Welcome to the course Control of Mobile Robots.
My name is Magnus Egerstedt, and I'm a
professor at
Georgia Tech in the school of Electrical
and Computer Engineering.
And I am delighted to be able to to
deliver this, this course to you.
The focus of the course is, how do you
make
mobile robots move in effective and safe
and predictable ways?
Meaning that they don't slam into
obstacles, they move smoothly and
nicely without having weird oscillations
in their motions.
And the trick to achieving this is
something called control theory.
An this first module of the course is
going to focus on controls and control
theory, and
what it is, and what the pieces are that
you need to be able to actually control
robots.
Before I dive in, though, I wanted to say
a few words about myself.
My research hammer
is control theory, so that, as I've
already said, deals
with how do you make things move or behave
well.
This could be stock markets.
This could be cruise controllers.
This could be limiting the spread
of epidemics, for instance, using
vaccination strategies.
So control theory doesn't really have a
tight, necessary coupling to robotics.
It's a general language for describing how
things behave.
Now, if that's my hammer, my nail is
robotics.
And I'm very, very interested in how do
you make robots do elegant and useful
things.
And, in my lab, we do really three
different types of robotics.
One is swarm robotics.
So how do you get lots and lots of robots
to do exciting things together.
The other is something called behavior
based control, or behavior based robotics.
And we will see quite a bit of this in
this course.
Behavior is basically a controller that's
in charge of something, like, going
towards
a landmark, or not slamming into
obsticles,
or following cooridors, or things like
that.
And the question is: how do you design
these behaviors and how do you patch them
together.
The third thing that I'm interested in is
field robotics.
Meaning how do you take robots and put
them out in
the unstructured, strange world outside of
the lab, meaning, if you
have a robot driving through the forest,
all of a sudden there
are no corridors or hallways or doors and
life becomes much more complicated.
So, before we start, I wanted to show a
little
video that came out of my lab a few years
ago.
So, here, this is is an example of swarm
robotics.
We have 15 robots that have been asked to
spell grits.
And grits stands for Georgia Robotics and
Intelligence Systems.
That's the robotics lab that I'm
directing.
And, these robots have to figure out where
they should go in the letter.
They need to figure out how they should go
there and they
need to figure out how to go there without
slamming into each other.
So, what's going on there is control
design at multiple levels of abstraction.
One each, one is the high level, where
should you go?
Then you have the lower level which deals
with, okay, I know where I want to be.
How do I actually get there without
driving
into other robots, for instance.
And as you can see, these robots,
with very limited information, meaning
they don't really
know where every other robot is, they
figure
out where they should go in this
formation.
And this video I like for multiple
reasons.
One is, because there's a lot of control
theory going on underneath the hood.
And the second is, you will be able to do
this at the end of this course.
So, let's talk a little about
the course.
Control theory will play a key role,
meaning, how do you actually control
systems?
And this will be done decoupled from
robotics.
But then, we're going to couple in the
robotic applications by
looking at robot models, because not all
robots are the same.
Some are hopping.
Some are slithering.
Some are rolling.
Some are flying.
And different type of robots require
different type of controllers.
We're going to look in general at
mobility,
meaning, how do you make the robots move?
That's the main question we're interested
in, in this course.
And then we'll end with applications.
So, what can you actually use these robot
for, robots for and where do
we need the kinds of solutions that we
will be going through in the course.
I do want to say though, that the course
is
not about artificial intelligence, which
is the high level planning,
meaning what should the robots actually be
doing.
We're going to be dealing with the
question of, when they've already figured
out what they should be doing, how should
they be doing it.
We're not going to deal much with
perception, meaning how do we
take computer motion, data and actually
turn it into something actable?
Or how do we actually design perception
algorithms?
We're going to assume that someone else
has done this.
And we're going to act on the perceptual
data that we already have.
And it's not a mechanical engineering
course.
Meaning, we're not going to be building
robots
or dealing with different kinds of robotic
designs.
However, there is a slight little caveat
to that.
Because we will indeed be looking at some
hardware aspects of robotics.
Here is a video from a course I taught on
campus at Georgia Tech.
And, this is a robot
that's running a lot of the things that
we're learning in this course, meaning
it's figuring out
how to get to a landmark, in this case, a
purple little tape mark on the floor.
But it has to do it while avoiding
obstacles.
So, it has to design different controllers
for different situations.
For instance, the left robot is not doing
so well.
The right robot here is doing even worse.
It's slamming into things.
But the point is that, at the end of this
course you will be able to do this.
Hopefully better than what we are seeing
right
now, because the robots are all over the
place.
Here we're seeing some other
instantiations of this.
See
how the robot is oscillating a little bit
too much, but it is able
to navigate the, the cardboard box and,
well, the right robot is driving into
chairs.
The other robot seems to be a little
confused.
So, these are examples of robotic
controllers
that aren't perfectly tuned and perfectly
designed.
Let's look at a final race.
We're going to see what the outcome would
be if you'd done the control design
properly.
See, they're going straight.
There's very little oscilation.
Well, one robot is oscilating but the
purple robot
there seems to be doing really well, going
quite straight.
And this, in my mind, is a successful
robotic control design.
So, what is actually in the course, or how
is this structured.
Well, we're going to have lectures.
There're going to be eight mini lectures
each week.
And, there are going to be seven weeks in
total.
Each week will be capped by a quiz.
And, there will also be glue lectures.
So, lectures that are designed to glue
what
I do, my slides, to the actual quizzes.
And then, as a bonus, there will be
additional material.
This is not mandatory.
But if you do it, your experience will no
doubt be more enjoyable.
First of all, we have a robotic simulator
that we use in the lab called Sim.I.am.
It runs on MATLAB and this will be part of
the course, meaning you can do it if you
want to.
Few words, MATLAB, is the program
environment we are using
because this is one of the standards in
the engineering world.
Now, you will be able to get MATLAB
for free as a student license through this
course.
So this will come at no extra cost to you.
But then, we'll also map this onto a real
hardware
platform, where, if you're interested you
build your own robot.
Because, I truly believe that there' s
nothing more exciting than doing
the theory right, and then seeing things
move in the real world.
So, if you follow along the course, do the
simulatin and also build the robot.
At the end, you will have an awesome
platform that can actually run around in
the world
without slamming into things.
Before we do this though, I want to make
one minor or maybe even major disclaimer.
To do robotics right and to do control
theory right
there is no way around the fact that you
need math.
So mathematics will play a key role.
And this may be a little uncomfortable to
some of
you at first but we're doing it for a
reason.
And I
will try my best to make sure that I
explain why
we're actually going through all the stuff
that we're going through.
But things like linear algebra, matrices,
and eigen values will show up.
Differential equations will play a key
role.
Calculus, derivatives and integrals will
be everywhere.
And this is just something you have to do
in able,
in a, in order to be able to do robotics
right.
And I hope you
will be following me along this ride for
this seven week adventure.
As a final note, before the actual course
begins, let me that
this opportunity to introduce the team
behind the course, they are all graduate
students in robotics and control theory,
here at Georgia Tech, and here
is Smriti Chopra, Smriti is going to be
running the, the glue lectures.
So these are lectures that are going to
map the general stuff
that I do in class onto the particulars of
the weekly quizzes.
And,
Smriti is also going to be highly involved
on the discussion forums.
And here we have J.P. de LaQua.
J.P. is going to be the, the master of
ceremony when it comes to running the,
the robotic experiments that we're
sprinkling in throughout
the course to highlight and stress
particular key points.
And JP is also the key force behind the
optional
simulation part of the course, and he will
be running
the weekly programming assignments.
And here lastly is Roland O'Flaherty
and Roland is the hardware guy.
He's going to show us how to build robots.
And this will also be an optional weekly
feature where we will be building
platforms that,
at the end of the day, we can implement
and run our controllers and code on.
And Roland tells me that he's actually
going
to produce a cooking show, but instead of
baking bread, we will be baking robots.
EmoticonEmoticon