0:00:00.539,0:00:03.358
I know this is going to sound strange,

0:00:03.382,0:00:08.365
but I think robots can inspire us[br]to be better humans.

0:00:09.245,0:00:12.339
See, I grew up in Bethlehem, Pennsylvania,

0:00:12.363,0:00:14.301
the home of Bethlehem Steel.

0:00:14.991,0:00:17.103
My father was an engineer,

0:00:17.127,0:00:21.446
and when I was growing up,[br]he would teach me how things worked.

0:00:21.470,0:00:23.564
We would build projects together,

0:00:23.588,0:00:26.356
like model rockets and slot cars.

0:00:26.732,0:00:29.640
Here's the go-kart that we built together.

0:00:30.208,0:00:31.952
That's me behind the wheel,

0:00:31.976,0:00:34.200
with my sister and my best[br]friend at the time.

0:00:35.875,0:00:37.801
And one day,

0:00:37.825,0:00:40.908
he came home, when I was[br]about 10 years old,

0:00:40.932,0:00:43.558
and at the dinner table, he announced

0:00:43.582,0:00:47.294
that for our next project, [br]we were going to build ...

0:00:47.318,0:00:48.468
a robot.

0:00:49.604,0:00:50.759
A robot.

0:00:51.120,0:00:53.183
Now, I was thrilled about this,

0:00:53.207,0:00:57.009
because at school,[br]there was a bully named Kevin,

0:00:57.033,0:00:58.922
and he was picking on me,

0:00:58.946,0:01:01.336
because I was the only[br]Jewish kid in class.

0:01:01.910,0:01:04.283
So I couldn't wait to get[br]started to work on this,

0:01:04.307,0:01:06.917
so I could introduce Kevin to my robot.

0:01:06.941,0:01:07.971
(Laughter)

0:01:07.995,0:01:12.238
(Robot noises)

0:01:17.552,0:01:18.944
(Laughter)

0:01:18.968,0:01:22.817
But that wasn't the kind of robot[br]my dad had in mind.

0:01:22.841,0:01:23.909
(Laughter)

0:01:23.933,0:01:27.565
See, he owned a chromium-plating company,

0:01:27.589,0:01:33.462
and they had to move heavy steel parts[br]between tanks of chemicals.

0:01:33.486,0:01:36.859
And so he needed[br]an industrial robot like this,

0:01:36.883,0:01:38.986
that could basically do the heavy lifting.

0:01:39.684,0:01:43.275
But my dad didn't get[br]the kind of robot he wanted, either.

0:01:43.870,0:01:46.300
He and I worked on it for several years,

0:01:46.324,0:01:51.069
but it was the 1970s, and the technology[br]that was available to amateurs

0:01:51.093,0:01:52.419
just wasn't there yet.

0:01:53.992,0:01:56.777
So Dad continued to do[br]this kind of work by hand.

0:01:57.595,0:01:59.023
And a few years later,

0:01:59.730,0:02:01.306
he was diagnosed with cancer.

0:02:03.822,0:02:05.392
You see,

0:02:05.416,0:02:08.012
what the robot we were trying[br]to build was telling him

0:02:08.036,0:02:10.192
was not about doing the heavy lifting.

0:02:10.216,0:02:11.367
It was a warning

0:02:11.391,0:02:13.895
about his exposure to the toxic chemicals.

0:02:15.184,0:02:17.477
He didn't recognize that at the time,

0:02:17.501,0:02:19.187
and he contracted leukemia.

0:02:19.827,0:02:21.719
And he died at the age of 45.

0:02:23.354,0:02:25.130
I was devastated by this.

0:02:25.703,0:02:28.465
And I never forgot the robot[br]that he and I tried to build.

0:02:30.195,0:02:33.748
When I was at college, I decided[br]to study engineering, like him.

0:02:35.081,0:02:39.629
And I went to Carnegie Mellon,[br]and I earned my PhD in robotics.

0:02:39.653,0:02:41.562
I've been studying robots ever since.

0:02:42.721,0:02:47.316
So what I'd like to tell you about[br]are four robot projects,

0:02:47.340,0:02:50.389
and how they've inspired me[br]to be a better human.

0:02:54.206,0:02:59.820
By 1993, I was a young professor at USC,

0:02:59.844,0:03:02.662
and I was just building up[br]my own robotics lab,

0:03:02.686,0:03:05.907
and this was the year[br]the World Wide Web came out.

0:03:06.487,0:03:09.472
And I remember my students[br]were the ones who told me about it,

0:03:09.496,0:03:11.813
and we would -- we were just amazed.

0:03:11.837,0:03:15.220
We started playing with this,[br]and that afternoon,

0:03:15.244,0:03:19.522
we realized that we could use[br]this new, universal interface

0:03:19.546,0:03:23.974
to allow anyone in the world[br]to operate the robot in our lab.

0:03:25.339,0:03:29.381
So, rather than have it fight[br]or do industrial work,

0:03:30.413,0:03:32.943
we decided to build a planter,

0:03:32.967,0:03:34.918
put the robot into the center of it,

0:03:34.942,0:03:36.620
and we called it the Telegarden.

0:03:37.736,0:03:42.253
And we had put a camera[br]in the gripper of the hand of the robot,

0:03:42.277,0:03:45.208
and we wrote some[br]special scripts and software,

0:03:45.232,0:03:47.266
so that anyone in the world could come in,

0:03:47.290,0:03:49.074
and by clicking on the screen,

0:03:49.098,0:03:52.719
they could move the robot around[br]and visit the garden.

0:03:53.416,0:03:57.267
But we also set up some other software

0:03:57.291,0:04:01.207
that lets you participate[br]and help us water the garden, remotely.

0:04:01.643,0:04:03.932
And if you watered it a few times,

0:04:03.956,0:04:06.231
we'd give you your own seed to plant.

0:04:07.167,0:04:10.643
Now, this was an engineering project,

0:04:10.667,0:04:14.779
and we published some papers[br]on the system design of it,

0:04:14.803,0:04:17.448
but we also thought of it[br]as an art installation.

0:04:18.638,0:04:20.787
It was invited, after the first year,

0:04:20.811,0:04:23.831
by the Ars Electronica Museum in Austria,

0:04:23.855,0:04:26.677
to have it installed in their lobby.

0:04:27.558,0:04:31.545
And I'm happy to say, it remained[br]online there, 24 hours a day,

0:04:31.569,0:04:33.400
for almost nine years.

0:04:34.369,0:04:38.144
That robot was operated by more people

0:04:38.168,0:04:40.065
than any other robot in history.

0:04:41.303,0:04:42.930
Now, one day,

0:04:42.954,0:04:46.302
I got a call out of the blue[br]from a student,

0:04:47.211,0:04:51.369
who asked a very simple[br]but profound question.

0:04:52.361,0:04:55.187
He said, "Is the robot real?"

0:04:56.599,0:04:58.885
Now, everyone else had assumed it was,

0:04:58.909,0:05:01.425
and we knew it was,[br]because we were working with it.

0:05:01.449,0:05:02.847
But I knew what he meant,

0:05:02.871,0:05:04.313
because it would be possible

0:05:04.337,0:05:07.122
to take a bunch of pictures[br]of flowers in a garden

0:05:07.146,0:05:10.785
and then, basically, index them[br]in a computer system,

0:05:10.809,0:05:13.436
such that it would appear[br]that there was a real robot,

0:05:13.460,0:05:14.610
when there wasn't.

0:05:15.088,0:05:16.612
And the more I thought about it,

0:05:16.636,0:05:20.215
I couldn't think of a good answer[br]for how he could tell the difference.

0:05:20.628,0:05:23.425
This was right about the time[br]that I was offered a position

0:05:23.449,0:05:24.750
here at Berkeley.

0:05:24.774,0:05:26.419
And when I got here,

0:05:26.443,0:05:28.805
I looked up Hubert Dreyfus,

0:05:28.829,0:05:31.756
who's a world-renowned[br]professor of philosophy,

0:05:32.582,0:05:35.312
And I talked with him[br]about this and he said,

0:05:35.336,0:05:39.438
"This is one of the oldest[br]and most central problems in philosophy.

0:05:39.462,0:05:43.479
It goes back to the Skeptics[br]and up through Descartes.

0:05:43.503,0:05:46.685
It's the issue of epistemology,

0:05:46.709,0:05:49.577
the study of how do we know[br]that something is true."

0:05:50.625,0:05:52.625
So he and I started working together,

0:05:52.649,0:05:55.520
and we coined a new term:[br]"telepistemology,"

0:05:56.757,0:05:58.893
the study of knowledge at a distance.

0:05:59.303,0:06:03.794
We invited leading artists,[br]engineers and philosophers

0:06:03.818,0:06:05.137
to write essays about this,

0:06:05.161,0:06:08.865
and the results are collected[br]in this book from MIT Press.

0:06:09.959,0:06:12.055
So thanks to this student,

0:06:12.079,0:06:15.398
who questioned what everyone else[br]had assumed to be true,

0:06:15.422,0:06:19.399
this project taught me[br]an important lesson about life,

0:06:19.423,0:06:22.410
which is to always question assumptions.

0:06:23.807,0:06:25.983
Now, the second project[br]I'll tell you about

0:06:26.007,0:06:27.999
grew out of the Telegarden.

0:06:28.023,0:06:30.844
As it was operating, my students[br]and I were very interested

0:06:30.868,0:06:33.209
in how people were interacting[br]with each other,

0:06:33.233,0:06:35.290
and what they were doing with the garden.

0:06:35.314,0:06:36.465
So we started thinking:

0:06:36.489,0:06:38.457
what if the robot could leave the garden

0:06:38.481,0:06:41.295
and go out into some other[br]interesting environment?

0:06:41.319,0:06:44.060
Like, for example,[br]what if it could go to a dinner party

0:06:44.084,0:06:45.545
at the White House?

0:06:45.569,0:06:47.068
(Laughter)

0:06:48.401,0:06:51.970
So, because we were interested[br]more in the system design

0:06:51.994,0:06:54.937
and the user interface[br]than in the hardware,

0:06:54.961,0:06:56.112
we decided that,

0:06:56.136,0:07:00.623
rather than have a robot replace[br]the human to go to the party,

0:07:00.647,0:07:02.877
we'd have a human replace the robot.

0:07:03.679,0:07:05.139
We called it the Tele-Actor.

0:07:05.978,0:07:07.986
We got a human,

0:07:08.010,0:07:11.118
someone who's very[br]outgoing and gregarious,

0:07:11.142,0:07:15.397
and she was outfitted with a helmet[br]with various equipment,

0:07:15.421,0:07:16.705
cameras and microphones,

0:07:16.729,0:07:19.794
and then a backpack with wireless[br]Internet connection.

0:07:20.882,0:07:23.026
And the idea was that she could go

0:07:23.050,0:07:25.500
into a remote and interesting environment,

0:07:25.524,0:07:27.541
and then over the Internet,

0:07:27.565,0:07:30.157
people could experience[br]what she was experiencing.

0:07:30.736,0:07:33.707
So they could see what she was seeing,

0:07:33.731,0:07:37.014
but then, more importantly,[br]they could participate,

0:07:37.038,0:07:41.788
by interacting with each other[br]and coming up with ideas

0:07:41.812,0:07:45.781
about what she should do next[br]and where she should go,

0:07:45.805,0:07:48.011
and then conveying those[br]to the Tele-Actor.

0:07:49.069,0:07:51.490
So we got a chance to take the Tele-Actor

0:07:51.514,0:07:54.322
to the Webby Awards in San Francisco.

0:07:55.129,0:07:58.369
And that year, Sam Donaldson was the host.

0:08:00.290,0:08:02.980
Just before the curtain went[br]up, I had about 30 seconds

0:08:03.004,0:08:06.746
to explain to Mr. Donaldson[br]what we were going to do.

0:08:08.119,0:08:11.636
And I said, "The Tele-Actor[br]is going to be joining you onstage.

0:08:11.660,0:08:14.014
This is a new experimental project,

0:08:14.038,0:08:16.633
and people are watching her[br]on their screens,

0:08:16.657,0:08:19.948
there's cameras involved[br]and there's microphones

0:08:19.972,0:08:21.758
and she's got an earbud in her ear,

0:08:21.782,0:08:24.099
and people over the network[br]are giving her advice

0:08:24.123,0:08:25.281
about what to do next."

0:08:25.305,0:08:26.700
And he said, "Wait a second.

0:08:27.806,0:08:28.979
That's what I do."

0:08:29.003,0:08:33.980
(Laughter)

0:08:34.004,0:08:35.876
So he loved the concept,

0:08:35.900,0:08:40.153
and when the Tele-Actor walked onstage,[br]she walked right up to him,

0:08:40.177,0:08:42.636
and she gave him a big kiss[br]right on the lips.

0:08:42.660,0:08:44.612
(Laughter)

0:08:44.636,0:08:47.558
We were totally surprised --[br]we had no idea that would happen.

0:08:47.582,0:08:50.200
And he was great, he just gave her[br]a big hug in return,

0:08:50.224,0:08:51.881
and it worked out great.

0:08:51.905,0:08:53.945
But that night, as we were packing up,

0:08:53.969,0:08:58.524
I asked the Tele-Actor,[br]how did the Tele-Directors decide

0:08:58.548,0:09:01.548
that they would give[br]a kiss to Sam Donaldson?

0:09:03.135,0:09:04.592
And she said they hadn't.

0:09:05.274,0:09:07.892
She said, when she was[br]just about to walk onstage,

0:09:07.916,0:09:10.890
the Tele-Directors still were trying[br]to agree on what to do,

0:09:10.914,0:09:14.014
and so she just walked onstage[br]and did what felt most natural.

0:09:14.038,0:09:18.552
(Laughter)

0:09:18.576,0:09:21.668
So, the success[br]of the Tele-Actor that night

0:09:21.692,0:09:25.945
was due to the fact[br]that she was a wonderful actor.

0:09:25.969,0:09:28.398
She knew when to trust her instincts.

0:09:28.422,0:09:32.351
And so that project taught me[br]another lesson about life,

0:09:32.375,0:09:36.022
which is that, when in doubt, improvise.

0:09:36.046,0:09:37.712
(Laughter)

0:09:38.664,0:09:43.583
Now, the third project[br]grew out of my experience

0:09:43.607,0:09:45.456
when my father was in the hospital.

0:09:47.284,0:09:50.637
He was undergoing a treatment --[br]chemotherapy treatments --

0:09:50.661,0:09:54.931
and there's a related treatment[br]called brachytherapy,

0:09:54.955,0:09:58.743
where tiny, radioactive seeds[br]are placed into the body

0:09:58.767,0:10:00.501
to treat cancerous tumors.

0:10:01.572,0:10:03.843
And the way it's done,[br]as you can see here,

0:10:03.867,0:10:08.234
is that surgeons[br]insert needles into the body

0:10:08.258,0:10:09.573
to deliver the seeds.

0:10:09.974,0:10:13.672
And all these needles[br]are inserted in parallel.

0:10:14.445,0:10:19.838
So it's very common that some[br]of the needles penetrate sensitive organs.

0:10:20.635,0:10:27.608
And as a result, the needles damage[br]these organs, cause damage,

0:10:27.632,0:10:30.041
which leads to trauma and side effects.

0:10:30.581,0:10:32.225
So my students and I wondered:

0:10:32.249,0:10:36.771
what if we could modify the system,

0:10:36.795,0:10:39.420
so that the needles[br]could come in at different angles?

0:10:40.395,0:10:42.238
So we simulated this;

0:10:42.262,0:10:45.707
we developed some optimization[br]algorithms and we simulated this.

0:10:45.731,0:10:46.882
And we were able to show

0:10:46.906,0:10:49.468
that we are able to avoid[br]the delicate organs,

0:10:49.492,0:10:54.545
and yet still achieve the coverage[br]of the tumors with the radiation.

0:10:55.313,0:10:58.787
So now, we're working with doctors at UCSF

0:10:58.811,0:11:01.078
and engineers at Johns Hopkins,

0:11:01.102,0:11:05.064
and we're building a robot[br]that has a number of --

0:11:05.088,0:11:07.687
it's a specialized design[br]with different joints

0:11:07.711,0:11:12.034
that can allow the needles to come in[br]at an infinite variety of angles.

0:11:12.483,0:11:16.205
And as you can see here,[br]they can avoid delicate organs

0:11:16.229,0:11:18.893
and still reach the targets[br]they're aiming for.

0:11:20.019,0:11:25.004
So, by questioning this assumption[br]that all the needles have to be parallel,

0:11:25.028,0:11:27.654
this project also taught me[br]an important lesson:

0:11:28.114,0:11:32.869
When in doubt, when your path[br]is blocked, pivot.

0:11:33.797,0:11:37.637
And the last project[br]also has to do with medical robotics.

0:11:38.187,0:11:41.757
And this is something[br]that's grown out of a system

0:11:41.781,0:11:45.129
called the da Vinci surgical robot.

0:11:45.866,0:11:48.310
And this is a commercially[br]available device.

0:11:48.334,0:11:51.307
It's being used in over 2,000[br]hospitals around the world.

0:11:52.013,0:11:56.386
The idea is it allows the surgeon[br]to operate comfortably

0:11:56.410,0:11:58.187
in his own coordinate frame.

0:12:00.762,0:12:06.531
Many of the subtasks in surgery are very[br]routine and tedious, like suturing,

0:12:06.555,0:12:08.822
and currently, all of these are performed

0:12:08.846,0:12:12.713
under the specific and immediate[br]control of the surgeon.

0:12:13.374,0:12:15.642
So the surgeon becomes fatigued over time.

0:12:16.086,0:12:17.522
And we've been wondering,

0:12:17.546,0:12:21.917
what if we could program the robot[br]to perform some of these subtasks,

0:12:21.941,0:12:23.322
and thereby free the surgeon

0:12:23.346,0:12:26.581
to focus on the more complicated[br]parts of the surgery,

0:12:26.605,0:12:29.405
and also cut down on the time[br]that the surgery would take

0:12:29.429,0:12:32.268
if we could get the robot[br]to do them a little bit faster?

0:12:32.958,0:12:36.380
Now, it's hard to program a robot[br]to do delicate things like this.

0:12:36.943,0:12:41.149
But it turns out my colleague[br]Pieter Abbeel, who's here at Berkeley,

0:12:41.173,0:12:46.653
has developed a new set of techniques[br]for teaching robots from example.

0:12:47.170,0:12:49.913
So he's gotten robots to fly helicopters,

0:12:49.937,0:12:53.191
do incredibly interesting,[br]beautiful acrobatics,

0:12:53.215,0:12:55.182
by watching human experts fly them.

0:12:56.152,0:12:58.058
So we got one of these robots.

0:12:58.082,0:13:00.622
We started working with Pieter[br]and his students.

0:13:00.646,0:13:04.826
And we asked a surgeon[br]to perform a task --

0:13:06.761,0:13:07.912
with the robot.

0:13:07.936,0:13:10.961
So what we're doing is asking[br]the surgeon to perform the task,

0:13:10.985,0:13:13.021
and we record the motions of the robot.

0:13:13.045,0:13:14.471
So here's an example.

0:13:14.495,0:13:17.616
I'll use tracing out[br]a figure eight as an example.

0:13:18.170,0:13:21.804
So here's what it looks like[br]when the robot --

0:13:21.828,0:13:24.896
this is what the robot's path[br]looks like, those three examples.

0:13:24.920,0:13:29.056
Now, those are much better[br]than what a novice like me could do,

0:13:29.080,0:13:31.774
but they're still jerky and imprecise.

0:13:31.798,0:13:34.739
So we record all these examples, the data,

0:13:34.763,0:13:37.624
and then go through a sequence of steps.

0:13:38.278,0:13:41.404
First, we use a technique[br]called dynamic time warping

0:13:41.428,0:13:42.918
from speech recognition.

0:13:42.942,0:13:46.279
And this allows us to temporally[br]align all of the examples.

0:13:46.918,0:13:52.120
And then we apply Kalman filtering,[br]a technique from control theory,

0:13:52.144,0:13:55.113
that allows us to statistically[br]analyze all the noise

0:13:55.137,0:13:59.259
and extract the desired[br]trajectory that underlies them.

0:14:01.283,0:14:03.833
Now we take those human demonstrations --

0:14:03.857,0:14:05.532
they're all noisy and imperfect --

0:14:05.556,0:14:08.154
and we extract from them[br]an inferred task trajectory

0:14:08.178,0:14:10.704
and control sequence for the robot.

0:14:11.181,0:14:13.497
We then execute that on the robot,

0:14:13.521,0:14:15.546
we observe what happens,

0:14:15.570,0:14:16.927
then we adjust the controls,

0:14:16.951,0:14:19.747
using a sequence of techniques[br]called iterative learning.

0:14:21.129,0:14:24.538
Then what we do is we increase[br]the velocity a little bit.

0:14:25.244,0:14:28.379
We observe the results,[br]adjust the controls again,

0:14:29.340,0:14:31.112
and observe what happens.

0:14:31.136,0:14:33.303
And we go through this several rounds.

0:14:33.327,0:14:34.508
And here's the result.

0:14:34.968,0:14:36.706
That's the inferred task trajectory,

0:14:36.730,0:14:39.951
and here's the robot[br]moving at the speed of the human.

0:14:39.975,0:14:42.080
Here's four times the speed of the human.

0:14:42.477,0:14:43.961
Here's seven times.

0:14:45.004,0:14:49.829
And here's the robot operating[br]at 10 times the speed of the human.

0:14:50.762,0:14:53.663
So we're able to get a robot[br]to perform a delicate task

0:14:53.687,0:14:56.573
like a surgical subtask,

0:14:57.081,0:14:58.962
at 10 times the speed of a human.

0:15:00.185,0:15:02.445
So this project also,

0:15:02.469,0:15:04.994
because of its involved[br]practicing and learning,

0:15:05.018,0:15:06.780
doing something over and over again,

0:15:06.804,0:15:09.381
this project also has a lesson, which is:

0:15:09.405,0:15:11.809
if you want to do something well,

0:15:13.084,0:15:17.524
there's no substitute[br]for practice, practice, practice.

0:15:21.186,0:15:24.673
So these are four of the lessons[br]that I've learned from robots

0:15:24.697,0:15:25.995
over the years.

0:15:27.312,0:15:32.819
And the field of robotics[br]has gotten much better over time.

0:15:34.319,0:15:36.595
Nowadays, high school students[br]can build robots,

0:15:36.619,0:15:39.503
like the industrial robot[br]my dad and I tried to build.

0:15:40.675,0:15:43.445
But, it's very -- now ...

0:15:43.954,0:15:47.255
And now, I have a daughter,

0:15:47.857,0:15:49.014
named Odessa.

0:15:49.673,0:15:50.926
She's eight years old.

0:15:51.682,0:15:53.331
And she likes robots, too.

0:15:53.970,0:15:55.366
Maybe it runs in the family.

0:15:55.390,0:15:56.630
(Laughter)

0:15:56.654,0:15:58.749
I wish she could meet my dad.

0:16:00.303,0:16:03.043
And now I get to teach her[br]how things work,

0:16:03.067,0:16:05.297
and we get to build projects together.

0:16:05.321,0:16:08.527
And I wonder what kind of lessons[br]she'll learn from them.

0:16:10.147,0:16:14.154
Robots are the most human of our machines.

0:16:14.983,0:16:17.937
They can't solve all[br]of the world's problems,

0:16:17.961,0:16:21.484
but I think they have something[br]important to teach us.

0:16:22.276,0:16:24.277
I invite all of you

0:16:24.301,0:16:27.879
to think about the innovations[br]that you're interested in,

0:16:28.712,0:16:30.955
the machines that you wish for.

0:16:31.660,0:16:34.267
And think about[br]what they might be telling you.

0:16:35.211,0:16:38.956
Because I have a hunch that many[br]of our technological innovations,

0:16:38.980,0:16:40.651
the devices we dream about,

0:16:42.166,0:16:45.015
can inspire us to be better humans.

0:16:45.984,0:16:47.135
Thank you.

0:16:47.159,0:16:49.071
(Applause)