Nova (1974–…): Season 43, Episode 8 - Rise of the Robots - full transcript

It's one of the most
ambitious challenges

in the history of technology.

Ladies and gentlemen,

start your robots!

The race to build robots that
can save lives in a disaster.

That can even rescue us.

But to do that,
they'll need to climb ladders,

walk over rubble, turn valves.

And no one knows
if it's even possible.

There's so many things
that could go wrong.

It's scary.



When you try to instill
human-like intelligence,

human-like manipulation skills
into machines...

it's just very, very hard to do.

What will it take
for robots to have...

the right stuff?

And are we ready
for the consequences?

It's too easy to look at them
and say,

"They're not there yet."

They will get to something
very powerful,

and then you have to say, "Well,
where will we have gotten to?"

They might actually have
the capacity

to be smarter than us.

What impact will robots have
on our future,

on how we work, how we interact,



even how we see ourselves?

"Rise of the Robots,"
right now on NOVA.

For decades,
roboticists have wowed us

with their creations.

Do you have any questions?

Especially in Japan, where bots
are starting to cook for us,

do the laundry,

even sing and dance.

Some look so human,
it's hard to tell us apart.

But looks can be deceiving.

People see these humanoid robots
developed in Japan,

all these fancy things,

and they had the expectation
that, "Oh, we have these robots

and they're going to, you know,
save the world."

But that didn't happen.

March 2011.

A devastating earthquake
and tsunami rocked japan.

Explosions and fires

at the Fukushima Daiichi
nuclear power plant

left the area
dangerously radioactive.

Going inside, even for a few
minutes, was life threatening.

It was the perfect time
for Japan's cutting-edge robots

to come to the rescue.

But none did.

The real tragedy was

if simply some valves
could have been turned

or some switches
could have been flicked

or hoses attached,

a lot of the meltdown
could have been prevented.

The question was, when robots
were needed the most,

just how come
they weren't effective?

And so I think that led
to a lot of self-reflection.

After decades of research,
where were our robot heroes?

For generations, science fiction

has portrayed robots
as our loyal servants.

Welcome to Altair IV, gentlemen.

I am to transport you
to the residence.

But it turns out our imagination

has taken us much further
than our technology.

It's hard to create robots
that can function in our world.

One of the challenges

with deploying robots
in the real world

is that the real world
is like the wild, wild West.

It's very rich and complex
and very unforgiving.

What will it take

for robots to make their way
out of the lab

and into the real world?

We're about to find out.

Ladies and gentlemen,

start your robots!

At the most ambitious
robotics competition in history.

We welcome you to the DARPA
robotics challenge.

This is where imagination
meets innovation.

You have completed the course!

Whoo!

In 2013, at a racetrack
on the outskirts of Miami,

DARPA, the research arm of
the U.S. Department of Defense,

challenged 16 teams
from around the world

to build rescue robots that can
help save lives in a disaster.

Very exciting, very exhausted,

very nervous and frustrated
at the same time.

Dealing with robots
is always like that.

It's the kind of challenge
DARPA is known for:

cutting-edge and high-risk.

DARPA was formed
back in the 1950s

in response
to the launch of Sputnik,

the world's first
artificial satellite

built by the Soviet Union.

Since then,
the agency has spent billions

developing military technology:

advanced weaponry
like stealth technology,

drones, and night vision.

But the spin-offs
from this research

have gone
far beyond the military.

The Internet, GPS,

bionic arms, even Siri,

were all fueled
with DARPA funding.

Cancel golf today.

It's off your calendar.

Good.

If these researchers succeed
in creating rescue robots,

the same technology
could be used to develop...

robots that take care
of the elderly,

babysit our kids,

clean up after us.

Robots in almost
every facet of life.

But first, can robots
really help out in a disaster?

To find out, DARPA has set up
an obstacle course for bots.

With human operators
controlling their every move...

Yes, we got it!

The robots must perform
basic tasks,

like opening a door...

All right, go, go, go!

Turning a valve,

drilling a hole in a wall,

walking over rubble,

even driving a car.

Bot after bot takes their
first steps into the real world.

And there's nothing easy
about it.

It's a monumental challenge.

I mean, DARPA calls it
"DARPA hard."

I mean, I almost said it was
"DARPA impossible."

Robots fall off ladders.

Some barely move an inch.

They even struggle
to open doors.

The things that a human can do
instinctively, easily,

you don't realize how hard
something like walking is

until you try to reproduce it
in a machine.

When my baby daughter
took her first step,

did she walk?

No, she took a step
and she fell down, right?

Over time,
she started to walk, right?

And it's kind of like
we're seeing here.

There are a lot of baby steps.

After two days of competition...

the results are not impressive.

But this is just the start
of DARPA's grand experiment.

The agency is giving
the roboticists

another chance to get it right,

and they're putting their money
where their mouth is.

The top scoring teams
in this first challenge

are receiving $1 million each
to continue their work.

In 18 months, they will return
for the final challenge

and the chance to win a grand
prize of $2 million.

Yes!

It's not clear to me how well
the teams are going to do.

It can't be too hard,
because then everybody fails.

It can't be too easy,

because then it's not
worth doing at all.

But if you make it just right...

Like Goldilocks, right?

I have tried to hit
the sweet spot of difficulty,

but I think risking failure
is the DARPA way.

Will the challenge push the
technology a giant step forward?

Will robots ever make their way
through our world the way we do?

No!

Back in the early '50s,

Alan Turing, one of the founders
of artificial intelligence,

said that the best thing
we could do was build a robot

with TV cameras for its eyes
and motors to drive its legs

and have it romp
around the countryside

and learn from the real world.

But he decided that was
technologically too hard

back in the '50s,
which it certainly was.

So he said, "Let's leave that
physical interaction until later

"and let's work
on more abstract problems,

the intelligence
abstract problems."

The field of artificial
intelligence, or A.I.,

has already built machines
that beat us at chess,

trade stocks
with lightning speed,

and search for anything we want
in an instant.

Yet when it comes to robotics,

progress has been
painfully slow.

Some of the biggest
problems robots face

are things we humans
usually take for granted,

like mobility, manual dexterity,

and the ability to see
and understand our environment.

These are the challenges
the robotics teams will tackle

in the DARPA competition,
beginning with mobility.

What's the best way
to make a machine

that can move through our world?

Does it need to walk on two feet
like us?

Some roboticists think
the answer is yes.

The shape of the robot is
dictated by what it needs to do.

There's a reason why the step
size in your home is this big,

there's a reason why your
door handle's this high,

because it's designed
for humans to move around.

So unless the robot is
the shape and size of a human,

it won't be able to navigate
and move around

in the environment
designed for humans.

But getting around on two feet
isn't easy, even for us.

It takes your average infant
almost a year

to go from crawling
to toddling to walking.

What does it take
to give a robot

this intrinsic human ability?

It's one of the biggest problems
facing roboticists today.

In Pensacola, Florida,

at the Institute for Human
Machine Cognition, or IHMC,

one of the teams competing
in the DARPA Robotics Challenge

is hard at work
developing its own software

to run this massive bot
named Atlas,

a 385-pound powerhouse.

DARPA funded the design
and development

of this rescue robot.

Several teams competing
in the challenge

are using this hardware,

but writing unique software
to guide their robot.

You might wonder why Atlas
is so top-heavy.

Most of its oversized head is
packed with cameras and sensors.

And while its feet
may look small,

they're designed to fit
in the kind of places we walk.

With bipedal walking robots,

there's still a lot of
strategies we have to determine.

And there's no known
textbook solution yet.

It's more art than science
still.

For team leader Jerry Pratt,

finding the best way
for a bipedal robot to walk

has not been easy or quick.

Well, I've been working

on bipedal walking robots
since 1994, so 21 years now.

He started back in graduate
school at the MIT Leg Lab,

where some of the most
bizarre-looking bots

hopped, jumped, and flipped.

But the robots Jerry built
were different

because they walked on two feet.

Using nature as his guide,

Jerry gave his bipeds
hips, knees, and ankles

that mimicked how animals move.

This one,
called Spring Flamingo,

had specially designed motors
that worked a lot like muscles,

varying the amount of power
each joint used.

Its legs and feet worked a bit
like shock absorbers.

They had a little give.

That robot was
a really good workhorse.

We had it working
for about three years,

probably walked
about 20 miles or so.

When it came
to designing a humanoid,

he focused on developing

the perfect combination
of hardware and software

that would enable his biped
to stay upright

as it calculates the best way

to shift its center of mass,

lift its leg, swing it forward,

and put its foot back down
in the right place

and with just the right amount
of pressure.

To make these elaborate
calculations,

the robot needs instructions,

lines of code that tell it
how to move.

These lines of code are written

in a programming language
that resembles English,

but for Atlas to use them,
its onboard computer

has a program
that translates them

into a language a machine
can understand: zeros and ones.

It takes more than two million
lines of code to run Atlas,

500,000 just to put one foot
in front of the other.

It's the interplay
of hardware and software

that keeps this bot on its feet,

an ability that took
hundreds of thousands of years

of human evolution to perfect.

We'll often look

at what strategies does
a human use in order to walk

because the physics in the world
that a human operates in

is the same as a robot.

But we wouldn't be able
to take a single step

without having
some pretty amazing senses.

When you walk, your eyes detect
the position of your body

relative to the world around it.

At the very same time,

a series of fluid-filled canals
in the inner ear

tells your brain the position
and motion of your head

so you know which way is up.

And a kind of sixth sense
called proprioception

uses your muscles and nerves

to detect where your arms
and legs are

in relation to each other.

All this information
comes together

in a part of the brain
called the cerebellum.

What kind of senses does
Atlas use to walk on two feet?

For eyes,
the bot has a stereo camera

and a couple of fisheye lenses

sticking out the side
of its head,

along with a spinning laser
called LIDAR

that scans everything
in the world around it

and creates a 3-D model
of its environment.

Atlas doesn't have an inner ear
to tell it which way is up.

Instead, it uses
a small cylinder in its butt

that contains gyros
and accelerometers

that tell it where it is
and how it's moving.

In addition, sensors
on each joint tell Atlas

where its limbs are
in relation to each other.

The result is an extraordinary
sense of balance,

allowing Atlas to stand
on one foot like a ballerina...

At least in the lab.

But in the real world,

Atlas's gyros and sensors
aren't always enough.

We only have a limited number
of sensors on the robot,

whereas with a human
or an animal,

you have thousands
of little force sensors

on every square inch
of your body.

Over 100,000 just on the soles
of your feet.

You can step on something
and detect what it is,

whereas the robots,
they don't know that.

They just know
they stepped on something.

At the final challenge,

if Atlas should step
on something the wrong way,

if the hardware and software
don't work perfectly,

it will fall.

We cannot get up
from a fall currently,

and we probably won't
survive a fall.

If we fall during the finals,

that may be the end
of our whole weekend.

Atlas isn't the only biped that
has trouble staying upright.

Meet HUBO,
designed and programmed

by cousins Paul Oh and Junho Oh.

At the first challenge,

HUBO clearly wasn't ready
for the big, bad world.

It's with fondness...

It's also with some pain...
That we think about it.

We went into the challenge
with quite a lot of enthusiasm,

but when we got there,

it was just like
one thing after another.

There were a lot of these
real-world instances

that we did not experience
in the lab.

That was a real awakening.

Across the ocean

at the Korea Advanced Institute
of Science and Technology,

Junho Oh is also doing
some soul searching.

After the shock wore off,

the cousins agreed
to make a radical change

in the design of their biped.

My cousin has come up

with this idea of adding wheels
to HUBO's knees,

as well as casters on its toes.

And I call it
the kneeling prayer mode.

Others like to call it
wheeled mode

or transformer bot mode.

Humans don't have wheels
on the knees,

but there's no reason why
we can't add that to a robot.

When the cousins return
for the final challenge,

they'll debut their humanoid,
transformed

with the help
of some clever engineering.

But for a robot to help out
in a disaster,

getting there is just the first
of many challenges.

To assist rescue workers
in the real world,

it needs hands

with the kind of strength
and dexterity it takes

to lift heavy hoses,

drill into walls,

solder and saw.

It needs a pair of these.

Our experience of the human hand

is that we can do everything
with it.

We can dig ourselves
through rubble,

we can make really nice
paintings, we can knit.

The human hand really
distinguishes our species

from all other species.

If you look at the way
humans manipulate objects,

we have this instinct
for tactile sensing,

for feeling the world.

Understandably so,

we've had over 200,000 years
of human evolution

to get to this particular point.

Is it possible to translate
a masterpiece of evolution

into motors, cables, sensors,
and thousands of lines of code?

On the outskirts of London,
behind an unassuming storefront,

a small group
of robot enthusiasts

are building a robotic hand

that they hope could one day
rival our own.

When we first started
building a hand,

we actually bought anatomy books

and tried to understand
how the hand works.

We thought it would be good
if we could copy,

as best we can, the human hand.

The Shadow Hand was built
with the idea of trying

to get as close as possible
to the human hand,

but as engineers.

Like the human hand,

this robotic version has
four fingers and a thumb.

It's about the same size
as the original

and can even shake like one,

but that's where
the similarities end.

We pack a huge amount of sensing
and actuation.

We have 25 joint
position sensors,

nine analog digital converters.

We have two tendons
coming from each joint

to a motor in the forearm.

20 motors in the forearm.

Each motor has
a temperature sensor,

a current sensor,
and two full sensors

so we can tell how hard
the motor is working

and how hard
it's driving the tendons.

We put contact sensing
in the fingertips

so we can tell that
we've touched something.

So there is a wealth
of computing power

just to get something that has

the same set of movements
as your hand.

And with this level
of dexterity,

it can do a lot more
than card tricks.

We have this hand handling
pipettes and lab equipment,

removing the human
from the risk...

For example, people that work

with very, very nasty bacteria
and viruses.

The Shadow Hand was designed
for delicate tasks,

not for dirty jobs like this.

When you look at where
robot hands get used,

you find people who want to do

something delicate
and precise...

and people who want something
big and rugged and solid.

For the robots
in the DARPA challenge,

strength and precision
are a must.

Finding the balance between them

has been a struggle
for team leader Brett Kennedy,

who developed this four-legged
rescue robot named Robosimian,

here at NASA's Jet Propulsion
Laboratory.

My first intuition is that this
needed to be a very robust hand,

and I was thinking,
"Well, robust means big."

"Why don't you go find out how
big Wilt Chamberlain's hand is

and make a hand that big?"

So if anybody's curious,

Wilt Chamberlain's hand
is very, very big,

and not only is it
very, very big

when you make
a robotic version of it,

it actually cannot deal
with normal human-scale tools.

So they modify their design,

making a smaller hand
with just three fingers.

This does everything we needed
it to in the competition.

It grabbed everything,
the human tools.

So we thought we were
in pretty good shape.

But then,
at the first challenge,

Robosimian had to open a door,
and things got messy.

So here's an individual finger,
and in that individual finger,

there is an artificial tendon,
a synthetic fiber.

When excessive forces happen,
it snaps.

When Kennedy realized
Robosimian's fingers

couldn't provide the strength
he was looking for,

he came up with a radically
different idea.

So this is the "cam hand."

To do most of the work
that we need to,

a simple hook works just fine

so that we can actually
get a grasp

and we can hold on
to most everything we have to.

The fact of the matter is this
is a very simple, dumb system.

It closes
until it encounters something,

and then that will hold it,
and it'll hold it securely.

In their quest to grasp victory,

the teams competing
in the DARPA Challenge

will rely on all kinds of hands.

Many are using this gripper,
designed with three fingers

that can wrap around
a variety of objects

and even pick up
something small.

If we look at tasks like,
for instance, opening a door

or using a drill,
using hand tools and things,

pretty much all these tasks you
can perform using three fingers.

Robotic grippers
have already made their way

onto the factory floor,
attached to massive robotic arms

that are as robust
as they are precise.

They build cars,
lift heavy boxes,

pack beer,

sort through anything
and everything,

from batteries to pancakes,

doing the kind of jobs
many people consider repetitive

and downright boring.

But some experts fear

as robots move beyond the
factory into the real world,

they'll take on a lot more.

I think that we will lose
at least 20% or 30% of our jobs

over the next 20 years,
and probably more.

Fast food workers are mostly
going to be replaced.

The cashiers at Wal-Mart,

their jobs aren't going to last
that much longer.

The first to go
are going to be drivers.

Uber is spending a lot of money
on driverless cars.

Google is spending
a lot of money.

Now Toyota is.

40 years from now, there are not
going to be a lot of jobs.

But not everyone agrees
the future looks quite so grim.

Over the past century,
while robots and automation

took over
many manufacturing jobs,

other kinds of jobs
have increased.

Some think that trend
will continue.

Machines in all of modern
economic history

have helped create jobs,

not taken them away.

Machines are complements
to workers, not substitutes.

They come together and enhance
the productivity of each.

They need each other.

Today, more and more jobs
require humans

to work side by side with robots
on the assembly line,

programming them

and repairing them.

Technology creates new jobs.

But only for those who have
the skills to adapt.

What happens to the workers
left behind?

The worker who was displaced
has to be compensated

through retraining
or through a pension.

So that's a social problem,

not a problem rooted
in technology.

People should already
be thinking

about what kind of society
we would want

if not everybody can have jobs.

Many ethicists say
we should also be thinking

about which jobs we want
to hand over to machines.

Much of the things
that we are creating

can be used for a whole broad
range of potential applications,

ranging from eldercare
and childcare robots

to healthcare robotic platforms,
even surrogate sex objects.

What is acceptable?

What impact these
high-tech machines will have

in the workplace
and our homes remains unclear.

But one thing's for sure:

robots are starting to take
their first baby steps

out of the lab
and into the real world,

learning how to manipulate
objects we use every day.

But for a rescue robot
to be truly useful,

there's another hurdle
researchers face,

the toughest one of all:

the challenge
of giving a machine

the ability
to understand its environment...

To give it a brain.

Our robot doesn't really have
what you'd call a brain.

He's not seeing the world,
he's not perceiving the world.

It's not thinking,
it's not reasoning.

You know, it's pretty dumb.

In fact, today's rescue robots
are so dumb,

DARPA permits human operators
to guide them, step by step.

The bot waits for instructions
that tell it what to do...

Those zeros and ones
that computers understand.

The instructions travel
through an elaborate pipeline

that connects every piece
of hardware,

every motor and sensor,

to a computer controlled
by a team of human operators.

The operators communicate
with their robots

via a wireless connection,

just like they would
at a real disaster.

And at the final challenge.

These operators
at Carnegie Mellon University

are about to guide their
rescue robot, named CHIMP,

through one of the skills
it needs to master

for the final DARPA Challenge:
turning a valve.

Using six different cameras

located on the front and back
of its head

and a spinning sensor
called LIDAR,

the bot sends data about its
environment to the operators.

A 3-D image of CHIMP's world
appears on their monitors.

CHIMP has no idea
what it's looking at,

so with the click of a mouse,

the humans tell it
exactly where the valve is.

Things like recognizing objects,

that's a very hard problem
in robotics.

So we let the human
tell the difference

between a cat and a dog,
a valve and a door.

Next, they show CHIMP
where to grab the valve,

which way,
and how far to turn it.

But it would be impossibly slow
and impractical

for the operators to tell CHIMP
how to move every joint,

every sensor, every motor
of its complex arm.

This job must be done
by the robot all on its own.

Using a process called
"motion planning,"

CHIMP determines the best path
for its arm to travel.

It shows the operator
what its intentions are,

what its path will be,
how it's going to get there.

There's a plan.

In the end, it's going to be

you say,
"Okay, I approve your plan.

CHIMP, go ahead and do
your thing," and off it goes.

As we move towards developing
this technology further,

we really want to push
on the robot autonomy

and let CHIMP do more things
on its own,

but we always want to keep this
as a tool for a human.

In other words, keep the human
in the loop and in control.

But that's not
every team's goal.

At the Massachusetts
Institute of Technology,

Russ Tedrake is taking
a different approach.

He is determined to give
his Atlas robot

as much autonomy as he can
to make it a whole lot smarter.

Our goal as researchers,

especially in this artificial
intelligence lab,

we want to solve
the long-term research questions

about how to make
autonomous robots.

This is the closest
we've ever come

to building an artificial
intelligence machine,

where this humanoid robot
is moving through the world,

solving real problems.

At a disaster,

where you can't always count
on a wireless connection,

the smarter the bot gets,
the more it can do on its own.

To create a more
autonomous robot,

Tedrake is developing software
that helps it find

and identify objects
with a bit more independence.

So if the robot is just looking
at my kitchen at home,

there are dishes everywhere,

and you ask the robot
to find a spoon.

That's a really hard question.

If the human just says,

"There's a spoon roughly
over here, click,"

and he just has to look
in a little patch of space

for something that's roughly
the same shape as a spoon,

that takes an extremely hard
problem of object recognition

in a complicated environment

and turns it into a very
simple problem of,

"Okay, I want to look
for spoon-shaped things

in this small region of space."

While CHIMP needs
its human operators

to tell it exactly where
a valve is and where to grab it,

MIT's Atlas is programmed
to recognize a valve on its own,

figure out the steps it needs
to take as it approaches,

then grab and turn it
with very little human help.

We can put it
in an autonomous mode

and basically just watch
the robot execute.

It shows us
what it's about to do.

We could always stop it

if it looked like it was going
to do something wrong,

but when things are going well,

the robot's operating
almost completely autonomously.

There's no real input

coming from us
to the robot at this point.

It's just going on
with its task.

A more self-sufficient robot

could potentially help save
more lives in a disaster.

But just how independent
do we want rescue robots to get?

What if an autonomous robot
enters a disaster

and has to decide
who lives and who dies?

It's the kind of moral dilemma
that gave Will Smith nightmares

in the movie I, Robot.

You are in danger!

When a bot chose
to save his life

over the life
of an 11-year-old girl.

Save the girl! Save her!

I was the logical choice.

It calculated that I had
a 45% chance of survival.

Sarah only had an 11% chance.

11% is more than enough.

A human being
would have known that.

While today's cutting-edge
machines,

like the thousands of drones
used by the U.S. military,

still have a human in the loop,

what happens in the future
if they don't?

There's conversations
going on right now,

conversations about, what are
the ethical laws of robots?

How far should we really push
this technology?

How much autonomy do we want
to give this new technology?

Even the semi-autonomous ones

raise certain
interesting issues,

like, who's to blame
when something goes wrong?

These systems will not be
foolproof,

they will not be perfect.

It's important to remember that.

How these robotic technologies,
autonomous or not,

will be used on the battlefield
of the future

remains an open question.

But now, just a week
before the final challenge,

the roboticists are starting
to feel the mounting pressure.

Brett Kennedy,
who leads the Robosimian team,

is no exception.

I really have no idea

how we're going to place
within the overall field

at the robotics challenge.

All the researchers
that are bringing their teams

are top flight,

so where we end up in that,
I don't know.

Will Jerry Pratt's years
of research with bipedal robots

finally pay off?

Will Paul Oh and Junho Oh
redeem themselves

after HUBO's poor performance
at the first challenge?

Is CHIMP's combination
of hardware and software

just what
its human operators need?

Or will more autonomy help
or hinder the team from MIT?

There're so many things
that could go wrong.

Geez, the chance that
the robot could break,

or something that worked
99 out of 100 times,

but we get unlucky...

It's scary.

June 5, 2015.

Our roboticists meet once again,

this time at the Los Angeles
County Fairgrounds.

Over the next two days,

they'll face off with finalists
from around the world.

Several teams come
from Korea and Japan.

This little robot comes
from Germany.

Most are funded
through government

or corporate sponsorship,
but some bots,

like this odd-looking one
called Cog-Burn,

are funded by the teams
themselves.

Yeah, of course.

Very stiff competition,

looks like there are
a lot of good teams,

probably be a lot that make it
all the way through the course,

so it will come down
to top speed.

To win, each robot must perform
a series of tasks

a lot like the ones they faced
in the first challenge,

from driving a car,
to drilling a hole in a wall,

to walking over debris,
to tackling a flight of stairs.

The robot that completes
the course

in the shortest time wins.

That puts a lot of pressure

on the operators
to keep their bots moving.

And DARPA has thrown
another wrench into the mix.

Just like at a real disaster,

the power
of the wireless connection

between the operators
and their robots fluctuates.

Sometimes, the robots
will receive a degraded signal,

just bits and pieces of the data
that tells them what to do.

Other times, no data,
no instructions at all.

DARPA is hoping
this will push the roboticists

to develop
more autonomous systems,

bots that can finish a job
without our help.

But what makes
the final challenge

downright nerve-wracking is that
unlike the first competition,

where the bots were tethered,

this time around, there are
no safety lines allowed,

putting these multimillion-
dollar machines

in real jeopardy.

What I was worried about

is that almost none of the teams

had tested their robots
off the safety tether.

We had no idea
when these robots fell

how badly they would break,

so many of the teams
were really scared.

Finally, the competition begins.

Each robot has two chances
to run the course.

IHMC starts their first run.

The robot drives
without a hitch.

Walks up to the door with ease.

All right, go through the door.

All is looking good.

Just two more tasks to go.

Nice.

No!

It's Jerry's worst nightmare.

After years of perfecting
his walking software,

a tiny misstep in the real world

brought one of the most advanced
robots on earth to its knees.

Nobody was really sure

if the Atlas robot
would survive a fall.

You know, we've never had Atlas
fall before today.

The team worries the bot
can't be repaired

in time for its final run.

For Carnegie Mellon's robot,
CHIMP,

the day is also filled
with ups and downs.

When he fell down
through the door,

you know, he's laying there
and we were just, "Oh no."

Back in the garage,
the operators scramble

to find a way to get the red bot
back up on its treads again.

But the bigger a bot,
the harder it falls.

Smaller robots fall
without breaking

and have truly bizarre ways
of getting back up again.

But they're too small to do
the kind of heavy lifting

rescue robots need to do
at a disaster.

Once you make it heavier
and larger,

you need stronger motors,

and stronger motor
means heavier motors,

and heavier motors increase
the overall weights.

CHIMP weighs over 400 pounds.

That's an awful lot of robot
to lift.

The crowd begins to wonder
if its run is done.

I found it fascinating

to watch the crowd
watching the robots.

This was really
an emotional moment.

They're collections of wires
and gears and motors,

and we were sympathizing
with them at a very basic level.

Then CHIMP's leg begins to move.

This little kid
down in the front screams,

"He's getting up!"

The bot not only gets up,
it completes all eight tasks,

becoming the first robot
to finish the course.

But not every team is so lucky.

As MIT starts its run,

hoping to show off
its robot's autonomy,

things run amok
in the control room.

We made a simple operator error.

When we went
to get out of the car,

we forgot to turn off
the driving controller,

so the foot tried
to push the throttle

when it was getting out
of the car.

In an effort to drive and get
out of the car at the same time,

the robot keels over
and breaks its right arm.

But the operators

were able to recover

and do basically
the entire course left-handed.

The robot is still able to do
most of the tasks,

but it faces
one insurmountable problem:

the team has programmed it
to use two hands

to pick up the drill
and turn it on.

It can't do that
when one of its hands is broken.

Without completing
all the tasks,

team MIT has no chance
of winning.

A weary team IHMC

has worked through the night
trying to fix its robot.

Somehow, the robot's
still working,

but the fall we took
did something with the robot

where he's kind of
all messed up a little bit.

There's something wrong
with its LIDAR,

the system that scans
the robot's environment.

The operator is going to have
to adjust on the fly

and compensate in his head
for the errors in our sensors.

Are we really leaning that much?

Our strategy is still
what it was yesterday:

eight points or bust.

The time has come
for their final run.

All right!

They make it through
all the tasks

and approach the dreaded debris

where they fell
on their first run.

All right, you gotta
swing the right first.

But this time, they succeed.

The operator guides their robot
all the way up the stairs,

finishing the course
in record time.

Yes!

You know, having a human
in the loop,

you've got a supercomputer
up here.

Humans can adapt
to just about anything.

Eight points confirmed!

The team is now in the running
for first place.

But a one-of-a-kind bot

could still give them
a serious run for their money.

Cousins Paul and Junho are each
competing with their own robot,

doubling the chances that
one of their humanoids on wheels

will win the day.

Paul's team makes it
as far as the drilling task

until the drill gets stuck,
overheats, and shuts down.

But cousin Junho's team
moves on.

Their robot completes
all the manual tasks

by rolling on its knees
or standing on its feet.

But when it approaches
its final task, the stairs,

it suddenly stops.

Back in the garage,
its human handlers

are double- and triple-checking
the instructions

they're about to send
their robot.

A mistake here will cost them
the competition.

Their job is done.

All they can do now
is watch and wait.

Finally, HUBO starts
to climb the stairs

unlike any human on earth,

with its head facing forward
and its feet facing back.

The humanoid on wheels
finishes all eight tasks

faster than any other robot.

Junho Oh's robot HUBO
wins the day.

IHMC comes in second.

Team CHIMP, from Carnegie
Mellon, comes in third.

These bots may have moved
slowly, but just like a toddler,

they're taking baby steps
into our world.

We tend to think

the world today is what
it's going to be like

in ten years and twenty years.

But if we look back ten years
what we have,

look back twenty years
what we have,

the world has been
totally transformed.

It's really hard for us

to imagine how different
it's going to be.

In some ways,
the change in the field

has been very incremental,
but in other ways,

it's been completely
unpredictable.

It's too easy to look
at them and say,

"Oh, they're not there yet."

Well, they will get
to something very powerful.

And then you have to say, "Well,
where will we have gotten to?"

In fact, a few months
after the finals,

Boston Dynamics, the makers
of the Atlas robot,

took this updated version
for a stroll

in a snow-covered woods.

While it stumbles,

it quickly recovers its balance
just as we would.

I think we just need
to approach this

with our eyes open
and understand our vulnerability

to a technology that we are
on the cusp of creating.

There's no question
that developing rescue robots

with the potential to save lives
makes a lot of sense.

But the potential for other
applications remains unclear.

The time is now to think
about their role in our lives,

as we face
the "Rise of the Robots."