills that are acquired throughout the life.
And it includes reading, comprehension and vocabularies for example, based upon facts
and rooted in experiences.
282
It becomes stronger with age as one accumulates new knowledge and new understand.
But this is essentially the accumulation of knowledge.
Whereas, the fluid intelligence is something which refers to the ability to think and
reason the ability to think and reason abstractly and solve the problems. And this is
independent of that what knowledge you are having. You can have enormous
knowledge, but you may not have that ability of the fluid intelligence to solve the
problems.
This is generally independent of learning, experience, and education. And some parts
which are important in it are like solving puzzles and coming up with problem solving
strategies.
Generally, this speaks in adolescence, and begins to decline progressively beyond 35 to

40 years of age. However, both the types of intelligence increase throughout the
childhood and adolescence, the crystallized intelligence and the fluid intelligence. So,
these are two different aspects of the intelligence.
283
Now, let us look into Spearman’s general intelligence. Now, Spearman, between you
know 1863 to 1945 is a famous British psychologist. He actually divided intelligence
into two parts; one is the general intelligence or the g-factor, and another is the specific
intelligence or the s-factor.
This is based on his observation that people who are generally well in one test tend to do
similarly well on other tests of mental ability. So, which means there must be something
general about the intelligence that is why this is possible that was the assumption based
on this observation.
However, it is also found that people are not equally skilled on everything. So, there are
some s-factors scores that measures a person’s ability in one particular area like in terms
of memory, in terms of attention, verbal comprehension, spatial skills and abstract
reason.
So, we have all these s-factors like numerical, mechanical, spatial, verbal. And in short
of overlap of them, but at the center lies something called a g-factor that is the general
intelligence, so that is the two factor theory that was the first kind of you know way of
defining what is called intelligence, the g-factor and the s-factor.
284
Now, this g-factor is responsible for all mental activity. And this is generally acquired by
the genes. It is present in all, but varies in terms of how much of g-factor one is having,
and it is assumed that this cannot be changed by education, practice and experience. This
is something which is kind of people are born with.
However, the s-factor is responsible for doing specific works, and this is not by birth,
this is not by birth, this is something which is acquired. So, specific intelligence is
related can be related to say scientific knowledge, singing, sports, these are all acquired.
This can be changed or improved by education, practice, and experience. So, the s-
factors according to Spearman’s theories acquired, and g-factors that is something that
comes as genetic; and the two together defines the intelligence.
However, not all psychologists have actually agreed with Spearman’s you know general
intelligence concept, this is something which is you know it is controversial, not all the
psychologists are agreed with it. As there are different views of the psychologists
particularly the view of that the g-factor is something which has been at a subsequent
stage quite challenged by the other psychologists, and they try to redefine these factors in
terms of more specific observable way of intelligence.
285
So, let us look into beyond these Spearman’s definition we would go into Thurstone’s
theories of intelligence first. We will look into Thurstone’s Modified g-factor
Intelligence Theory.
And in this case Thurstone, it is from 1887 to 1955 has actually developed this theory
which is based on certain more you know distinct definitions of some of the aspects like
associative memory, numerical ability, perceptual speed, reasoning, spatial visualization,
verbal compensation, comprehension and word fluency.
286
So, Thurstone’s focus are for example, associative memory which means how much you
are able to actually memorize and you know this memory is not just from the
hippocampus, but you know association of one memory aspect with other previously
learned aspects of the memory. So, that is you know the ability the more you can do that
association that it goes beyond the hippocampus area, so that is something that actually is
defined in intelligence is kind of a telltale sign of this associative memory.
Then numerical abilities like the ability to solve arithmetic problems. Perceptual speed
the ability to see the differences and similarities around objects. Reasoning, you know
ability to find that there are some inherent rules in a system. Special visualization, the
ability to visualize the relationships and between the things. And the verbal
comprehension, the ability to define and understand the words, and the word fluency that
is the ability to produce words rapidly in terms of explaining something.
So, according to him that the g-factor that spearman has defined is something which
needs to be more specific and he has defined these seven specific areas by which the g-
factor can be actually defined. And this cluster of abilities is what he has focused on. So,
even though he has not entirely challenged the Spearman’s theory, but he had modified
or sort of taken out the vagueness from the definition of Spearman.
287
Now, the point here is that is intelligence, Is it really one general trait or many specific
abilities because what Thurstone has said that the general intelligence itself has many
specific abilities.
Now, in today’s context neither spearman nor Thurstone entirely denied the ideas of each
other, but differed in what they thought which mattered most. But most psychologist
today agrees that there is a positive correlation between various stay you know aspects of
the cognitive ability.
And those aspects that we had just discussed things like associative memory things like
perception, things like you know numerical aptitudes all these have it has a positive
correlationship with the cognitive ability and that can be defined in terms of g-factor, but
it is probably made up multiple general ability factors which Thurstones has said.
And it is the definition is still not very close there can be many other factors which can
go inside. Now, let us look into some of the more advanced let us look into some of the
more advanced theories of intelligence. So, beyond Thurstone as the focus is on multiple
intelligence we look into Howard Gardener’s point of view of intelligence now.
We will now look into Howard Gardener’s Multiple Intelligence Theory.
288
Now, Howard Gardener is from 1943 to present, he is living psychologist one of the very
famous and from Harvard University. He has proposed that the traditional idea of
intelligence, based on so called IQ testing that typically only account for linguistic
logical and spatial abilities, this did not fully and accurately depict a person’s ability.
So, he proposed eight different intelligences based on skills and abilities that are valued
in different cultures. Now, let us look into what are these eight different intelligences.
289
Well, the first one is what is called bodily kinesthetic intelligence. So, what it means is
that the ability to control the body movements and to handle objects in a very skillful
manner, is something like the athletes will be doing, something like the dancer will be
doing or a craftsman will be doing, so that is called bodily kinesthetic intelligence.
Then there is something called interpersonal intelligence – the capacity to detect and
respond appropriately to the moods, motivations, and desires of others; something which
politicians do, a salesman or a clinical psychologist do, so that is what is called
interpersonal intelligence.
Then there is something the third factor so bodily kinesthetic intelligence that is the first
important factor, interpersonal intelligence the second important factor. And then there is
a third important factor which is intrapersonal intelligence. So, this is something like the
self awareness, and it is in tune with inner feelings, values, beliefs, and thinking process;
and that is something which generally will observe in philosophers, essayist, etcetera.
The intra personal intelligence something which is the inner feeling of a person.
And then he has talked about logical or mathematical intelligence. So, this is as the name
itself suggests that the capacity to think conceptually and abstractly, capacity to discern
logically or numerical patterns; something which a scientist would do, a mathematician
or a navigator would do, so that is what is the fourth intelligence factor the logical
mathematical intelligence.
Then there comes the musical intelligence, that means, the ability to produce and
appreciate rhythm pitch and timbre which is as you know people like musicians,
composers, or singers you will generally find it with them. The next type of intelligence
he has noted is called naturalistic intelligence something which refers to the ability to
recognize and categorize animals, plants, and other objects in nature like geologists, like
botanists, like ecologists etcetera.
Then of course, there is the verbal linguistic intelligence like well-developed verbal skills
and sensitivity to sound, meanings and rhythm of words something which you will find
in poets, in writer, in public speakers, so that is what is the verbal linguistic intelligence.
290
And finally, the visual spatial intelligence, the visual spatial intelligence naturally talks
about images and pictures, how to visualize accurately and abstractly like a sculptor
does, a painter, chess player or architects. So, these are the eight categories.
The bodily kinesthetic intelligence, interpersonal intelligence, intrapersonal intelligence,

logical mathematical intelligence, musical intelligence, naturalistic intelligence, verbal
linguistic intelligence and visual spatial intelligence, these are the eight factors according
to Howard Gardener which actually defines the core intelligence.
Now, this itself once again is not the final theorem. This itself is not the final theorem of
intelligence. There is further you know we talk about Sternberg’s the point of view of
intelligence, and that is where we will try to complete the psychologists point of view.
So, let us look into that what is Sternberg’s point of view of intelligence.
We will now look into Robert Sternberg’s Triarchic theory of intelligence. Now, this is
from the psychologist Robert Sternberg at Cornell University who defines intelligence as
mental activity directed towards purposive adaptation, selection and shaping of real-
world environments relevant to one’s life.
So, what Sternberg says is that whatever gardener is telling in terms of various types of
intelligences, these are nothing but certain skills. And they all refer to in a sense all of
them together refers to that what is the capacity of an individual towards purposive
291
adaptation towards selection or towards the shaping of the real world environment, you
think of any of the capacities that we have discussed like say interpersonal abilities.
So, this talks about that how a person will be dealing with his environment in a way that
actually improves the success of his or her ability to survive in the environment
essentially acknowledging that the homo sapiens are in always in a socialistic animal. So,
they are in a group. So, in a group survival depends on how you are actually convincing
your group towards your points of view.
Like that if you look at other you know aspects of it, let us just took look at say
something like intrapersonal the philosophical point of view. Again the person who is
having an intrapersonal kind of skills will be able to abstract the various things that
actually happen in life; and through that abstraction the person may actually gain you
know significance in the society.
And that is the same way you look at all other abilities the ability of say for example the
word, the vocal abilities or the other these are all some of them are related to
entertainments. For example, the visual abilities or some of them are related to help
homo sapiens to establish themselves in you know in the nature among the nature you
know how to survive.
So, all of these are actually whatever you know has been defined by Gardener, this
according to Sternberg are actually skills. And these skills are essentially certain aspects
of mental activity which are directed towards the purposive adaptation, and shaping of
the real-world environment, so that is what is his point of view.
He stress is that both the universal aspects of intelligent behavior and the importance of
adapting to a certain social and cultural climate these are the important aspects of
intelligence. So, that is what is the foundation of his Triarchic theory of intelligence.
292
Now, basically grouping you know regrouping all the things and abstracting it these are
the three that he found to be important in terms of intelligence that is analytical
intelligence that is the ability to evaluate information and solve problems; creative
intelligence that means the ability to come up with novel ideas; and practical
intelligence, ability to adapt to a changing environment.
So, these are the Triarchic these are the three you know things which are important in
terms of successful intelligence parameters so that is what Sternberg believes that a
balanced measure of all the three forms of intelligence would actually result in greatest
life success, and that would actually ensure the continuation of the project, so that is
what is Sternberg’s point of view in terms of intelligence.
293
So, if you ask me that what is the best theory in terms of intelligence, we would say that
Spearman’s theory is generally most recognized. However, there are Gardener’s Theory
or Sternberg’s Theory which gives us more new and promising insights.
We can also say that while IQ tests remain the standardized test of intelligence, this may
not always be the case. And as our world changes and new theories of intelligence will
continue to emerge as we will be able to understand the world better. We may find that
our assumptions about human intelligence itself is not quite what we once thought.
So, this is where we will put an end in today’s talk on psychologist’s point of view of
intelligence. In the next talk, we will talk about the other points of view of intelligence
particularly from the robotics or from the artificial intelligence point of view.
Thank you.
294
Lecture - 12
Welcome to the course of Foundations of Cognitive Robotics. We have reached the final
week and in this final week in the first lecture I have talked about the various theories of
intelligence particularly from the psychologist’s point of view. Now, there is also another
perspective to intelligence which is very important for us from the robotics point of view
and that is the perspectives of artificial intelligence.
There are distinctly two groups of artificial intelligence perspective. It starts with the soft
AIs, where the AI is the Artificial Intelligence is described in terms of soft rules. And,
then there is a further development towards the hard rules of AI where the AI is
embedded in robotic system.
So, we will see that how this theory is developed and with that perspective we will see
that what are the different theories of artificial intelligence, a glance of it. We will not be
able to have all the things in this lecture, but a glance of some of the important theorems
of artificial intelligence; we will see it now.
295
In this lecture we will cover these following you know famous personalities and their
theories related to artificial intelligence. We will first talk about Marvin Minsky’s
contribution in terms of Society of Mind that is the book that I have chosen and that is in
1987.
Then we will talk about Allen Newell’s point of view that is the Unified Theories of
Cognition, that is after the Minsky’s work and also John Anderson’s ACT theory. It has
undergone several evaluations, although the first part of the theory, the first initial
architecture was in 1983.
Now, up till these this will be mostly computationally dominant artificial intelligence.
So, I will simply call it as the soft AI and then we will discuss about some of them which
are like hard AI. So, this will encompass the Dynamic Field Theory from Schoner and
also a very important book that I have covered many times in this lecture series that is
Rolf Pfeifer and Josh Bongard’s Body Shaping Cognition.
So, both of them and they will be towards the hard domain and finally, we will talk about
something which will be mix of the tool that is the LIDA and the cognitive architecture
corresponding to it.
So, let us call this to be a mix of soft and a mix of soft and hard rules and maybe we can
classify it as the hybrid technique. So, that is the hybrid technique ok. So, this is what we
will be covering soft, hard and hybrid. Let us first start with the soft theories of artificial
intelligence.
296
Now, I will start with Minsky because no theory of artificial intelligence can remain
complete without any discussions of Minsky’s view on intelligence. And, in a very
simple manner Minsky has covered intelligence as simply the ability to solve hard
problems. Now, the question is that what are these hard problems? Well, our minds
contain this exactly in Minsky’s words contain processes that enable us to solve
problems we consider difficult.
Intelligence is basically a term it is our mind for whichever of those processes we do not
yet understand that you know why these are the things that we are able to solve. So, that
kind of hard problems for which there is still lack of understanding Minsky wanted to
say that this is what we actually refer to the term through intelligence. Now, here also he
has made a division between an expert because as an expert he said it is essentially an
agent which needs a large amount of knowledge of only relatively few varieties.
So, it is a large amount of knowledge with relatively few varieties. Something like if you
consider that an intelligent chess player that is or an intelligent chess agent a robot. Now,
this kind of agents will be only expert of chess that intelligence we will not get reflected
in other problems. So, in contrast if you consider an ordinary person’s common sense
you would see that would involve a much larger variety of different types of knowledge
and that requires more complicated management systems.
297
So, according to Minsky that instead of an expert point of view of development of the
agents, we should, if we want to look for intelligence we should broad our scope and we
should start to see how these you know larger variety of different types of knowledge can
be actually tackled under one umbrella.
In fact, many times Minsky has mentioned that it is more difficult to emulate a you know
some of the children’s actions in a robot than to do something which is a very
industrially standard dexterous kind of a work.
So, that is what you know is related to the common sense and what is so easy for us may
not be at all easy for the robot; because what looks so easy may actually involve a lot of
complicated processes inside it.
Now, some of the interesting questions here to raise are how each separate parts of an
intelligent system actually work? If you consider the brain itself as a source of intelligent
system each one of the parts of the brain how does it work? And, how does each part
interact with the other parts? And how all these local interactions is going to get
combined to accomplish what the system does?
So, there are three questions that if there is a subdivision then how each subdivision
works? How the subdivisions interact among themselves? And how these interactions
298
finally get combined whenever a system has to do something which is like a final
deliverable from the system?
In connection with human brain this will also raise the questions like how the brain cells
work or how cells of each type would interact and how our billions of cells are organized
into societies? Which he has mentioned as the society of mind, how they are organized in
terms of doing some work?
Now, again and again there are two concepts that has been used by Minsky and that is
very important even in today’s context of artificial intelligence which is agent and
agency. Agents are considered to be simple experts having specific skills used for
carrying out an operation.
So, they are essentially treated like a black box, let us say that you have to add some
numbers. So, if you have an organ which carries out this addition. So, it carries out this
addition of any number you give say 4 and 5 and you immediately get an answer that the
answer is 9.
Now, we do not know how exactly the brain is working in order to get this number 9. We
will simply say we know and that we know is because of an agent who is carrying out
this addition operation inside the brain and that works pretty much like a Blackbox. So,
this is one important concept related to the agent.
299
Now, there is another part which is called agency. Now, agency term is has a very broad
convocation in English. For example, you want to plan a travel, then where do you go?
You go to a travel agency or let us say you want to plan some savings. So, you go to
some savings agency.
So, agency suggests the image of an organization that consists of several interacting
subagents. Now, whenever it is a question of agency then there is an image map that is
available for us, but for agent it is a black box; that means, we do not know how the
whole thing works.
So, essentially the society of mind according to Minsky or any intelligent system consists
of several agents and agencies. And, maybe some of these agents work under an agency
in that case it will be like a black box and sometimes there can be independent agents.
But, these are the two things which is very important in order to understand how an
intelligent mind or on how an intelligent system could be developed.
Now, another important point is that there almost invariably exists a kind of an interface
between the deep brain and the brain that interacts with the world. So, with the brain that
interacts with the world we will call it as A brain and the brain that is that can only
interact with A brain is known as the B brain. So, we have for example, different
somatosensory systems and this somatosensory system interact with the environment and
pick up the information.
300
Now, that is done by a brain which would actually have some knowledge of what they
are looking at, but mostly they themselves are actually controlled by the B brain. So, B
brain looks at that what is what are the informations that A brain has brought and B brain
simply kind of guides the A brain to look for the informations, to change the resolution,
to change the perspective and everything. So, you see that is what is the kind of the work
of the B brain.
Now, some of the activities that A activities and B activity; for example, A it seems
disordered and confused while carrying out an activity. Then B may inhibit that activity
or A appears to be repeating on some activity.
Then B may say that make A stop, A does something B considers good, make A
remember this or A is occupied with too much of detail, make A take a higher level view
or A is not specific enough focus A on lower level details. So, this is what the B controls.
So, this is the B’s role over A.
Now, this kind of a hierarchy according to Minsky can actually help in terms of
developing an intelligent system. We have seen the Minsky’s view on artificial
intelligence. Now, almost contemporarily there is Allen Newell who had also looked into
this problem and helped to flourish the field of artificial intelligence. And, Allen’s point
of view was in terms of developing a kind of a unified theory of artificial intelligence.
Because, there is this group of psychologists and there is the group of computer scientists
and Allen kind of worked as a binding agent between the two. And, he has developed the
first framework under which according to him this kind of integration is possible. So,
there are certain demands from his unified model and let us look into those demands.
301
Well, Allen Newell’s Demands of Unified theory of Cognition. There are actually 13 of
such and he considers that if these 13 demands are fulfilled, we can consider that the
system is intelligent system or the prerequisites of the intelligent system has been taken
care of. Now, this starts with first of all behave flexibly as a function of the environment;
we will come to this point again and again that how the analysis of the environment
plays an important role for any intelligent system.
So, there is something called an ecological niche and every intelligent system is to find
out what is that niche, where it can prosper. And, in order to do that it has to be adaptive;
it has to behave in a flexible manner. So, that is what is the very first and important
thing. And, the second one is also correlated with these exhibit adaptive rational and goal
oriented kind of a behavior. The third important thing is that the intelligent system
should operate in real time.
It should not you know happen in a virtual world, it should happen in the real time and it
should operate in a rich, complex and detailed kind of an environment. So, the
environment should be a kind of an environment which for example, a living system you
know faces. It should be complex, detailed environment. This system, the artificial
intelligent system that one has to develop must use symbols and abstractions just like the
living counterpart.
302
Use language which should be both natural and artificial. Natural language is particularly
very very important, because we will see later on that that increases the power of
communication immensely. Learn from the environment as well as learn from the
experience. Acquire capabilities through the environment, operate autonomously, but
within a social community where there is a group of some you know things which are
working together.
Be self aware and have a sense of self. This is a big philosophical task of course, that
how can one be aware, how can an agent any agent, any robot be aware of itself and can
carry a sense of self. Be realizable as a neural system. So, it should be you know one
should be able to physically realize such a system be construable by an embryological
growth process. This is something which is also something which is not achieved till
today; that means, can a robot generate a new robot, procreate a new robot.
So, that is the embryological growth process and the other important point that is arise
through evolution. So, if all these 13 points are met, then actually we can say I mean
according to Newell that you can say that you have developed a truly cognitive
intelligent robot. Now, we will see that much of these cannot be actually fulfilled in one
shot. But, the next important milestone towards this is actually John Anderson’s ACT
theory. ACT refers to Adaptive Control of Thought - Rational.
303
So, sometimes it is called ACTR or ACTs are stars at different levels it has been
developed jointly by: Anderson and Christian Lebiere at CMU or Carnegie Mellon
University. Now, according to Anderson who’s a very much a proponent of a single
theory or cognition; a common cognitive system for higher level he thought it to be very
important. Because, it is supported by short evolutionary history of human intellectual
functions and it’s also the plasticity in acquiring the functions.
It is the kind of plasticity that we observe and common features among different
cognitive processes. So, all these three things that the human intellectual functions have
developed in a short evolutionary span only and it has a plasticity in terms of acquiring
functions, different parts of the brain. It has been shown many times that even if one
parts get injured or get damaged, the other parts come into action and work on these
directions.
So, that kind of plasticity it can quickly show and different cognitive processes have
certain common features. These are the three things that are actually kind of supported
that there must be a unified theory of cognition.
So, an important metaphor of the unitary theory is in the modern general purpose
computers for example, because for a general purpose computer it does not take help of
different subsystems. One single processor can handle let us say the word processing can
handle the internet, can handle the data analysis and things like that.
So, that is something which says that yes unitary theory is possible. The ability of homo
sapiens to acquire complex skills is actually not driven by any specific faculty of the
brain, that the brain is generic that is why these complex skills are achieved. And, the of
mind is a general pool of basic structures and processes. This was the basis of
Anderson’s ACT theory.
304
Now, if you look further into it, there is this ACT framework and one important concept
of the framework is the concept of production which is a set of condition action pair.
And, it is something like once you have this knowledge about how to do a specific thing
and then you know this is like an agent it continuously does that very specific things.
So, for example, you know if a logic is included like this, that if person 1 is the father of
person 2 and person 2 is the father of person 3, then the definition of grandfather is that
person 1 is the grandfather of person 3.
So, finding out the grandfather from this logical relationship is a production process.
Now, Anderson’s concepts is that essentially there are always three types of memories
with us. The first is the working memory.
So, that is where you know we essentially interact with the outside world, but the
working memory is limited in scope. Now, if something is too big then the working
memory stores it in the declarative memory, where the items category wise are stored.
And, whenever needed working memory can actually retrieve from the declarative
memory.
And whenever some function is to be actually carried out then the working memory
sends these to the production memory and the production memory has all these kinds of
condition action pairs. As I told you so, the production memory goes to that and it say it
305
executes it and it sends the data back to the working memory. So, that is what is the
structure that this three memory interaction with the outside world and that is the
architecture, the basic architecture of the ACT framework.
Now, we have talked about the soft rules in terms of developing intelligence. So, we
have seen Minsky’s point of view, we have seen Newell’s demands and we have also
seen some algorithm that Anderson had developed you know in terms of the ACTRs.
Now, the soft intelligence has limitations because, it is if you look at Allen Newell’s
logics that it is not interacting with the real complex world. In order to interact with the
real complex world, a system must have sensors, must have motors, must have
processing systems and a body.
That means the system the intelligent system must look like a physical robot. Now, this
the moment you develop this kind of a physical robot; what you will not be able to avoid
once this physical robot starts to interact with the environment is the dynamics of the
robot. So now, I will talk about some very fundamental work by Schoner and his groups
on the dynamic field theory. So, let us look into that.
We will now look into Schoner’s: Dynamic Field Theory and a very easy example
because we are just introducing the concept view that has been done by Schoner is by
taking the examples of Braitenberg’s taxis vehicle. Now, you know that taxis is
306
something which is very popular among the biologists like phototaxis for some of the
flowers or chemotaxis for some other plants.
So, taxis denotes that a gradient based movement. Now, we are considering here a
Braitenberg’s taxis vehicle and as I was telling that this vehicle is having some sensory
system. So, this embodiment is there then it has some nervous system and then it has
some motor system. Now, all these things are not hanging in thin air. So, it has a body
which actually you know kind of houses all these three things: the sensory system,
nervous system and the motor system.
Now, let us say that this sensory system has to move towards a particular source by
measuring the intensity through the sensors and because the two sensors are placed at
two different positions.
So, they will be having different intensities because one which is nearer to the source
will be having higher intensities. One which is away from the source will be having
lower intensities and you can see that the activation and the intensity in any structured
environment. So, it is actually inversely related in the sense that as the intensity is in
increasing the activation is going low.
So; that means, that here the intensity is more. So, this is higher. So, if we this is higher
here the intensity so, the activation will be actually lower here. So, the motor system here
will be actually you know if the same signal is brought here; then these motor signals
will be less here, there will be a less turning. On the other hand, if the intensity is lower
here then the activation will be more and that means, that the motors will be turning here
at a higher speed.
Because activation is directly proportional to the wheel motion; so, the higher the
activation the more will be the wheel motion. So, what will happen is that this whole
system will then rotate because in this side they will be having higher activation. So, the
motors will turn faster and this side they will be having lower activation. So, the motors
will turn slower and as a result they will be driving towards the source. So, this is you
see a simple Braitenberg’s taxi vehicle model.
Now, the interesting thing here is that from the vehicles point of view the only logic that
is given for the vehicle is these two logics; that if you get high intensity you should
307
activate low and if you get low intensity you should activate more. And, also this motor
logic that if you are having high activation of the signal then the wheel motion should be
higher, if you have low activation then the wheel motion should be lower.
But, if you look at this system from outside you will see that just simply following this
logic how interesting is its behavior. Then at that as soon as there is a source it
understands that how to target itself towards the source just by through this kind of a
thing. And of course, there is a structured environment here, we have not talked about
any presence of disturbance or multiple sources etcetera. So, the dynamics of the system
if we can actually develop then that itself then we will guide us towards the behavior of
it.
Now, what Schoner has found out is that there are some very important ingredients for
these kind of dynamics. And in the dynamic you know field theorem first of all you must
be having a model of the environment available with us, environment is very important.
Then based on the environment you should design a sensor model through the sensor
characteristics, a motor model through the motor characteristics, a model of the body
because the body is housing the sensors and the actuators and imposing certain
constraints on it.
308
It is linking for example, the turning rate of the vehicle that is the rigid body dynamics in
this case to the difference in turning rate of the wheels. So, that is where is the
significance of the body and also a model of the nervous system which is needed because
from the sensor the activation is going to the motor through the nervous system. So,
these are the five very important ingredients that you must have in terms of developing a
dynamic field theory based intelligent robot.
Now, the DFT can be also that is the greatest you know advantage of DFT that it can be
also expressed in a mathematical form. So, a higher degree of abstraction is possible. For
example, in this particular example that I have shown you that the robot turning towards
the source, if the heading angle is denoted by theta then one can actually write it like a
theta dot, this is theta dot or that is so to say d theta d t that is the rate of change of the
heading angle. So, theta dot is equal to some kind of a function of theta.
So, theta dot is some kind of a function of the heading angle. So, rate of change of
heading angle is a function of the heading angle itself. And, if you look at it that for
example, in this equation this is a non-linear this can be a non-linear equation.
And, in this equation there is a fixed point here and if the robot goes towards the left side
of the fixed point then there is a positive turning rate which will take it towards the fixed
point. If it goes towards the right side of the fixed point then there is a negative turning
rate which will take it towards the fixed point.
309
So, the turning rate of the vehicle will be such that it will be always guided towards an
attractor, this attractor is very important. So, this is for a single source system, this kind
of a taxis vehicle model has one attractor and it has to always go into that attractor. Their
presence of the attractor also tells us that there is a presence of stability in the system.
Stability is a very important concept for any intelligent system and for a system to be
enormously intelligent; we will later on see that it has to have multiple attractors which
should give multiple stability configurations to the system. So, this is an important you
know kind of mathematical abstractions that is possible through the dynamic field
theory.
Now, let us consider a situation where you have not just one source, but you have two
sources. Again in this case because you have two sources; so, you will be having
essentially two fixed points. And, the logic will be such that depending on the initial
orientation of the robot you will either go to one fixed point or you will go to another
fixed point and there exists a plane in between which will work like a repellor.
So, it will drive the system towards one of the fixed points. So, that is once there are
multiple sources then there are multiple points of stabilities. Now, what if the source 1
and source 2 are very close? In that case it may happen that these two bimodal system
will be so close that you can apply a mean field theorem and by using the mean field
310
theorem you can actually consider a unimodal system. Again in this case you will be
having a single fixed point towards which the system will be driven continuously.
And, this whole thing can be beautifully described with respect to this kind of a you
know bifurcation diagram which I had earlier discussed with you, that there are such
non-linear bifurcations that are telltale signs of an intelligent systems.
That if the sources are too close then there will be a single attractor system; this is a
single attractor system. So, we will be in this range and the moment the sources are far
apart the distance crosses a critical point that critical point is of course; this is the critical
point of bifurcation.
Then you are we will be going to get two attractors and in between there is this repellor
plane which is this repellor plane I have talked about. So, and it will become a bimodal
kind of a solution. So, depending on the separation distance between the sources we are
going to see this intrinsically complex behavior from an apparently very simple system
that it is either going to go to one single solution or it is going to go to two solutions, two
points of stabilities. And, in this case it will be depending on the initial orientation of the
system.
So, this is you can see that how this same model can develop more amount of behavioral
complexity. We have seen the Schoner’s model of dynamic field theorem and we have
seen that in these field theorem how the embodiment is absolutely important; you must
have the environment, you must have the sensor, you must have the actuator, the nervous
system connecting them and the body in which the whole thing must be dwelling.
Taking these further we will see Rolf Pfeifer’s and Josh Bongard’s and they are those
point of view from MIT that how the body can control the mind itself or so to say the
intelligence itself how that can be governed by the body. So, let us look into their
propositions in terms of intelligent system.
311
We will now look into the body shaping cognition which was developed by Rolf and
Rolf Pfeifer and Josh you know in MIT. Now, here said that the intelligent agents as I
told you that agent would mean any intelligent system living or non-living. Now, they
should exploit their ecological niche that is what one of the proposition of from the Allen
Newell and they should exhibit diverse behavior following certain soft and hard rules.
So, the compliance to the diverse behavior is something very important for an intelligent
system.
Now, these diverse behaviors a part of it is actually the earlier soft rules that Minsky,
Newell’s and Anderson’s their can. And, one of them is that the robot should be able to
use natural language and because that gives enormous flexibility in terms of
communications and interactions. So, soft rules are that the mode of interaction should
be as diverse as possible and the hard rule is that all intelligent agents must comply with
the laws of physics.
Now, this directly means that there must be this intelligence embodied in a real system
which is subjected to the laws of physics. For example, the robot is subjected to the laws
of gravity. It is subjected to the laws of you know whenever we will be talking about
motors etcetera, laws of electromagnetics or other laws like for smart materials we have
discussed the piezoelectric laws of piezoelectricity.
312
So, all these laws of physics it has to comply; there is no escape from it. You can only
escape it in a virtual world, but for a real robot you cannot escape it. So, that is the first
important thing that we have to keep in our mind.
Secondly whenever we will talk about these robots, there has to be a frame of reference
that is important for us the frame of reference. The agents can exploit physical laws even
if they are not aware of them. In this sense intelligence is something which may not be a
property of an agent, but the person is observing the agent.
Now, there is a very famous analogy there that if you consider a ant moving by the side
of a beach which is having lots of poodles and small mouths etcetera. Now, you may see
once the ant has moved you may see that it has at the end when it reaches from
destination a to destination b; you may see that the ant is having a very complex path,
this is by the side of the sea.
So, you may see a very complex path, but you know all that the ant must have had done
is that if there is an obstruction at a particular position, it has always turned to the other
direction, it has changed itself just like this Rolf just like these other Braitenberg’s taxis
device. So, just by following the algorithm of the obstructions it must have had actually
develop this kind of a complex path.
313
And, hence the intelligence in the ant from the ant’s point of view may not be that high,
but it is from person is watching this ant path it is a very complex path. Behavior in
animals and robots cannot be reduced to internal mechanisms alone.
When you are looking at the behavior you are looking at their complex interactions with
the environment. So, behavior is the result of an agent interacting with the real world
which includes not only the agent’s neural system, but also its entire body because
without body it is not interacting.
How the sensors are distributed, what is the material properties of the muscle tendon
system and the joints. Remember that is why I told you that for the new age robots you
need to know how the smart materials, the smart muscles are going to behave; you
cannot exclude that fact if you have to develop an intelligent robot.
And, seemingly complex behavior can result from very simple neural mechanisms or
simple control programs. In fact, I have shown you the taxis vehicle example where a
kind of a complex behavior is actually happening; although the logic was very simple.
So, the frame of reference is important that are you watching it from the robot’s point of
view or are you watching it from outside.
Now, there are certain characteristics which rob said that an intelligent agent should be
having. First of all, it must obey as I had told you many times that it must obey the laws
314
of physics like energy dissipation, friction, gravity. It generates sensory stimulation
through motion and generally through interaction with real world. This generation of
sensory stimulation is very important and through interactions it should generate it. And,
this we have to also keep in mind that suppose there is a living system, suppose there is a
living dog moving on your lawn.
Once the dog has moved then there are the footprints of the dog on the lawn. So, there it
will affect the environment also that we have to keep in our mind. And, these are
complex dynamical systems which when they interact with the environment it will have
attractor steps.
I have shown you simple abstractor steps right, if you remember in terms of that
unimodal system. So, there can be many attractor steps. In fact, for animal locomotion
this is also known as gaits. So, there can be many gaits on the system and these are
always the result of interaction of three systems: agents body, its control system or brain
and its environment.
So, that would define what are these attractors are and the more the number of attractors
the more complex is the system. The other important thing is that they must perform
certain morphological computation. Now, by morphological computation what we mean
is that for example, if you consider the human leg; the muscles and tendons are elastic so
that the knee, when the leg impacts the ground while running performs all small adaptive
movements without any neural control.
So, you know you do not always need neural control, you may say that this is something
which is like a behavior which is given to the system a priori. So, that it can actually
work without interference by a control system. And, it is the environment which
invariably always closes (Refer Time: 44:30) that is the characteristics of intelligent
agents. Now, we have looked into the embodied intelligence, the hard intelligence and
you know an aspect of embodiment of intelligence and also we have looked into the soft
rules.
Now, we are going to kind of assemble these things together and we are trying to
propose some kind of a cognitive architecture. So, I will just give you one such example;
in literature there are many. I have chosen one which is a very popular example of LIDA,
a LIDA system.
315
And, that is what we are now going to look at that how can be how a cognitive
architecture can look like. And, you would see that this is quite different from what we
generally do in robotics control, the cognitive architecture system. So, let us look into
this LIDA cognitive architecture system.
LIDA which is also known as Learning Intelligent Distribution Agent, it is having two
important characteristics. First of all, it works in a cycle and it is roughly is like a
clockwise way that is shown here, but there is a cycle. You will see in all intelligent
systems there is a rough kind of a you know approximate time cycle that always is
involved. And secondly, it has three important phases: understanding phase, attention
phase and acting and learning phase.
So, it has these three important phases and the entire system is having many so, to say
you know the according to Anderson’s these productions kind of you know or the agents
you know which carries out many specialized actions. So, first of all such a system
would work with a sensory memory and then this sensory memory will send the data, we
will call it in form of queues to something which is known as Perceptual Associative
Memory or PAM.
Now, PAM also continuously sends its kind of you know directions to the sensory
memory. If you remember that A brain and B brain structure if you remember it there
that is something similar and this is the priming that is happening.
316
So, PAM works something like A brain which interacts through sensory memory to the
external environment. So, this is where is the internal and external environment and then
PAM with the help of certain codelets which are like structure building codelets, it
actually develops the current situational model.
Now, when it is developing the current situational model, it also takes care of the
conscious contents queue so, that it can update itself. Then from the current situation
model and at this stage the current situational model can also get updated by the spatial
memory and the transient episodic memories. And, everything can get consolidated in
the declarative memory and then certain aspects of these are chosen for the global
workspace through the attention codelets. So, this is another kind of a procedural
memory.
So, it comes to the global workspace and finally, from global workspace the actions
other procedural memory starts to work and the action selection and the sensory motor
memory comes into picture. And finally, the motor plan gets executed which works on
the environment. Now, there is this blue part which is preconscious level and there is this
red part which is happening in the conscious level.
Let us now look into it with a little more details. So, as I told you that cognitive cycle
hypothesis is one of the most important feature of LIDA. It claims that there is a
317
continuous iteration of similar you know flexible cognitive cycles which would
approximately take 200 to 300 milliseconds.
So, it is not exactly a clockwork of a computer, but it has some kind of a cycle in which
it continuously works on. And as I told you that agents life consists of continuous
sequence of three activities: understanding, attending and action selection.
So, these three activities have to happen continuously in the system. Now, these three
phases of LIDA; let us look into it one by one again. The understanding phase which we
have just discussed; so, in this phase the cycle begins with sensor stimuli from sources in
the agents external and internal environment. And, they start to recognize certain low
level features and then pass the activation to higher level feature such as what object is it,
what category, what is the relationship, what are the events, what are the situations
etcetera.
And these are done under the constant monitoring of nodes which are known as
perceptual associative memory or so to say PAM. So, you remember the PAM here just
this is where is this PAM part. So, this is continuously interacting with the sensory
memory.
Now, features recognized precociously, this is the preconscious phase and represented by
PAM node structure make up the percept that is passed asynchronously to the
318
workspace, where a model of the agent’s current situation called the current situational
model is assembled or updated.
So, you remember that this is where is the CSM or the Current Situation Model. Now, at
time 0 there is always a model that is present and the model gets updated by the sensory
memory and the PAM. So, the model continuously gets itself updated and this is done
with the help of the structure building codelets.
So, a new model of the agents’ current situation is assembled from the percepts
associations. These associations come from other may come from other spatial
memories, episodic memories etcetera. And, the undecayed parts of the previous model
and this assembling process typically actually takes the help of structure building
codelets. It actually assembles or updates the you know this particular the current
situation.
Then these codeletes also continuously monitor the workspace for opportunities to fulfill
their particularly specified task, because codelets are something like an agent. They may
draw upon perceptual memory and even sensory memory to enable the recognition of
relations and situations, analogies and similarities.
Now, this newly assembled model constitutes the agents understanding of its current
situation within its world. And, it has made a sense of incoming stimuli that is important
319
whatever is reaching you, it will make a sense of it and the understanding phase is
complete. So, that is the first phase.
Next we come to the attending phase where the entire model is not presented because the
entire model could be very complex or it can be very rich in terms of information. So,
one has to choose that where the attention is needed. So, only a portion of the model is
actually choosing. And, special purpose attention codelets they bring certain perceptual
structures of concern to the particular attention codelet into the global workspace.
And, a representation of the contents of the winning coalition because there will be many
coalitions, but the winning coalition then is broadcast globally bringing its contents to
consciousness. So, this is where the consciousness starts and that completes the attending
phase of the cycle. So, understanding attending is over.
320
Now, we are conscious because we have to do the action. So, now comes the action and
the learning phase. One major purpose of all this processing that has happened so far is
to help the agent choose what to do next, the other being the several forms of learning
also.
The primary recipient is procedural memory which stores templates of possible actions it
is having that if this happens, this action is taken including their context and possible
results. So, templates whose contexts intersect sufficiently with the contents of the
conscious broadcast instantiate copies of themselves with their variables specified to the
current situation. These instantiations are passed to the action selection mechanism
which chooses a single action from one of these instantiations.
The chosen action then goes to sensory motor memory finally, where it is executed by an
appropriate algorithm or a motor plan. And, the action taken would affect the
environment as I told you earlier and also it will affect the internal representation or both
and the cycle gets complete, you start the next cycle again. So, this is how the whole
system according to Franklin and their group the LIDA cognitive architecture is.
So, we can build similar cognitive architectures for a cognitive robot. So, this is where
we will end today’s lecture.
Thank you.
321
Lecture - 13
Good morning students, welcome to the course of Foundations of Cognitive Robotics. In

this final week of the third lecture, I am going to introduce you to a little bit of a
mathematical concepts of how the neurons develop networks among themselves. You see
in the last few lectures while discussing about intelligence and intelligent system
development, there was one concept that we had continuously discussed that started from
the Minkowski’s concepts of agents and agency development.
And these agents they themselves are nothing, but in the terms of the brain, they are
nothing, but the group of neurons and similarly of course, from the other side for cognitive
robotics perspectives one can develop a group of artificial neurons in a networked system.
Now what is important here is that each neuron having its own way of responding to the
environment which is in terms of as we have earlier studied in terms of firing sequence of
the neuron, it can be modeled as an oscillator we have seen that.
Now, if there are group of neurons which are connected; that means, there are actually a
group of connected oscillators together what will be the characteristics or what will be the
behavior of such group of neurons when they start to fire and oscillate together.
Is there any mathematical model possible so, that we can you know see that whether there
are concepts like synchronous or totally incoherent working things like that whether it is
possible from the behavior of the neurons, that is what we are going to look into today
through certain classical models.
So, let us look into the behavior of group of neurons the mathematical model of such kind
of a system.
322
The outline of this lecture is that I will first talk about the concept of synchronization which
is very important; that means, as I told you that a group of neurons working like oscillators
do they synchronize or not what do we mean by synchronization? I will talk about some
of the background models that has been developed in order to explain this concept of
synchronization and then I will talk about classical Kuramoto model which is the standard
model in terms of such neural behaviors.
We will also touch about the application of the Kuramoto model and the neuronal
synchronization we will talk about some modifications that are possible in the Kuramoto
model and some of the other variations of the Kuramoto model. So, that is the outline of
the talk.
323
Now let us first discuss about that what we mean by what we call the synchronization
phenomena. So, we have talked about synchronization.
Now, this word syn means same and the chronous part of it; that means, the time. So,
synchronization refers to that about the oscillators a set of oscillators which has a similar
timed response sharing the same time; that means, their response is very much the same
and this would actually in our case it would refer to what we will be calling that adjustment
of the rhythms in any coupled oscillatory system and that is possible due to weak
interaction between the phase oscillators.
So, if each one of the oscillators like this is one oscillator and this is another oscillator,
now they if they are weakly linked through this system they are weakly linked it is possible
that under certain conditions they will develop synchronization in their response. Now this
is not only possible in terms of a mechanical model, but also in terms of many natural
events like synchronous flashing of fireflies.
It’s a very interesting one I will show you and then cardiac pacemaker cells or say
synchronization of the power grids, there are many such applications where you will see
that the similar phenomena happens in the system. This is the phenomena which was first
studied by Huygens in the mid of 17th century even though you know this kind of
phenomena like synchronous flashing of fireflies etcetera they were quite ubiquitous in
nature.
324
However, a systematic mathematical study only started in about last 50 years or so, starting
from Winfree and then Kuramoto model.
Now let us look into the synchronization in nature first. So, here what I am going to show
you is the mutual synchronization of certain biological oscillators. For example, in this
case if this is about the fireflies, but similar things you can also you will be able to
appreciate in the networks of neurons when we will discuss particularly in the context of
circadian pacemaker and hippocampus.
And also you will see similar concepts in insulin secreting cells of the pancreas and crickets
that chirp in unison you will see particularly this monsoon season will be coming up and
you will see suddenly the crickets are all chirping at times in unison. So, these are all
examples of synchronization of different types of synchronization in nature or
synchronization in manmade systems.
Now, if you look at this particular system for example. So, this is a particular case in which
you know this happens in actually you know in Thailand and there are these you know
fireflies in a jungle and if you look at the video what you would see is that there are laser
pointers which are actually kind of oscillator. So, they are oscillating you know or LEDs
they are oscillating and the fireflies respond to that oscillation and they start to
synchronize.
325
So, let us look into this video and then we will be able to see that you know what is the
significance of this kind of a system is.
So, let us look into the video. As you can see that this is a remote village area and you can
see that there are lots of these fire flies along with there will be some lights which will be
oscillating in nature.
326
And you can already see that there is these lights which are oscillating in nature all around
you can see the oscillatory lights and if you wait a bit. So, this is the oscillatory LEDs.
327
You would soon see that the fireflies they would start to respond to this oscillatory light
now you can see them.
328
All around they are oscillating and very interestingly if you wait for some time you would
see that their oscillation frequency is very much the same as the LED induced oscillation
frequency.
329
So, this is an example of oscillation synchronization in nature you can very clearly see it.
Now you have seen that how in nature you know the fireflies they start to synchronize with
respect to some kind of an induced excitation you know frequency of light. A very simple
example you would see I will show you from one of the UCLAs you know laboratory
demonstration on metronomes. Metronomes are the systems which are used in order to
measure the frequency or to drive an oscillator at a particular frequency.
330
Now, in this experiment there will be many such metronomes and these metronomes are
weakly coupled over a plank which is supported in a very you know movable manner ok.
And you would see that as all the metronomes will initially be at their own frequencies
own characteristic frequencies and you will see that what happens after a few minutes are
passed. So, let us look into this example also.
So, in this example this is from UCLAs department of physics and astronomy let us look
into this example.
331
As you can see that there are metronomes here, each metronome is actually excited at their
own frequencies and initially they are all out of phase you can see one is going right
another is going left etcetera and you can also see that they are weakly coupled because
this whole base is moving.
Now, as you can see gradually you see all of them are going towards the same direction at
the same time. So, which means that they have actually synchronized. So, imagine that
each one of these metronomes we are talking about each one of them is a neuron and
imagine each one of these neurons are connected by some kind of a weak link then this
weak link can actually induce a situation based on certain conditions where all the neurons
will be synchronized to each other. So, that is the problem that we are going to look at.
332
Now, there is a timeline as I told you the evolution of the study 1958 this was first studied
by Weiner mathematically we will soon see Weiner’s way of you know putting this he
recognized the ubiquity in the natural world and speculated that it was involved also in the
generation of alpha rhythms in the brain. So, that was the for the first time that somebody
speculated that the source of alpha rhythm in the brain maybe that is because of neuronal
synchronization.
Now, later on Winfree in 1967, he has formulated the problem in terms of large population
of interacting limit-cycle oscillators. So, he has considered that problem of each one of
these oscillators and they are weakly coupled. So, how they will exchange the energy with
each other that was the problem that he has actually considered weakly coupled oscillator
problem.
And afterwards in 1975 Kuramoto came out with a better model of the oscillator and that
has been globally accepted in the scientific community where he has shown actually that
these oscillators when they were run at arbitrary intrinsic frequencies to begin with and
they are essentially coupled through the sine of their phase differences.
And based on certain conditions and coupling at the degree of coupling etcetera it may be
possible that there will be a synchronousness or partial synchronousness that will be
developed in the system. However, Kuramoto did not address the stability problem and
333
which has been later on addressed by Strogatz and he has also worked on the partial
synchronization of the phases that was also another contribution by Strogatz.
And he has shown that the experimental results are more matching with respect to the
certain connectivity which we are going to discuss that Strogatz has explained. If we
consider that type of connectivity, then actually the time for synchronization matches with
the experimental results. At a subsequent phase the concepts of network science were
introduced by Watt and Strogatz in 2003.
And we have seen such works like in 2010 by Gardenes where these you know modified
Kuramoto model is used for understanding synchronization in the corticocortical network
of cat’s brain. So, we can see that the mathematical concept is getting used in terms of
modeling the network of neurons in brain.
Well we have seen the timeline of development of this type of you know modeling and it
took as I told you that nearly from 50 to 70 years to reach into the stage in which we are
today. So, let us start first with the Winfrees model and then to the Kuramoto model and
see the mathematical modeling how it gradually evolved.
So, as I told you that Winfree was the first to develop the mathematical model of the
oscillators which are coupled to the collective rhythm. So, that was Winfree model and
this is analogous to a mean field approximation in physics.
334
So, what is mean field approximation? Well what it says is that if there are you know
certain variables which are random in nature, then the simplified approach to actually
model this is in terms of the mean of the variation and that is why in this case also the
oscillators oscillating frequency even though it is random in nature, but by applying similar
to the mean field theorem you can think of an average oscillating frequency in the system.
Now here he has developed certain interesting things for example, the oscillator is actually
dynamics is described in terms of the phase of the oscillator. So, these are the phases
𝜃̇𝑖 = 𝜔𝑖 + 𝐾 (∑ 𝐴𝑗𝑖 𝑋(𝜃𝑗 )) 𝑍(𝜃𝑖 )

𝑗=1
And K is actually the coupling strength because if you remember that particular example
where the oscillators the metronomes are connected through a wooden plank, now
depending on the fixity of the wooden plank the coupling will be there.
So, that is what is the coupling strength and 𝐴𝑗𝑖 is a measure of the communication capacity
between ith and jth oscillator and 𝑋(𝜃𝑗 ) is the measure of the influence of other oscillators
on ith oscillator that is 𝑋(𝜃𝑗 )and 𝑍(𝜃𝑖 ) is actually the sensitivity of the ith oscillator and in
this model he has considered N number of oscillators which are networked together.
Now this model was not widely accepted by the scientific community simply because it
lack certain symmetries like symmetries of translational invariance what happens if the
phases are shifted if theta is shifted from theta to say theta plus theta 0 under such condition
the model we need to remain invariant.
So, these are something that this kind of a model was not able to show. So, even though
he has understood the crux of it that the oscillators are to be described in terms of its rate
of change of phases.
But the relationship is not mathematically very rigorous in this case which has been later
on improved by Kuramoto. Now the classical Kuramoto model which is a very successful
model of the same system this was attempted by Kuramoto in 1975.
And this was very well acknowledged in the community as you can see that the relationship
is very much the same here also we have these rate of change of the phase
335
Kuramoto represented a large number of oscillating subsystems by the set of 𝑁 coupled
differential equations:
𝑁
𝜃̇𝑖 = 𝜔𝑖 + ∑ 𝛤𝑖𝑗 (𝜃𝑗 − 𝜃𝑖 ), 𝑖 = 1, … . , 𝑁

𝑗=1
𝐾
𝛤𝑖𝑗 (𝜃𝑗 − 𝜃𝑖 ) = sin(𝜃𝑗 − 𝜃𝑖 )
𝑁
[𝛤𝑖𝑗 : phase interaction function]
So, what it is telling is that the rate of change of the phase not only depends on the natural
frequency of every oscillator, but also the effect of other oscillator on this oscillator.
And this is defined in terms of parameters like coupling strength K, the number of
oscillators that we are considering, the more the number of oscillators the lesser will be
this effect and the sin(𝜃𝑗 − 𝜃𝑖 ) which is essentially the relative phase this is the measure
of the relative phase with respect to the particular oscillator that we are talking about. So,
this is a relative phase that we have to keep in our mind ok.
So, this is the you know very simple, but so, to say elegant description of the dynamics of
a group of oscillators which are weakly coupled.
Now, if you look at these you would see that as K essentially increases. So, if you go back
these coupling constant as it is increasing, you will be finding that its effect is more coming
on the dynamics of the system.
So, you will see that the dispersion the basic dispersion of natural frequencies omega i
could be overcome due to the increase of K and that will actually result in terms of
coherence in the system. Initially if it is incoherent because of high K it will gradually
become partial coherence and then it will gain full synchronization.
336
Now, the other important point is that in order to quantify this degree of synchrony we can
consider a centroid vector. Once again apply the mean field theorem a centroid vector
which has a magnitude of r and a phase which is 𝜓. Now naturally this depends on the
average response as I told you by using the mean field theorem.
As K increases, the interaction functions overcome the dispersion of natural frequencies

ωi resulting in a transition from incoherence, to partial and then full synchronization.
To quantify the degree of synchrony, centroid vector of this phase distribution is

considered as,
𝑁
1
𝑟𝑒 𝑖𝜓
= ∑ 𝑒 𝑖𝜃𝑗 … … … … … . (1)
𝑁
𝑗=1
Here, 𝜓: mean phase of set of 𝜃𝑖 ,
𝑟: order parameter, captures the degree of phase coherence in the system
Here r(t) with 0 ≤ r(t) ≤ 1 measures the coherence of the oscillator population, and ψ(t) is
the average phase.
Now, a very nice way. So, first of all r equals to 0 would mean that all the oscillators in a
phasor diagram they are arbitrarily spread and r equals to 1 means they are all very closely
spaced as you can see here. Now Foncesa and Abud in 2018 has given a very nice
description of the system.
337
He has considered 1000 neural oscillators and have chosen 25 of the neural oscillators
arbitrarily from them and then he has plotted their phase rate of change of the phase
corresponding to different coupling strength as you can see that there is coupling strength
unity and then gradually increase it to 2.1 then to 3.5 and then to a very high value order
of magnitude at 35.
Now what you can see is that initially the phase distribution is quite a big you know rate
of phase distribution, but gradually you can see as you are increasing the coupling strength,
there is a grouping that is happening. As you have increase the coupling strength further
you can see that there are distinct groups most of the groups are somewhere here whereas,
there are groups here there are there is a group here there is a group here.
And if you are increasing it to even more to K equals to 35 astoundingly that even if
initially they are at a very wide you know natural frequency distribution, they will be all
coming their instantaneous their instantaneous frequency of response that frequency will
be very close to each other. So, that is what is very interesting. So, that you know has been
studied by Foncesa and Abud that is one of the beauty of the Kuramoto model.
Now, if you further explore these mean field approximation of the Kuramoto model, then
we can think of that this is the coupling Kij is in some sense the average of the all coupling
and which has to be positive in nature.
The original analysis of synchronization was accomplished by Kuramoto in the case of

mean-field coupling, that is taking Kij = K/N > 0
 Multiplying both sides of eqn. (1) by 𝑒 −𝑖𝜃𝑖 , we get
𝑁
𝑖(𝜓−𝜃𝑖 )
1
𝑟𝑒 = ∑ 𝑒 𝑖(𝜃𝑗−𝜃𝑖 )
𝑁
𝑗=1
 Equating imaginary parts,
𝑁
1
𝑟 sin(𝜓 − 𝜃𝑖 ) = ∑ sin(𝜃𝑗 − 𝜃𝑖 )
𝑁
𝑗=1
𝜃̇𝑖 = 𝜔𝑖 + 𝐾𝑟 sin(𝜓 − 𝜃𝑖 ) , 𝑖 = 1, … . , 𝑁
 Phase 𝜃𝑖 seems to evolve independently from each other here, but interaction
is set through 𝑟 and 𝜓.
 Effective coupling now proportional to the order parameter r, creating

feedback relation between coupling and synchronization
338
So, you can see that even though in the last equation it was in terms of summation of some
sinusoidal by using this average phase psi and with respect to that the deviation we can get
an even simpler relationship by using the mid field approximation of the system.
1
 Order parameter given by, 𝑟𝑒 𝑖𝜓 = 𝑁 ∑𝑁
𝑗=1 𝑒
𝑖𝜃𝑗
can be rewritten as
𝜋 𝑁
𝑖𝜓
1 𝑖𝜃
𝑟𝑒 = ∫ 𝑒 ( ∑ 𝛿(𝜃 − 𝜃𝑗 )) 𝑑𝜃
−𝜋 𝑁
𝑗=1
 In the limit of infinitely many oscillators, they may be expected to be
distributed with a probability density 𝜌(𝜃, 𝜔, 𝑡).
𝜋 ∞
𝑟𝑒 𝑖𝜓 = ∫ ∫ 𝑒 𝑖𝜃 𝜌(𝜃, 𝜔, 𝑡)𝑔(𝜔)𝑑𝜃𝑑𝜔
−𝜋 −∞
 If 𝐾 → 0, 𝜃 ≈ 𝜔𝑡 ⇒ 𝑟 → 0 𝑎𝑠 𝑡 → ∞ (Reimann-Lebesgue Lemma)
(oscillators not synchronized)
 If 𝐾 → ∞, 𝜃𝑖 ≈ 𝜓, oscillators synchronized to their mean phase, 𝑟 → 1

 For intermediate couplings, 𝐾𝑐 < 𝐾 < ∞, part of oscillators are phase locked
𝜃̇𝑖 = 0, and part are rotating out of synchrony with the locked oscillators-
state of partial synchronization (𝟎 < 𝒓 < 𝟏).
Stationary synchronization for mean field coupling
 A continuity equation can be written for the oscillator density as,

𝜕𝜌 𝜕
+ (𝜌𝑣) = 0
𝜕𝑡 𝜕𝜃
Where 𝑣 = 𝜔𝑖 + 𝐾𝑟 sin(𝜓 − 𝜃𝑖 ) is called angular/drift velocity.

 The above equation solved together with an appropriate initial condition and
the normalization condition gives:
𝜋
∫ 𝜌(𝜃, 𝜔, 𝑡)𝑑𝜃 = 1
−𝜋
1
 Trivial stationary solution: 𝜌 = 2𝜋 , 𝑟 = 0, corresponding to an angular
distribution of the oscillators with equal probability in the interval ሾ−𝜋, 𝜋ሿ ⇒
Incoherent solution.
339
Stationary synchronization for mean field coupling
A simple solution corresponding to oscillator synchronization

 A typical oscillator moving with velocity v will become stably blocked at an angle
such that
Kr sin(θi − ψ) = ω and − πൗ2 ≤ (θ − ψ) ≤ πൗ2
 Oscillators with frequencies satisfying ȁωȁ > Kr can’t be locked, their stationary
density obeys 𝛒𝐯 = 𝐜𝐨𝐧𝐬𝐭𝐚𝐧𝐭
 This is a stationary state of partial synchronization.
The corresponding stationary density is therefore,

𝜔
𝛿 ቀ𝜃 − 𝜓 − sin−1 ቀ ቁቁ 𝐻(cos 𝜃), ȁ𝜔ȁ < 𝐾𝑟
𝑘𝑟
𝜌=൞ 𝐶𝑜𝑛𝑠𝑡𝑎𝑛𝑡
, 𝑒𝑙𝑠𝑒𝑤ℎ𝑒𝑟𝑒
ȁ𝜔 − 𝐾𝑟 sin(𝜃𝑖 − 𝜓)ȁ
Equivalently, 𝜌 = ξ𝐾 2 𝑟 2 − 𝜔 2 𝛿(𝜔 − 𝐾𝑟 sin(𝜃 − 𝜓))𝐻(cos 𝜃)

for −𝐾𝑟 < 𝜔 < 𝐾𝑟
Now, order parameter in the state of partial synchronization,

𝜋ൗ ∞
2 𝜔
𝑟 =∫ ∫ 𝑒 𝑖(𝜃−𝜓) 𝛿 ቀ𝜃 − 𝜓 − sin−1 ቀ ቁቁ 𝑔(𝜔)𝑑𝜃𝑑𝜔
−𝜋ൗ2 −∞ 𝑘𝑟
𝜋
𝐶𝑔(𝜔)
+∫ ∫ 𝑒 𝑖(𝜃−𝜓) 𝑑𝜃𝑑𝜔
−𝜋 ȁ𝜔ȁ>𝐾𝑟 ȁ𝜔 − 𝐾𝑟 sin(𝜃𝑖 − 𝜓)ȁ
Second term vanishes because of symmetry of 𝑔(𝜔) and the first term becomes
𝜋ൗ
2
𝑟 = 𝐾𝑟 ∫ cos2 𝜃 𝑔(𝐾𝑟 sin 𝜃)𝑑𝜃
−𝜋ൗ2
Two solutions, one 𝑟 = 0 for incoherence and second branch of solutions, for partially
synchronized phase:
𝜋ൗ
2
1 = 𝐾∫ cos2 𝜃 𝑔(𝐾𝑟 sin 𝜃)𝑑𝜃
−𝜋ൗ2
This branch bifurcates continuously from 𝑟 = 0 at a value 𝐾 = 𝐾𝑐 obtained by setting
𝑟 = 0,
2
𝐾𝑐 =
𝜋𝑔(0)
340
For a general frequency distribution 𝑔(𝜔), an expansion of right-hand side in powers of
𝐾𝑟 yields the scaling law
8(𝐾 − 𝐾𝑐 )
𝑟~√
−𝐾𝑐 3 𝑔"(0)
The partially synchronized phase bifurcates super-critically for 𝐾 > 𝐾𝑐 if 𝑔"(0) < 0 and
sub-critically for 𝐾 < 𝐾𝑐 if 𝑔"(0) > 0
We have seen the mathematical modeling of the basic Kuramoto model. Now let us look
into some of the applications of this model specifically in the context of applying these for
the neuronal networks.
If we look at the application of these kind of Kuramoto model, as I told you earlier itself
that there are many field where we can apply this concept.
For example, for power grids for networks of disordered Josephson junctions for
modelling of self other integration. Now self other integration means that when there are
of number of agents these agents may be some living and some non living so, but a number
of agents and they are interacting with each other.
So, it is a self and, but they are to be you know agents which are having the cognitive
capability and there you can get a self other integration during the synchronous ness of
these kind of oscillators and of course, the neuronal networks itself and also sometimes
fields like seismology can get benefited by the same Kuramoto model.
341
Now, if you look at the Kuramoto model for the neuronal network. As you can see that we
have discussed about this particular figure many times that this is a single neuronal the
dendron’s this is the soma part of it and this is where is the axon part.
And you also know that there is this nodes of Ranvier there and this is this parts are
myelinated that will increase the speed of the signal travel and then the signal comes to the
axon terminals from here it starts to actually fire the chemicals which is absorbed by the
next dendrite.
Now, the question is that clearly each neuron there is an oscillator. The question is that can
they also have a kind of a synchronousness as has been predicted long before well indeed
this is a possibility, I encourage you to look into this particular you know biological neural
model.
Now collective synchronization of the Kuramoto model is certainly possible for this kind
of neuronal systems and the dynamics of the relative phase variables of the oscillator plays
an important role here and the order parameter R versus the coupling intensity this also
plays an important role in this kind of a system.
342
Modified Kuramoto model to study the synchronization between neurons
𝐾
𝜽̇𝒊 = 𝝎𝒊 + ∑𝑵𝒋=𝟏 𝜞𝒊𝒋 (𝜽𝒋 − 𝜽𝒊 ), 𝒊 = 𝟏, … . , 𝑵, 𝛤𝑖𝑗 (𝜃𝑗 − 𝜃𝑖 ) = 𝑁 sin(𝜃𝑗 − 𝜃𝑖 )
 ‘All-to-All’ connectivity  Small world Connectivity
 Fixed natural frequency  Time-Varying natural frequency
 Fixed coupling strength  80:20 coupling/ STDP Based coupling
 Time delay
 External stimulation
Time delay: finite conduction speed of electrical signals down the axon
𝑑𝑖𝑗
𝜏𝑖𝑗 = + 𝜏𝑡𝑟𝑎𝑛𝑠𝑑𝑢𝑐𝑡𝑖𝑜𝑛
𝑣
External Stimulus: Models the foreign currents which are applied to the neuron
1
𝜃̇𝑖 = 𝜔𝑖 (𝑡) + 𝑋𝑖 (𝑡)𝑆(𝜃𝑖 ) + ∑ 𝐾𝑖,𝑗 sin(𝜃𝑗 (𝑡 − 𝜏𝑖,𝑗 ) − 𝜃𝑖 (𝑡)) , 𝑖 = 1, … , 𝑁
𝑘
𝑗∈𝜗(𝑖)
Now, if you look at the modifications that are possible in a Kuramoto model, the first
important modification that is needed is in terms of that the connectivity is in considered
in Kuramoto model is all to all connectivity. Whereas, the connectivity that is generally
feasible is actually a small world connectivity which means local neurons are connected
to each other.
The fixed natural frequency of each neuron may not be also realizable. So, you need a time
varying natural frequency and the fixed coupling strength may be changed by 80-20
coupling or STDP based coupling.
And then there is this time delay that happens in the system and external simulation effect
all these things will actually modify the behavior of the neuron. Now if I take the time
delay into consideration the finite conduction speed of electrical signal actually creates the
time delay as you can see that there is a time delay that is involved in the system. Again if
there is external stimulus in the system the that would model the foreign currents. So, this
is where is the external stimulation in the system that can happen ok.
So, there are various such things that we can actually introduce into the Kuramoto model
in order to make it more realistic in nature.
343
Now, if you look at one of the connectivity called the Watts Strogatz connectivity in terms
of a small world network, this essentially tells is that the neurons which are close in the
neighborhood they are actually connected more rather than the neurons which are very far
apart.
So, suppose N are these number of nodes which are there and k is the mean degree of
connectivity of each neuron. So, the probability of each neuron connected with you know
k such neuron on an average ok. So, that is what is the k is the coupling and the k satisfies
the condition that its value is maximum k is N that is all to all coupling and between log
N and N the value of k and beta is a special parameter which satisfies the condition that
beta has to be in between 0 and the unity.
Now, under such a situation you see when beta equals to 0 you have actually all to all
connection every single one is connected. If beta equals to 0.5 you start to develop a kind
of a you know small world kind of a thing where one of the neuron is connected to your
neighbors when beta equals to 1 you can see that almost there is no long distance
connection. So, each one of them are connected to the neighboring neurons. So, there are
various such possibilities that are there.
344
Now, beta equals to 0.15 once again you would see that there are high synchrony clusters
which can appear based on this kind of a small network connectivity. So, a typical case for
which beta equals to 0.15 in a Watts Strogatz graph with N equals to 10 and the degree as
k as 4 you can see that if we would have considered all to all connectivity, then this would
mean that there will be many yellow points yellow points actually are synchronous and
they are distributed all over essentially.
On the other hand, you can see; you can see that there is a distinct banded nature and the
phase locking has actually improved in terms of synchronousness of the neuronal
oscillators.
Other variations of Kuramoto model
There are some other variations possible of Kuramoto model for example, we can
introduce a bit of stochasticity in the model, where the it is not just in terms of the natural
frequency variation, but in terms of this 𝜉𝑖 (𝑡) which is an independent white noise process,
an independent arbitrary input coming into the oscillator system.
Stochastic Kuramoto Model: The mean field model including white noise forces
(accounts for the contribution of different stochastic forces)
𝜃̇𝑖 = 𝜔𝑖 + 𝐾𝑟 sin(𝜓 − 𝜃𝑖 ) + 𝜉𝑖 (𝑡) , 𝑖 = 1, … . , 𝑁
345
 𝜉𝑖 (𝑡) independent white noise processes with expected values
⟨𝜉𝑖 (𝑡)⟩ = 0, ⟨𝜉𝑖 (𝑡)𝜉𝑗 (𝑡′)⟩ = 2𝐷𝛿(𝑡 − 𝑡′)𝛿𝑖𝑗
 The Fokker-Planck equation for the one-oscillator probability density

𝜌(𝜃, 𝜔, 𝑡) is
𝜕𝜌 𝜕 2𝜌 𝜕
=𝐷 2− (𝑣𝜌)
𝜕𝑡 𝜕𝜃 𝜕𝜃
So, you need to add that in the form of Fokker Planck equation in order to get a stochastic
Kuramoto model. This is one possibility the other possibility is that in order to consider a
second order Kuramoto model. So, here the model with frequency adaptation where both
phases and frequency evolve in time and having synchronization that slowed down by
inertia can solve such problems. So, that is a second order Kuramoto model.
 Second order Kuramoto model: The first order Kuramoto model approaches too
fast the partial synchronized state compared to experimental observations, infinite
coupling strength for 𝑟 = 1
 The model with frequency adaptation, where both phase and frequency
evolve in time and having synchronization slowed down by inertia, can
solve such problems.
 𝜃̈𝑖 = −𝛼𝜃̇𝑖 + 𝜔𝑖 + 𝜆 ∑𝑁
𝑗=1 𝐴𝑖𝑗 sin(𝜃𝑗 − 𝜃𝑖 )
So, in a summary in this lecture we have seen that how the dynamics of a group of neurons
can be developed in terms of a set of first order differential equations and can be described
in terms of a physical model called Kuramoto model.
We have also seen that the Kuramoto model has its weaknesses particularly in terms of all
to all connectivity which can be changed by you know small world connectivity this each
neuron is connected with only the neighboring neurons. So, that is a small world
connectivity.
And we have seen that with respect to the small world connectivity you can develop a
better partial synchronization in the system. We have also seen that you can extend in
future this Kuramoto model in terms of many parameters, but some of them that we have
346
discussed are like you know white noise excitation of the system or in terms of the coupling
of the system or delay in the system. So, these are some of the things that in which you
can actually extend the Kuramoto model.
I think we will put an end there, in the final lecture we will see an experiment where we
will show you that how a robot and a kid can actually interact and some kind of a phase
locking can happen in the system.
So, this is where we would like to put an end. I must acknowledge my student Mr. Anurag
Dwivedi who has completed his MTech thesis in this direction recently for many of the
slides.
Thank you.
347
Lecture - 14
Good morning students, welcome to the course of Foundations of Cognitive Robotics.

We have reached the last lecture in this series of foundations in the cognitive robotics;
where we will specially focus on some experiments and experimental efforts that we are
carrying out at IIT Kanpur.
So that, you will be able to know that if you have to develop such robotic systems what
are the general experimental setups that are needed. And also you will get yourself
familiarized with the different types of experiments that are possible in this direction.
So, our focus today will be a few case studies through the experiments related to the field
of cognitive robotics.
Well, we have chosen the first important part in terms of the experiments in the field of
child robot interaction. Now, you know already about Human Robot Interaction or HRI
which is a common, very common you know topics in robotics, where the robots
basically interact with adult human beings. And both ways the robots basically learn
from the adult human beings and also it helps in many cases like for geriatric
348
applications, it helps adult human beings in terms of different types of let us say you
know requirements, functional requirements for elderly people.
The point of child robot interaction is quite unique. Generally, for child robot interaction
we choose children of the age of 3 to 11 and as you can see here that there is an
interaction between a Nao robot and a child so that is taking place. So, the cognitive
development for this type of kids is generally immatured and it is still at the developing
phase. When we have discussed about the human brain, we have talked about this aspect.
Now, one very important thing is that much unlike the adults a child does not foresee a
robotic system as a simple mechatronic device. It actually imposes a significant degree
of human like nature. In fact, we have a term for it, it is called anthropomorphisation in
their mind map. So, the perspective of a child and 3 to 11 year children we are talking
about, the perspective of a child towards the robot is totally different than an adult
human being and that is a very important point that we have to keep in our mind in child
robotic interaction.
Also that this is fundamentally different from the adult robot interaction is, because
children are not just small adults, their neuro physical, their physical and their mental
development are actually ongoing. So, this also makes a challenge in terms of the human
robot interaction. So, that is why the applications that I will show you today in terms of
cognitive robot development will be mostly will be actually in the field of child robot
interactions. So, that is one important.
349
Now, the major components in this interactions you know if you consider that how a
child is interacting with a robot, it will be in terms of different types of sensing both from
the child’s point of view. Somatosensory sensing of the robot as well as the visual and
audio related sensing and that is somewhat true for the robots point of view also.
It should be able to then interact with the audio visual and other types of sensory
systems. Sometimes touch is also included and then the other component of this
interaction is in terms of perception, how the child is perceiving the robot and how the
robot is perceiving, how is it making a differentiation between a child and an adult and of
course, in terms of the interactions.
These interactions you can think of bringing you know aspects like that the robot shows
anger, fear, happiness, and sadness and these how the child is perceiving all these
emotions, emotional perceptions as these things are happening. And of course, these
perceptions we can quantify it in terms of things like valence, arousal, these are all
certain aspects of measuring the emotions and also engagement.
So, from the visual data that the robot gathers we should be able to actually measure this
level of perceptions and accordingly, we should be able to find out the relative
importance of various types of interactions.
350
Now, the objective of these experiments will be first of all identifying this interaction
related parameters that is very important. And develop different models that would
implement smooth and untethered interaction between the subject and the robot. So,
model development; identifying several of the robot design parameters; let us say the
interaction speed, rhythm, interaction pattern whether it is visual, audio, audio based or
somatosensory.
So, identify these robot design parameters that affect child robot interactions and try to
come out with recommendations for suitable modifications to improve the same. And
complement the AI and machine learning based learning models, because in this case the
robot has an embedded intelligence.
Develop and re-explore different cognitive models of child learning and find ways to
integrate the same in the robot development. Interaction design and perform experiments
and perform exhaustive data analytics on the experimental data. So, these are the kind of
objectives that we keep when we design experiments related to child robot interactions.
351
Now, when we talk about interaction what do we actually mean by this word
interactions? Well, it is, if the word carries an extensive meaning, but we would be
mostly focusing on expression and actuation. When I have initially talked about the
smart materials and their you know availability in the robot, in this mechanical actuators
we have in mind like rotation, like motion swinging, etcetera related to robots various
degrees of freedom.
When we talk about expressions these are emotional inputs like joy, anger, surprise, fear,
sadness, disgust, etcetera. This can be given either through auditory or through the visual
cues.
Sometimes you can have a composite action like some dancing sequence that robot may
carry out with a child and which involves both expression as well as actuation and many
times these composite actions may actually form the basis of interaction design while
designing an interactive robot.
352
So, the methodology that will adopt towards this direction is; we will note down some of
the very important child related parameters, if you have to carry out such experiments;
like emotional expression of the child, understanding of the robot, reaction to robot
malfunctions particularly, attention, movement, imitation, Bi-directional
communications.
On the other hand, from the robots point of view; robot action, robot performance, and
robots use characteristics these are very important. From the interaction related
behaviours physical interaction is one important thing, engagement levels and interest of
the interactive interaction from both the sides. So, these are certain things that we
generally try to study in child robot interactions, whenever we carry out such
experiments.
353
There are various novelties in these experiments for example, these are complement to
basically the AI based system. As I have told you that these are embedded intelligence,
so these are complement to AI based system and it helps in developing more conscious
memory in robots. So, using systems like reinforced learning, etcetera; we can also tune
embodiment design parameters through this experiment and we can label of various
cognitive models of learning and development and we can improve the perception of the
robot.
And in terms of long term interactions tutoring, delivering therapy, or treatment of

autisms, spectrum disorder are some of the long term goals that one tries to achieve
through this kind of child robot interaction. Now, when we have discussed about the
entire paradigm, this is really quite broad in terms of the you know its perspective. So, let
us try to see what exactly would be the canvas of this interactions between a child and
the robot that we will be trying to address today.
354
In our canvas, in one side we have the robot which is a sophisticated version of a toy let
us say of course, as I told you that for children. For the children it has a different
connotation, they do anthropomorphize such a system, but we will talk we will try to
figure out about the forms; organic, inorganic, or replicas of real life, functions, inert in
level to the functional level, low function to different types of functions.
From the child’s perspective the age of the child, the development, psychophysical
parameters, these are the things which will be important for us and this will be studying
through experiments related to behavioural studies, eye trackers, and EEG experiments
of which, EEG I have already discussed with you.
The interactions can be in terms of oral and touch and variable modalities may be visual
of course, and what we will try to figure out are the some of the characteristics in terms
of latency, physical properties, modalities goals, etcetera. So, this is the framework in
which we will try to work.
355
Now, whenever we talk about the interactions, there is an entire gradation of interactions
as has been shown. In this particular slide you can see, Edward Hall’s work on a system
for the notation of proximity behaviour there are certain distances that one has to keep in
mind whenever the context of robot to child or robot to adult interaction takes place.
If it is in the public by that we mean that this must be 12 feet plus distance, if it is just a
social interaction then it is generally between 4 to 12 feet, if it is personal interactions
with the robot then it is generally 1.5 to 4 feet, but sometimes these dyadic interactions
they become very intimate with the kid and in that case it will be between 0 to 1.58.
So, the physical distancing would definitely play an important role and one has to decide
that in the very beginning itself. Now, in our experiments, we generally try to keep it at
the personal level and some extent at the social level.
356
Now, another important thing is that when we consider this entire interaction from a
child’s perspective usually, a child will be in a what you call default mode state, default
mode network state and at that stage it is a self referential mental activity.
Now, as you bring a robot for interactions, the salience network develops which actually
switches the child’s attention and it goes to central executive network and this is
cognitively demanding activity. So, that is something that we have to keep, you know
you have to keep it with us.
Now, when we talk about the first experiments we will work on actually aesthetic
experiments and it is found out in literature like some of these trends in cognitive
science. I have given one of the references that aesthetic experiences are somewhat
similar to the default mode networks ok, because this is also some kind of a purposeless
activity and play also when we talk about play that is also a purposeless activity. So, for
play for aesthetic experiences, etcetera we expect that the neural system will be at the
default mode network.
357
So, keeping this point in mind, because we have to know that from where we expect the
sensations to come, the aesthetic content of forms can be designed such that we can try to
have you know bring different types of aesthetic experiences. We can create play
experiences and compare and analyse them with aesthetic experiences, we can use EEG
and eye tracker experiment for this purpose.
EEG, I already told you that it actually measures different you know responses that is
happening in the brain with respect to the stimulus. Eye tracker is specially focused on
the eye the pupil itself and there are certain parameters which I will discuss soon like
fixation time, like the dilatation, pupil dilatation, these are the things that we actually
look at in the eye tracker experiment and both of them essentially helps us in terms of
knowing whether there is enough attention of the kid towards the robot.
Now, we always try to take EEG during free play instead of putting them enclosed inside
an enclosure like for adult age, well that has positive and negative sides. The positive
side is that this is most you know kind of a realistic situation, because whenever a child
is in free play he is not pretentious.
So, he is you know we are expected to find out the default mode network and the other
point is that the flip side is that during the free play the child moves a lot and that
actually collapse many times the signal. So, you may not get a good signal and at that
358
time this eye tracker is very useful wherever you lose signal you can at least keep track
of the attentions etcetera.
Now, generally for engineering experiments we carry it out with greater than three type
of samples statistically, but for cognitive science it is very high it is sometimes ten times
higher than that and that is, because the you know there are a lot of subjectivities that
comes into picture. So, the tradition is that you should do it as many for as many number
of subjects as possible the same experiments before making any meaningful you know
conclusion of that.
Now, the devices that we use generally in our experiments are SMIs for eye tracker
system. So, that is RED 500 that is a pretty standard eye tracker system, then we use
Mobilab Gtech wireless, 8 electrode set and also nautilus wireless 16 channel dry
electrode set and also trigger circuit based on the experiments.
Now, why trigger circuit is needed is; because we are having two different recording
systems EEG and the eye tracker in order to have the same time stamp for both the
instruments we need to trigger them at the same time that is why a trigger circuit is very-
very important for this kind of experiments.
359
Now, when we will be carrying out experiments with eye tracker with the EEG settings?
In this particular case for an 8 channel limitation. I have shown you that what are the
regions that we generally choose. We choose the entire you know the central part the and
in that the frontal and the parietal part we choose and also, because there are these visual
parts there so we choose these occipital parts.
So, there are these 5 points that we definitely try to put and then these three so, these are
total 8 channels and these are based on wet electrodes and based on self preferencing, we
take a notch filter around 48-50 Hertz that actually takes care of the electrical signals
around us, the line frequencies and we also put a band pass filter between 0.1 to 60 Hertz
and the sampling rate is about 250 Hz and we take reference from left mastoid and Fpz
ground.
So, these are some of the things that one has to keep in mind that what will be our setup,
what will be the distribution of electrodes that will be actually carrying out and these are
the things that one has to determine. So, in this case these were our choice of the 8
channels.
360
Now, the experiment that we want to carry out is called an experiment of anticipation.
So, how a child starts to expect certain things to happen and what if that does not happen
in the play? How does it affect his or her behaviour and also how does it bring certain
things like ERPs evaporated potentials for example, P 300 type of elicitations. So, that is
what we try to find out and in this particular case, we have chosen in this experiment 18
children out of which 12 are male and 6 are girls with an average age between 5 years
and 6 months.
So, mostly around 4 to 5 years of age in fact and we have connected it from inside this
IIT community, it was and out of this group finally, EEG was acquired from 9 children,
because in most of the other cases the EEG signal was point. And of course, the signals
were also administered through the eye tracker and the children were invited to the lab
and explained about the game and they were given gifts as an incentive. So, this is the
way we have carried out the design of the experiment.
361
Now, in this particular experiment we have chosen some cartoon characters, because we
know that cartoons are very close to a child’s heart. So, we have chosen 6 cartoon
characters and we have told them to actually find out one from this 6 which is the most
preferred one from the child’s perspective.
And then we start to show these figures many times on the visual screen and tell the child
to count that how many number of times this is happening. So, the child will remain you
know attentive and there will be always an anticipation that whether my favourite
character is appearing or not.
And at the same time we are recording it to the eye tracker about the attention, about the
fixation of the child and the EEGs also telling that what is happening to their brain at that
particular instant. So, we have chosen these six images and we made 10 runs of them and
there was a kind of a 500 millisecond break before the experiment started.
And each one of them there was a 1000 milliseconds between these you know
simultaneous characters like 500 milliseconds character to character and continuously
changing. We bring an oddball paradigm where the child we you know expects that the
character to appear and it does not appear or it appears wrongly. So, that is the oddball
paradigm and what we note here is actually the recording from the eye tracker also
simultaneously.
362
Now, for eye tracker there are some important points we have to keep in our mind
whenever we are using it, because I have not discussed about this with you earlier, that in
the eye tracker it always looks into the gaze points; that means, the points where the
child is really looking at.
If you look at it invariably fine wherever there are some actions happening or the eyes,
etcetera these are generally or the head, these are generally very important gaze points.
And we, the eye tracker starts to collect the data with a certain kind of a sampling rate let
us say if it is 100 Hertz it would mean that 100 individual gaze points per second will be
recording.
And if we find that a series of gaze points are very close; it can be in time, it can be in
space; that means, the child is focusing on something. So, this gaze cluster will constitute
a fixation. Denoting a time period or a space where the child has spent more time, like
you can see that there is more fixation on these you know head part of it ok.
So, that is very important that where exactly the child is paying attention to and this
fixations time this time of fixation like here you can see that the fixation time is 300
millisecond and above for child. So, this is a measure of his attentiveness his or her and
also we measure that how much of pupil diameter is taken place which is in this case the
diameter has increased by 0.2 millimetre.
363
Now, what we found from this experiment is that, there was in none of these final nine
cases, there was any error from the child point of view which means he has computed the
number of times the figure you know appeared in a very correct manner and there was
indeed an increase in pupil diameter for the targets. So, this actually tells that there is an
interest that goes towards the target.
Fixation time on an average was actually more for the targets and there was epochs of
800 milliseconds after stimulus administration was used for EEG analysis and we have
learned through this exercise about the design of trigger circuitry, the P 300 algorithm
itself and that the eye movement can contaminate EEG the realization of that and for
targets the EEG data was very-very useful and that actually tells us a lot about the
characteristics of the kid and the eye tracker information also you found it to be more
reliable in these circumstances.
Now, here what you also notice is that many times while playing this game, the kids will
be using something like a memory chunk which is like a baddley type of memory chunk.
As you know that in general, the short term memory is only between let us say 5 to 9
items, that is 7 items plus minus 2.
Now, the memory, the short term memory can be increased and baddley proposed that
you can increase it in terms of a working memory and that working memory
enhancement is what is this memory chunk development and that is something that we
found that can happen again and again if we play this game. So, some kinds of expertize
gets developed in the child and that actually helps in terms of this memory chunk, you
know type of application by the child in a very effective manner.
So, we have noted what experiment, which is related to a sort of practice that you can
also try if you have all these experimental units related to EEG characterization of a
child. And also related to the study of fixation times, the attention, the area of focus, area
of attention, and the corresponding EEG, nature of EEG pattern in the child. So, these are
this was the focus in this particular style.
Next we will study a game and this is a game of rhythm. We will try to find out through
the game that what kind of a rhythm is most you know attractive to a kid, because
accordingly then you can design the cognitive robot. So, let us look into these
experiments to find out the rhythm.
364
So, we are trying to find out, what are the pleasurable interactions and what is the effect
of these rhythms.
For this case we have chosen more number of EEG setups we have used basically 16
channel, but it is dry electrode system and you have you can see the locations that has
been chosen and you already know that how to actually number this. I told you about the
EEG experiments. So, that is along the vertical axis and along this axis also you have
seen that what are the particular electrodes that we have chosen and both in the frontal as
365
well as in the parietal part and then we have used all this information’s in terms of
building up the model of the kid.
Now, here also we have used a notch filter at 40 to 50 Hertz and a band pass filter
between 0.1 to 60 Hertz, the sampling rate was 250. references left mastoid and the
ground was in right mastoid.
Now, the name of this game we call it as a ding dong game, it is like a pat a pat game,
where the robot says something let us say ding and the child has to respond by telling
dong. Now, the robot can start to play very slowly this game and then it can tell the child
and it can try to increase the speed faster and faster until at a point where the child loses
the track robot.
So, we through this interaction as I told you our purpose is to find out what is the best
kind of the interval level which is good for the child robot interaction. Our rational for
this experiment is that a play has rhythmic components with subtle you know changes.
So, we wanted to see whether there are any behavioural artefacts towards this direction
and such a thing you can see in tennis, today in case of swings or say some hammer
hitting’s or you know anything that involves such rhythmic action.
366
Now, subjects that are chosen here, there are some boys and some girls. So, 9 girls and 9
boys with 18 children is the sort of you know subjects that we have used and the robot
that we have used for this study is a nao robot.
So, let us look into the game first of all we carry out a familiarization phase for the kids
so that we can say that you know. So, that we can get the child familiarized with the
robot, because many times whenever we bring the kid in contact with the robot she starts
to ask questions like can he play certain game with me etcetera.
So, getting familiarized, knowing what it can do or cannot possibly do etcetera is very
important and that is the familiarization phase also we record the EEG just to see that
whether it is in the DMA, default mode network.
367
Now, if you look at that at each and every instant we have gradually reduced the
stimulus, inter stimulus interval by actually 25 milliseconds. So, starting from 2850 and
at every stage reducing it by 250.
We can get the response from the child and we can see here that gradually the child is
taking these interactions and the lowest interaction also you can see here in that terms
beyond which again the robot increases its speed and interestingly as the robot is
increasing its speed. So, the response time also changes for any kid. So, that is very
important and the inter stimulus interval the correct design of it is very-very important.
368
Now, we also put a time stamp and check the eye fixation parameters during these
stimulus progress and for prominent eye there is a fixation period and there has to be a
fixation of the robot eye and fixation on a human eye kind of system. What we have
noticed is that instead of the initial part where it is actually a very slow part of the game.
The game becomes interesting at this high speed level between 5 to 10 to and it is at this
level we have seen inevitably at that particular level there is a sweet spot and invariably
all the children reported that they liked that particular reason in the play itself. Now, this
is where we are showing the response time graphically and what you can see is that
before peak the average response time is only about 0.3.
369
We can see that the average response time before peak is say about 709.59 milliseconds,
for the first three trial blocks and after that the sweet spot starts as you can see that it
actually comes down to below roughly 500 milliseconds level and then again it increases
and the average response time after peak is about 660.70 milliseconds.
So, we can see that after the sweet spot the child gets some kind of a lesson so that the
response time of the child improves, it is like 66.70. So, that is something that is
important for us.
370
Now, here I am showing these three blocks again and I am mapping here. The three
waves separately; that means, the theta wave, alpha wave, and the beta wave. What you
can see is that the theta by alpha ratio or the theta by beta ratio both of them show
sufficient amount of you know to say that un durations.
But for the block between 9 and 10 the theta by alpha becomes an important point and
beyond that you know integrable theta pickup beta type of the thing that actually bring
the this particular you know learning into action.
So, we can see that theta alpha is something which gets prominent you can see in this
region onwards. So, this according to us indicates the possibility that there is some kind
of a learning that has taken place in the child.
Now, let us summarize what we have learnt in this week. We have talked about initially
about the concept of intelligence, where we have talked about Spearman’s two factor that
like the g factor and the s factor theory. We have talked about many other theories and
we have seen there are opposite theories like Sternberg’s triarchic theory, which talks
about analytical creative and practical intelligence and we said that there is no single
definition of intelligence you know that can be expected today.
In fact, broadly speaking it is from two perspectives one is from the psychologist point of
view and another is from the AI’s point of view and where we have shown some of these
371
intelligence definitions on the AI perspective like Minsky’s, where he is talking about
this as the ability to solve hard problems. Minsky has also given the concepts of agents
which are like black box experts and agencies like organization of conglomeration of sub
agents.
We have also seen that whenever these sub agents, developments, etcetera even it is very
similar in the brain also it takes place. It involves the networking of neural oscillations,
there are issues of synchronizations there which has been modelled by Winfree initially
and later on a very famous model, we have discussed about it that is the Kuramoto model
for the coupled oscillators that has come out.
And what are the factors that affect the synchronization? Things like say time delay,
things like the coupling strength, there are this 80-20 coupling, and things like the
synapses, there are different such parameters which actually control the nature of
coupling and the possibility of synchronization.
And finally, we have talked about the child robot interactions. So, this was the summary
in this four week. Now finally, the introductory lectures are all over, let us summarize
the entire you know this introduction of the foundations of cognitive robotics. What we
have discussed let us just try to summarize it under certain bullet points.
372
In the very first week, I have talked about the definition of cognitive robotics, if you
remember from various perspectives and I have talked about if you remember, the
organismoid robots and you know completely organic type and I said that right now, we
can be at best at the organismoid robotic space. I have also talked about the building
materials for cognitive robots that has the smart materials, different types of
piezoelectric, magnetostrictive, shape memory alloy, and electroactive polymers.
So, we first we devoted in terms of defining the cognitive robotics and the building
materials for cognitive robotics that was our you know concentration on the first week.
When in the second week we have talked about the architecture of the brain, because we
said that until and unless you know that how one of the most fascinating living machines
the human brain itself you understand it, you will not be able to know that what you have
to achieve in terms of the robot.
So, we have discussed about the architecture of the brain, the centre of cognitions, we
have also discussed about the architecture of the neurons. So, that was our week two and
then while discussing about the architecture of the neuron we have talked about the
modelling of the neural impulse and in the next week we started discussing about Huxley
Hodgkin model of neuron.
And also we have said how you can further expand that model in terms of our current
understandings particularly, in terms of the action that you can model it as a piezoelectric
system and also I have talked about various types of experiments that you can do using
EEGs. So, a little bit of introduction to EEG signals I have given you.
Finally, in the last week we have talked about the intelligence. How to define
intelligence and what is expected from cognitive robotics along with a few very
preliminary experiments that we have carried out. So, that is our final journey in this
week. So, let us hope that all these things together will give us a kind of a platform,
based on which we can do the advanced studies on cognitive robotics.
373
Now, I must acknowledge for this particular lecture my students Aravind Sanmugam
and Anirudha Bhattacharjee for their research which I have presented through this one.
Thank you.
374
THIS BOOK
IS NOT FOR
SALE
NOR COMMERCIAL USE
(044) 2257 5905/08

nptel.ac.in
swayam.gov.in

Foundations of Cognitive Robotics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Foundations of Cognitive Robotics

Uploaded by

Copyright:

Available Formats

INDEX

S. No Topic Page No.

(Refer Slide Time: 00:21)

(Refer Slide Time: 11:17)

(Refer Slide Time: 13:01)

(Refer Slide Time: 14:34)

(Refer Slide Time: 15:27)

(Refer Slide Time: 16:58)

(Refer Slide Time: 20:02)

(Refer Slide Time: 26:14)

There can be further improvement of structural coupling in terms of historical coupling.

(Refer Slide Time: 33:23)

(Refer Slide Time: 33:52)

(Refer Slide Time: 02:40)

(Refer Slide Time: 04:25)

(Refer Slide Time: 04:29)

(Refer Slide Time: 06:11)

(Refer Slide Time: 08:16)

(Refer Slide Time: 11:44)

If you make your robot even softer like; high-density polyethylene or

(Refer Slide Time: 14:21)

(Refer Slide Time: 15:49)

(Refer Slide Time: 17:09)

(Refer Slide Time: 19:17)

(Refer Slide Time: 20:36)

(Refer Slide Time: 24:19)

(Refer Slide Time: 26:34)

(Refer Slide Time: 28:37)

(Refer Slide Time: 34:37)

(Refer Slide Time: 35:04)

(Refer Slide Time: 35:06)

(Refer Slide Time: 35:16)

(Refer Slide Time: 35:26)

(Refer Slide Time: 35:32)

(Refer Slide Time: 04:36)

(Refer Slide Time: 08:06)

(Refer Slide Time: 09:29)

And it is cost-effective in terms of a low-frequency band, it is not as high as a frequency

(Refer Slide Time: 15:05)

(Refer Slide Time: 17:18)

(Refer Slide Time: 17:52)

(Refer Slide Time: 18:12)

(Refer Slide Time: 18:28)

(Refer Slide Time: 20:06)

(Refer Slide Time: 21:11)

Essentially, these directions it would be different in comparison to the other two

Applications, as I told you, is phenomenal; it is increasing every day; mechanical,

(Refer Slide Time: 26:37)

(Refer Slide Time: 27:29)

(Refer Slide Time: 28:32)

(Refer Slide Time: 30:35)

(Refer Slide Time: 01:01)

(Refer Slide Time: 01:25)

(Refer Slide Time: 03:14)

(Refer Slide Time: 04:17)

Now, in terms of the electroactive polymers, if we compare the electroactive polymers

(Refer Slide Time: 16:49).

(Refer Slide Time: 19:48)

(Refer Slide Time: 20:26)

There is another very interesting application of this agonist-antagonist motion. By using

(Refer Slide Time: 21:07)

(Refer Slide Time: 21:14).

(Refer Slide Time: 21:34).

(Refer Slide Time: 22:09).