Picking up again from
your first love Mark.

Today we're very lucky to have
Tucker Tucker's associate

here actually here and
eight hundred ninety eight

for one year and then for two years before

returning to Georgia Tech in two thousand
and one Tucker never writes the Borg.

Well yes to attack his
research interests include

collaborators in communication and
multifocal teams and activity right.

Moderately live social systems.

Tucker has published over
seventy research articles and

did two books on a personal
note he likes boiled peanuts

and he said and a former F.B.I.

team thanks.

The poster for
my talk talked about using robots for

education and I'm going to talk about
that but I thought I would just

before that show you some of the projects
we're working on in robotics here.

Overall.

A couple interesting things that
occurred over about the last two years.

One of them is we've created
a Research Center for Robotics.

There are about thirty professors and

other researchers here I enjoy the tech
and of Engineering and Computing the do.

I work in robotics and we got

together to form this center
that now you know arguably.

We're among a handful of the top schools
and robotics anywhere because we sort of

pulled together in this way we also
now offer a Ph D. in robotics.

Or one of only two universities
that offer a Ph D. in robotics.

I have to pull asterisk pending
Board of Regents approval which.

You all know how that goes but
they're meeting actually next week.

On that.

So we think there's not going
to be a problem with that.

So we're really excited about that and
we're also pushing robotics more and

more into the undergraduate curriculum and
that's the topic of.

Sort of my last.

And my last set of slides.

But let me let me show you
just a few of the things.

We do here related to.

Robotics.

And many of you may have heard about.

The urban grand challenge.

This is the robot that we entered now.

And you know unfortunately we didn't
when the finals were conducted.

This Saturday but.

We I think we really did a fantastic job
within a year so we we went to centrally

from nothing to a fully autonomous
car that could operate in traffic.

In about a year and.

It.

No it didn't know.

Well it finished some
courses I should say.

Well just let it run and fortune.

Jump a little bit tell you what it is so
this is from art testing of the car

at the Georgia Public
Safety Training Center.

Down in Forsyth there is a.

Test Center where actually
the police in Georgia train and

there's a large highway like track and

then an urban course which is
where we're running right here.

And this.

This other cars acting as as traffic for
us.

So it's got I believe a total
of eight laser sensors that act

like radar five actual radars six.

Six cameras and then in the back.

We've got.

Eight server computers that process
all the data and the way we've got it.

Architected is.

Each computer.

Is paired with a sensor system so.

For instance one computer is devoted to
radar and others devoted to laser and so

on and the sensors connect
directly to those computers

they process whatever they need and
share it.

With the others and eventually.

To the main control computer.

The vehicle could go or
can go thirty miles an hour

stop at stop signs and look for other cars
turn on its turn signal when to turn.

The.

Infected it can do all the tasks that are
necessary for the Grand Challenge of what.

Caused this big problem at the qualifying
event for the Grand Challenge was in.

First run just as the robot started
the serial connection between

the G.P.S. and I.M.U.
and the control computer froze.

So the robot didn't believe it was moving.

Actually it's a.

Normally very reliable connection.

And we're not sure what
happened is just a normal

nine pen serial connection but it froze.

So the robot thought it wasn't moving.

So it applied some gas.

Still not moving.

So it applied some more gas.

And if you look at the sensor logs.

It looks as if from the robot's point of
view this concrete wall ran into it you

know that it was sitting there minding its
own business and home and it was actually.

Recorded at twenty four miles an hour
when it hit the concrete barrier.

So we actually.

Rebuilt it.

And it and it worked fine but that.

Accident early on sort of.

I don't know.
DARPA didn't like it put it that way and

and so we were qualified for the final.

However I think our our
vehicle actually performs.

Just as well as as many the vehicles
that were in the final.

So anyways.

We're going to continue our
work research with this robot

a lot of folks ask you know what.

What is the goal or purpose for
research like this and it.

It's not just for autonomy cars.

There's many many spin offs for
this technology for instance.

Being able to process the input
from all the sensors and

know what's around you can help
a driver and in many instances.

For instance.

The the car can sort of alert you that
you're about to do something that might

put you in harms way.

So you know magine it's
late at night some driver.

Coming up the intersection doesn't have
their lights on and you don't see them.

While the car could see it.

See this other car with
a radar laser an alert you.

And other using vision the vehicle can
see the lane that you're in those where

the center of the lane is very well and
just a simple sort of tactile feedback.

You could feed back to the steering wheel
that such that when you go to change lanes

or so little bit of resistance to move and
then zip you go over the next lane and

kind of snap into the next lane so you
can safely keep you in lanes and so on.

So what lager is about is.

All the performers there
are eight eight groups

participating get an identical.

Robot.

And.

Then have to set solve navigational
tasks and in general we have to do is.

Each month we send

to the government a Compact Flash with
our software on it and they test it.

They run it three times over this course.

Now what we've done recently
is look at the problem of.

How can a robot.

Navigate through.

Sort of complex obstacles.

And not say through planning or

through sort of classical control
theory but through learning.

So let me see.

As we get over to the robot Let
me show you what it looks like.

So that will be.

A little or an herb red cup was the.

The goal.

So this is the robot and my.

Right into that my student.

Was driving it.

In this pause there.

OK It's kind of hard to see but it's
got sort of the stalk on which there's

four cameras actually two stereo pairs.

So it uses these cameras to
see what's in front of it and

to build a model of what
obstacles might be in front of it.

So right now there's no obstacles.

And it knows by G.P.S. where the goal
is so it's happening now while

my student drives it to the goal is
it's getting all these examples of OK if

the goal is there and there's no obstacles
around me I should drive to the goal.

And the very back to the thing
he did at the very start.

Beginning here.

Notice that we started it
facing away from the goal and

then he very slowly turned
it around to face the goal.

The reason for that turn is for
training it for

that turn is we found out in our
first couple tests that if it got

if it only been training just
driving straight to the goal

that somehow found itself in a situation
where the goal was off to the side.

It didn't know what to do because it
didn't have any training example.

So that's why the beginning
we do that little turn.

So now it.

Now knows OK if this open terrain and

goes ahead I mean just drive
directly to the goal but

what if we make a lot more complicated and
put a hay bale in front of it.

Well it hasn't been trained so

it doesn't know any better than
to just drive into the hay bale.

So now we train it.

To drive around hay
bales when it sees them.

So he's he's got the radio control and
this little switch he turns that

turns it on to learn and
then when he drives around.

So now the next time.

It sees a hay bale.

In those drive around it.

Not as much more complex.

You know terrain than
just a single hay bale.

So we set up a more
complicated course with

lots of hay bales and
there it is doing that final turn.

So this is you know the robot's all
you know just driving by itself.

So we trained it for
about fifteen minutes among hay bales and

now these are sort of sped up replays
of what it what it learned how to do.

You know just and make a motion sick.

Now I think in fact I'm
pretty sure we can.

So some somehow in that earlier part.

I mucked up the.

Movie we quit that now
start my Power Point.

We can we can use the same
approach to to train a robot say

how to find lanes on the highway and and.

It can be used for vision for
stereo for all sorts of things.

So we're going to be expanding that.

OK now this is the advertised
topic of my talk and.

In conjunction with this new
robotics research center.

We've created something
called the Institute for

personal robots in education.

Now so Georgia Tech is an institute that
we have a center in their institute and

inside the center.

I don't know where the official
rules are for what you can.

Name things but I suppose.

And we could have a lab within
the institute and maybe a center.

I don't know.
It was the reason it's called Institute as

we got funding from Microsoft for
this and they really wanted

whatever it was to be called an institute
so we said OK we'll work on Institute.

What it's about is to
make computer science

education more interesting and
exciting and.

But a little bit about
how we lead up to this.

I mean Georgia Tech
undergrads are in here.

You know it's a few.

Dirge tech is one of the few and

certainly one of the first to
require that everybody all

undergraduates take a course
in computer science.

And that started back
in the early ninety's.

And that you know I think
that's a great idea.

The problem however was when college
of computing started teaching.

Computer science introductory computer
science to every single student about

thirty percent of them failed.

And the percentage was even higher for
women and some minorities.

So we had this big problem.

You know namely that students had take
this course over again we were already

teaching about one thousand a year.

In introductory computer science if they
had taken over again that meant to since

they were teaching you know
close to two thousand a year.

Those were eating
everybody's resources and

ticking off the College of Engineering and
so the issue was how to address that.

Well one thing we did is
created a new version of

the course that centers on media.

And by take the media
version of the course.

OK.

It was the way that courses
pitched to students

taking it is OK we're in a learning.

And images and sounds and movies.

And you're going to learn how
to make a little movie and

the way after you've made these movies and
sounds and so

on you you sort of inadvertently
learned computer science and

that became very effective and
now about thirty percent of the students

take that version of the course and
it's got a much higher pass rate and.

For some reason women take it in much
higher percentages than the other

than the other Course so.

Around the same time we also began this.

Well in the early two thousand as we
began this big effort in robotics and

we began to think hey can we use robots to
help teach computer science in the same

way that we used media before and that.

That's that's what we've
been doing let me.

Jump ahead here.

So the.

The experiment is that we're going
to give in fact this semester.

There's there are two sections of this
course every student gets their own

little robot that looks like this guy
down here in the bottoms about this big.

It's for sale in the bookstore and
it's required.

You don't have to buy a textbook.

But you do have to buy a robot.

And.

The idea is so the textbook actually
is online as a electronic document.

The idea is.

So we're not the only ones who thought of
this idea to teach computer science using

robots.

In fact there is and
they did do something similar at MIT.

The what's different about our approach
that I think is better is that

we have these inexpensive robot.

And everybody gets their own
MIT they've got a lab with

you know five to ten
very expensive robots and

the students have to work in groups
of four they have to reserve time to

use the robots and so on and our approach
you get your own you take it home in fact.

They now have cameras on them and

you know lots of kids that are in dorms or
so on in the robot down the hall.

You know to spy on their neighbors and
you know take pictures and

then becoming situations and
put on the Web So they're having fun.

You know they're showing it off
to their friends and so on.

The.

The current robot.

Plus the dongle you need to buy
is about one hundred twenty.

The robot is made by the the robots
made by a company called parallax.

We have our own dongle
that you plug into it.

That adds vision and Bluetooth
communication back to your laptop.

So that dongle costs about sixty
dollars robots about sixty so

it's one hundred twenty
are our follow on robot which

will be available in the fall
is going to collapse.

What's on our dongle and in the in this
blue robot into one and we think will be

able to sell that all together for about
eighty dollars or so in the sensor go.

Yeah.

It's not being taught this spring actually
will be taught in the fall there and

there is a junior level robotics course
being taught in the spring though.

Actually two of them.

So let me remind you though
the course I'm talking about is.

Is computer science thirteen
No one you know the first

computer science course
that you that you take

it there's lots of questions that
we don't know the answer to.

And and
actually this is kind of an experiment.

And fortunately a big
research universities you can

conduct experiments on undergrads so.

You have to file some paper saying that
you know you won't burn their eyes

with lasers.

But as long as you know.

No nobody is physically damaged it
seems you can but I'm exaggerating.

Of course.

The we were taught this course last
spring and we taught in the summer and

now we're teaching at this fall.

Two hundred eighty students and the result
from the spring was essentially.

We did no harm in other words
the students coming out of that course.

Liked it.

About as much as students taking
the non robot version of the course.

I think part of the reason we didn't
have sort of a stupendous success.

Last spring was it was sort of new and
actually we were used to teaching it.

Where'd getting pretty good
at teaching it now this fall.

So I think we'll have more interesting.

Results but it was there are a lot of

research questions about you
know how do you really teach

computer science with with these robots
that that we're pursuing and I mentioned.

I mentioned a few minutes ago that at

Georgia Tech that we were
among the first to require

computer science for
all all majors and now we have.

Before the robot version we actually
had three options available sort of

the standard.

Java based approach.

There's a Matlab based approach and

then this media approach and
what we're doing is we're replacing this.

Traditional Java based one with
the with the robot version but

anyways the results as of.

A few years ago before the robot.

Was introduced is that we went
from twenty eight percent

withdraw failure rate
down to sixteen percent.

And the students who take
who have taken these sort

of alternative courses like
media are doing just as well.

Later on in the program.

As their as their peers.

So there's there's
evidence that using this

what we call a context
based approach works.

I mentioned already that.

A key component is everybody
has their own robot and

I see like you not in your
head like yeah that's cool.

So I mean part of it is
kind of a cool factor but.

You know we need to do
everything we can to get.

Students interested in this.

Now something I hadn't mentioned yet
in this presentation is that.

Not only.

Do we have trouble with with people not
faring well in computer science one

now nationwide enrollment
computer science is dropping.

In fact.

It's down about fifty percent from two or
three years ago.

So that means for instance my wife
teaches at University of Georgia.

And she used to teach her operating
systems course to twenty students.

Now there are six students in that class.

And that's not because
they don't like her.

That's just how many students are around.

Georgia Tech is one of the very few.

Who's in Rome and
computer science is actually growing.

But in fact we may be the only one
maybe the only major university for

which that's happening and.

This this idea of using robots
is also something we hope

to we're going to send out
to other schools to make.

Make it interesting for students in other
schools to get into computer science.

OK.

The question is what should
the robot look like you know

us with the manufacturer was and so on.

Whenever people think about a robot for
education they think of Lego and

the biggest question we get is
why aren't you using Lego's But

let me throw that back at you.

Why would why would you think
we're not using Lego's.

Yeah so that right.

So you kind of hit it on the head
there were teaching computer science.

Not sticking Lego blocks together.

Now Lego.

Is actually a great platform.

I think for teaching say a junior or

senior level engineering course in which
you're trying to show how to integrate

computing mechanics and
syncing say like is great for

that but you imagine your teacher you
get two hundred students in the class.

You're in the first day of class and
say OK everybody open up your Lego box and

then you open up this thirty seven page.

Construction manual and
you know you start with step one.

And you know clearly that's impractical
The other reason it's impractical

is Lego's are great but they they break.

And.

If you've got your program on
the robot and you try to run it.

It doesn't work is that.

Because your software is bad or because
you didn't have a little connector in

the right spot and it's just
becomes extremely frustrating for

students who are just trying
to learn programming.

So the idea is we wanted something that.

Is kind of like an i Pod
on Wheels in other words

it's all we're already assembled
you open up the box boom it's done.

You turn it on and
it's easy easy to make it work.

Now you could do that with
very expensive robots.

Which is kind of the way mits going.

But again everybody.

Reserved time for the robots and so on.

So this is why we've gone with this

approach this is a picture what it looks
like I forgot to bring a robot with me.

I'm sorry.

But it was every student
gets one of these robots.

It's this is the version we were
using last year that has this little.

Electronic attachment and
what that is is that say U.S.B. connector.

So it connects back to the laptop.

That's on the another aspect that's
very important here is the software

the students write.

Runs on their laptop.

It doesn't run on the robot.

Now they're couple reasons for
that the you know the standard model.

You know a standard approach
like what one people use for

Lego is you write software your laptop
compile it download it to the robot.

Now that's.

That's fine except.

Imagine you're running this program on
the robot and it does the wrong thing.

How do you know what's going on.

Well the normal thing tell a student as
well just put some print statements in

there and you can see what's happening.

Well robots don't have anything that
you can see a print statement on so

the best you could do is
like turn on an L.E.D.

or something to get feedback for
what the robot's doing this is why it's

much better if it runs everything
runs on a laptop because you

can see exactly what it's doing
in a print statements and so on.

Python.

This just talks about the sensors and
so on that they the robot has.

This is now the new version you can see
this thing kind of stuck on the front

the the little black circular thing

poking out is a cell phone camera
that's on that little longer.

And on the bottom you see.

This is a picture taken by that camera and

this is doing a little bit of processing
to find just the orange pixels.

You can find things like soccer balls and
so on.

So this is the version we're
going with this semester and for

the next few semesters.

And this is sort of what we want
the final version to look like

something about the size
of what's in your hand.

And it detection going to have
someone look exactly like that but

that's about the size or we're aiming for.

These are just some pictures of students.

Using the robot.

Now one thing I noticed when
I was an undergrad here.

Everybody saved.

P. E. for
their senior year I don't know why.

But back then we had to take
this drown proofing course.

Which we literally had to do this
they would tire hands together in our

feet together and they would put a brick
around our neck and we had to somehow

survive in the pool for twenty minutes and
eventually that course got outlawed.

Not sure.

But it was that was of course everybody
saved till their senior year.

It turns out that computer science one for

some majors is the course that
they save until their senior year.

And these two women were graduating.

I think systems engineers.

Who've you know their senior year their
senior their last semester they took this

course and they just had a great time and
they told us.

Gee I wish I had taken this when I was

a freshman I might have
switched into computer science.

There are of course people who hated it.

And but for the most part I think people
people have had a good good time with it.

So he also mentioned we're doing
this with Brant Mark college.

And part of the reason that we're.

Paired with them is they're an all
women liberal arts college.

We are in almost all
male Technical College.

And so the.

The idea is let's develop and test it
in these two extreme environments and

sort of distill the lessons so that it
can apply in many different environments.

And starting next fall.

We're also rolling it out to
University of Georgia Georgia State.

Maybe Spelman and other colleges as well.

So the idea is that take the same
curriculum and try lots of places and

see if it can sort of survive the test
in these other places as well.

This is just some example code.

Let me tell you how.

Just real quick.

How you teach a fundamental
computer science concept

on this robot one of the early things
you teach is something called iteration.

So something for example something you
want to do over and over and over again.

It's called iteration.

So one of the first assignments is
OK make the robot drive in a square

and one of the things that's nice about
this robot it's got a hole in the middle.

That's exactly the same
diameter as a sharpie pen.

So you drop the little Sharpie and

put some poster board on the ground and
you can draw.

All kinds of designs.

So what they're really
projects is OK draw a square.

So we show them that.

OK.

You know right.
The step to draw draw them drives

forward a foot then turn right.

Knight agrees and then do that.

Four times and you have a square.

So you've got eight lines of
code go Turn turn to a turn.

So that's an important concept
in other words that you write

computer programs by showing step by step
sort of a recipe of what happens and

it happens in order but Haiti generation.

Well what you can show is OK.

Instead of having these two lines
repeated over and over again.

You just have a little for loop and
you do the same two lines four times and

so that's that's how you
teach with these robots and

then the idea is that it's grounded.

You know something physical moving around.

And sort of comparison
sort of classic first

programming assignment is
write a program to print the.

Fema not she number sequence.

And that's all well and good except.

Students don't get why does that matter
it's just a completely artificial thing.

So if you ground in a robot
this moving around.

It's more fun.

Is the chapters in our book.

These are just some of the example.

Projects.

This is this is one of
the brain Mar students project.

So their task was to turn the sound off.

Their task was to program the robot to
go through a maze and reach a light.

And this is one of their solutions.

What's kind of.

Cute about it.

I turn down the music but she.

She made sort of a dance video out of it.

And then.

Thank you thank her.

Professor.

Yes.

So if you just start at
robot Education dot org.

And then you can get to the book.

This is this is another project where the.

The robot.

Follows a light.

I guess it seems a little bit impractical
that you have to stick the light.

You know right in its forehead but it was
it has three light sensors in front and.

This is just showing that you
can connect with the brightest

light sensor you know says to the motors.

Let me.

Yeah that's

the aerial I should add that these.

The company.

That makes these little blue robots.

For some reason they don't like them.

So when we called them and said we
want to use this little blue robot for

the course they said no no you don't
want that you want this other one and

I said no we want to the blue one and
the other one that he wanted

us to get is something called the bow bot
and it's the classic you know you get

a screw together and connect the wires and
we you know we didn't we don't want that.

So anyways when we started this about
a year and a half ago they had.

Two thousand of those
robots now they have zero.

And they can't understand you know
the robot wasn't selling they had it for

five years now they have zero in a certain
you know we're telling look it's because

of this you know we we've
been to lots of conferences.

You know where people teach computer
science and they're buying these robots.

So now he's.

Ordering ten thousand of them.

So that's that's

that's kind of nice to know they were
having an effect on the economy.

But the the new robot they
were going to make is going to

going to going to replace that one and
I think be much nicer but

it was the was finally
getting around to saying is.

The same robot can work for many courses
not just intro computer science in fact.

It Carnegie Melon they're using it for
senior level intro to robotics course.

Because you can teach certainly
much more complex stuff on

on the same robot and
here at Georgia Tech.

It's used in a mechanical engineering
intro to robotics course and

strangely enough they've figured
out how to control it from matlab.

So they teach this course
in robotics in matlab and

they can drive the robot around in my lab.

Anyways that's sort of the end of my two.