WEBVTT

00:00:00.498 --> 00:00:03.097
Five years ago, I was a Ph.D. student

00:00:03.097 --> 00:00:04.741
living two lives.

00:00:04.741 --> 00:00:07.057
In one, I used NASA supercomputers

00:00:07.057 --> 00:00:09.785
to design next-generation spacecraft,

00:00:09.785 --> 00:00:12.364
and in the other I was a data scientist

00:00:12.364 --> 00:00:14.783
looking for potential smugglers

00:00:14.783 --> 00:00:18.022
of sensitive nuclear technologies.

00:00:18.022 --> 00:00:20.881
As a data scientist, I did a lot of analyses,

00:00:20.881 --> 00:00:22.348
mostly of facilities,

00:00:22.348 --> 00:00:24.619
industrial facilities around the world.

00:00:24.619 --> 00:00:27.307
And I was always looking for a better canvas

00:00:27.307 --> 00:00:29.373
to tie these all together.

00:00:29.373 --> 00:00:31.239
And one day, I was thinking about how

00:00:31.239 --> 00:00:33.363
all data has a location,

00:00:33.363 --> 00:00:35.096
and I realized that the answer

00:00:35.096 --> 00:00:36.984
had been staring me in the face.

00:00:36.984 --> 00:00:39.622
Although I was a satellite engineer,

00:00:39.622 --> 00:00:42.793
I hadn't thought about using satellite imagery

00:00:42.793 --> 00:00:44.359
in my work.

NOTE Paragraph

00:00:44.359 --> 00:00:46.306
Now, like most of us, I'd been online,

00:00:46.306 --> 00:00:48.088
I'd see my house, so I thought,

00:00:48.088 --> 00:00:50.386
I'll hop in there and I'll start looking up

00:00:50.386 --> 00:00:51.803
some of these facilities.

00:00:51.803 --> 00:00:53.772
And what I found really surprised me.

00:00:53.772 --> 00:00:55.446
The pictures that I was finding

00:00:55.446 --> 00:00:57.347
were years out of date,

00:00:57.347 --> 00:00:58.415
and because of that,

00:00:58.415 --> 00:01:00.086
it had relatively little relevance

00:01:00.086 --> 00:01:02.546
to the work that I was doing today.

00:01:02.546 --> 00:01:04.007
But I was intrigued.

00:01:04.007 --> 00:01:07.243
I mean, satellite imagery is pretty amazing stuff.

00:01:07.243 --> 00:01:09.546
There are millions and millions of sensors

00:01:09.546 --> 00:01:10.990
surrounding us today,

00:01:10.990 --> 00:01:13.950
but there's still so much we
don't know on a daily basis.

00:01:13.950 --> 00:01:17.703
How much oil is stored in all of China?

00:01:17.703 --> 00:01:20.636
How much corn is being produced?

00:01:20.636 --> 00:01:24.707
How many ships are in all of our world's ports?

00:01:24.707 --> 00:01:27.266
Now, in theory, all of these questions

00:01:27.266 --> 00:01:29.510
could be answered by imagery,

00:01:29.510 --> 00:01:31.237
but not if it's old.

00:01:31.237 --> 00:01:33.695
And if this data was so valuable,

00:01:33.695 --> 00:01:35.627
then how come I couldn't get my hands

00:01:35.627 --> 00:01:38.177
on more recent pictures?

NOTE Paragraph

00:01:38.177 --> 00:01:41.312
So the story begins over 50 years ago

00:01:41.312 --> 00:01:43.183
with the launch of the first generation

00:01:43.183 --> 00:01:46.783
of U.S. government photo reconnaissance satellites.

00:01:46.783 --> 00:01:48.479
And today, there's a handful

00:01:48.479 --> 00:01:50.674
of the great, great grandchildren

00:01:50.674 --> 00:01:52.339
of these early Cold War machines

00:01:52.339 --> 00:01:54.348
which are now operated by private companies

00:01:54.348 --> 00:01:57.057
and from which the vast majority of satellite imagery

00:01:57.057 --> 00:01:59.882
that you and I see on a daily basis comes.

00:01:59.882 --> 00:02:02.796
During this period, launching things into space,

00:02:02.796 --> 00:02:05.473
just the rocket to get the satellite up there,

00:02:05.473 --> 00:02:10.074
has cost hundreds of millions of dollars each,

00:02:10.074 --> 00:02:11.826
and that's created tremendous pressure

00:02:11.826 --> 00:02:14.115
to launch things infrequently

00:02:14.115 --> 00:02:15.459
and to make sure that when you do,

00:02:15.459 --> 00:02:19.098
you cram as much functionality in there as possible.

00:02:19.098 --> 00:02:20.709
All of this has only made satellites

00:02:20.709 --> 00:02:23.126
bigger and bigger and bigger

00:02:23.126 --> 00:02:24.724
and more expensive,

00:02:24.724 --> 00:02:29.761
now nearly a billion, with a b, dollars per copy.

00:02:29.761 --> 00:02:31.460
Because they are so expensive,

00:02:31.460 --> 00:02:32.927
there aren't very many of them.

00:02:32.927 --> 00:02:34.272
Because there aren't very many of them,

00:02:34.272 --> 00:02:36.728
the pictures that we see on a daily basis

00:02:36.728 --> 00:02:38.232
tend to be old.

NOTE Paragraph

00:02:38.232 --> 00:02:41.514
I think a lot of people actually
understand this anecdotally,

00:02:41.514 --> 00:02:44.053
but in order to visualize just how sparsely

00:02:44.053 --> 00:02:45.937
our planet is collected,

00:02:45.937 --> 00:02:48.003
some friends and I put together a dataset

00:02:48.003 --> 00:02:50.590
of the 30 million pictures that have been gathered

00:02:50.590 --> 00:02:54.074
by these satellites between 2000 and 2010.

00:02:54.074 --> 00:02:56.829
As you can see in blue, huge areas of our world

00:02:56.829 --> 00:02:59.619
are barely seen, less than once a year,

00:02:59.619 --> 00:03:01.891
and even the areas that are seen most frequently,

00:03:01.891 --> 00:03:05.639
those in red, are seen at best once a quarter.

00:03:05.639 --> 00:03:08.555
Now as aerospace engineering grad students,

00:03:08.555 --> 00:03:11.983
this chart cried out to us as a challenge.

00:03:11.983 --> 00:03:15.438
Why do these things have to be so expensive?

00:03:15.438 --> 00:03:18.173
Does a single satellite really have to cost

00:03:18.173 --> 00:03:22.936
the equivalent of three 747 jumbo jets?

00:03:22.936 --> 00:03:25.099
Wasn't there a way to build a smaller,

00:03:25.099 --> 00:03:28.271
simpler, new satellite design that could enable

00:03:28.271 --> 00:03:30.289
more timely imaging?

NOTE Paragraph

00:03:30.289 --> 00:03:33.740
I realize that it does sound a little bit crazy

00:03:33.740 --> 00:03:35.251
that we were going to go out and just

00:03:35.251 --> 00:03:37.098
begin designing satellites,

00:03:37.098 --> 00:03:39.061
but fortunately we had help.

00:03:39.061 --> 00:03:41.996
In the late 1990s, a couple of professors

00:03:41.996 --> 00:03:45.477
proposed a concept for radically reducing the price

00:03:45.477 --> 00:03:47.249
of putting things in space.

00:03:47.249 --> 00:03:49.360
This was hitchhiking small satellites

00:03:49.360 --> 00:03:52.376
alongside much larger satellites.

00:03:52.376 --> 00:03:55.285
This dropped the cost of putting objects up there

00:03:55.285 --> 00:03:57.283
by over a factor of 100,

00:03:57.283 --> 00:04:00.369
and suddenly we could afford to experiment,

00:04:00.369 --> 00:04:01.837
to take a little bit of risk,

00:04:01.837 --> 00:04:04.103
and to realize a lot of innovation.

00:04:04.103 --> 00:04:07.192
And a new generation of engineers and scientists,

00:04:07.192 --> 00:04:08.791
mostly out of universities,

00:04:08.791 --> 00:04:11.134
began launching these very small,

00:04:11.134 --> 00:04:13.725
breadbox-sized satellites called CubeSats.

00:04:13.725 --> 00:04:16.128
And these were built with electronics obtained

00:04:16.128 --> 00:04:20.026
from RadioShack instead of Lockheed Martin.

NOTE Paragraph

00:04:20.026 --> 00:04:22.866
Now it was using the lessons
learned from these early missions

00:04:22.866 --> 00:04:25.561
that my friends and I began a series of sketches

00:04:25.561 --> 00:04:27.250
of our own satellite design.

00:04:27.250 --> 00:04:30.034
And I can't remember a specific day

00:04:30.034 --> 00:04:31.894
where we made a conscious decision

00:04:31.894 --> 00:04:34.529
that we were actually going to
go out and build these things,

00:04:34.529 --> 00:04:36.959
but once we got that idea in our minds

00:04:36.959 --> 00:04:39.120
of the world as a dataset,

00:04:39.120 --> 00:04:41.610
of being able to capture millions of data points

00:04:41.610 --> 00:04:44.663
on a daily basis describing the global economy,

00:04:44.663 --> 00:04:47.183
of being able to unearth billions of connections

00:04:47.183 --> 00:04:50.348
between them that had never before been found,

00:04:50.348 --> 00:04:51.996
it just seemed boring

00:04:51.996 --> 00:04:54.903
to go work on anything else.

NOTE Paragraph

00:04:54.903 --> 00:04:57.876
And so we moved into a cramped,

00:04:57.876 --> 00:05:00.855
windowless office in Palo Alto,

00:05:00.855 --> 00:05:02.822
and began working to take our design

00:05:02.822 --> 00:05:05.760
from the drawing board into the lab.

00:05:05.760 --> 00:05:08.418
The first major question we had to tackle

00:05:08.418 --> 00:05:10.752
was just how big to build this thing.

00:05:10.752 --> 00:05:13.935
In space, size drives cost,

00:05:13.935 --> 00:05:15.948
and we had worked with these very small,

00:05:15.948 --> 00:05:18.042
breadbox-sized satellites in school,

00:05:18.042 --> 00:05:20.593
but as we began to better
understand the laws of physics,

00:05:20.593 --> 00:05:22.715
we found that the quality of pictures

00:05:22.715 --> 00:05:25.799
those satellites could take was very limited,

00:05:25.799 --> 00:05:27.503
because the laws of physics dictate

00:05:27.503 --> 00:05:30.609
that the best picture you
can take through a telescope

00:05:30.609 --> 00:05:32.848
is a function of the diameter of that telescope,

00:05:32.848 --> 00:05:34.510
and these satellites had a very small,

00:05:34.510 --> 00:05:36.216
very constrained volume.

00:05:36.216 --> 00:05:38.011
And we found that the best picture we would

00:05:38.011 --> 00:05:40.496
have been able to get looked something like this.

00:05:40.496 --> 00:05:42.741
Although this was the low-cost option,

00:05:42.741 --> 00:05:44.414
quite frankly it was just too blurry

00:05:44.414 --> 00:05:47.983
to see the things that make
satellite imagery valuable.

00:05:47.983 --> 00:05:50.129
So about three or four weeks later,

00:05:50.129 --> 00:05:52.946
we met a group of engineers randomly

00:05:52.946 --> 00:05:54.587
who had worked on the first

00:05:54.587 --> 00:05:57.028
private imaging satellite ever developed,

00:05:57.028 --> 00:05:59.146
and they told us that back in the 1970s,

00:05:59.146 --> 00:06:01.328
the U.S. government had found a powerful

00:06:01.328 --> 00:06:02.971
optimal tradeoff --

00:06:02.971 --> 00:06:06.059
that in taking pictures at right
about one meter resolution,

00:06:06.059 --> 00:06:08.943
being able to see objects one meter in size,

00:06:08.943 --> 00:06:11.634
they had found that they could not
just get very high-quality images,

00:06:11.634 --> 00:06:14.527
but get a lot of them at an affordable price.

00:06:14.527 --> 00:06:16.503
From our own computer simulations,

00:06:16.503 --> 00:06:18.508
we quickly found that one meter really was

00:06:18.508 --> 00:06:20.272
the minimum viable product

00:06:20.272 --> 00:06:23.351
to be able to see the drivers of our global economy,

00:06:23.351 --> 00:06:24.912
for the first time, being able to count

00:06:24.912 --> 00:06:27.633
the ships and cars and shipping
containers and trucks

00:06:27.633 --> 00:06:30.457
that move around our world on a daily basis,

00:06:30.457 --> 00:06:34.261
while conveniently still not
being able to see individuals.

00:06:34.261 --> 00:06:35.882
We had found our compromise.

00:06:35.882 --> 00:06:37.486
We would have to build something larger

00:06:37.486 --> 00:06:39.049
than the original breadbox,

00:06:39.049 --> 00:06:40.700
now more like a mini-fridge,

00:06:40.700 --> 00:06:43.461
but we still wouldn't have to build a pickup truck.

00:06:43.461 --> 00:06:46.085
So now we had our constraint.

00:06:46.085 --> 00:06:47.842
The laws of physics dictated

00:06:47.842 --> 00:06:51.166
the absolute minimum-sized
telescope that we could build.

NOTE Paragraph

00:06:51.166 --> 00:06:54.220
What came next was making the rest of the satellite

00:06:54.220 --> 00:06:55.961
as small and as simple as possible,

00:06:55.961 --> 00:06:58.613
basically a flying telescope with four walls

00:06:58.613 --> 00:07:01.554
and a set of electronics smaller than a phone book

00:07:01.554 --> 00:07:04.992
that used less power than a 100 watt lightbulb.

00:07:04.992 --> 00:07:06.892
The big challenge became actually taking

00:07:06.892 --> 00:07:09.090
the pictures through that telescope.

00:07:09.090 --> 00:07:12.134
Traditional imaging satellites use a line scanner,

00:07:12.134 --> 00:07:13.895
similar to a Xerox machine,

00:07:13.895 --> 00:07:16.295
and as they traverse the Earth, they take pictures,

00:07:16.295 --> 00:07:18.491
scanning row by row by row

00:07:18.491 --> 00:07:20.166
to build the complete image.

00:07:20.166 --> 00:07:23.170
Now people use these because they get a lot of light,

00:07:23.170 --> 00:07:25.001
which means less of the noise you see

00:07:25.016 --> 00:07:27.752
in a low-cost cell phone image.

00:07:27.752 --> 00:07:30.113
The problem with them is they require

00:07:30.113 --> 00:07:32.353
very sophisticated pointing.

00:07:32.353 --> 00:07:34.899
You have to stay focused on a 50-centimeter target

00:07:34.899 --> 00:07:36.651
from over 600 miles away

00:07:36.651 --> 00:07:38.883
while moving at more than
seven kilometers a second,

00:07:38.883 --> 00:07:41.632
which requires an awesome degree of complexity.

00:07:41.632 --> 00:07:45.069
So instead, we turned to a new
generation of video sensors,

00:07:45.069 --> 00:07:48.160
originally created for use in night vision goggles.

00:07:48.160 --> 00:07:51.059
Instead of taking a single, high quality image,

00:07:51.059 --> 00:07:52.446
we could take a videostream

00:07:52.446 --> 00:07:55.369
of individually noisier frames,

00:07:55.369 --> 00:07:57.113
but then we could recombine

00:07:57.113 --> 00:07:58.579
all of those frames together

00:07:58.579 --> 00:08:00.575
into very high-quality images

00:08:00.575 --> 00:08:03.168
using sophisticated pixel processing techniques

00:08:03.168 --> 00:08:04.689
here on the ground,

00:08:04.689 --> 00:08:07.826
at a cost of one one hundredth a traditional system.

00:08:07.826 --> 00:08:08.979
And we applied this technique

00:08:08.979 --> 00:08:11.953
to many of the other systems on the satellite as well,

00:08:11.953 --> 00:08:14.793
and day by day, our design evolved

00:08:14.793 --> 00:08:18.282
from CAD to prototypes

00:08:18.282 --> 00:08:21.170
to production units.

NOTE Paragraph

00:08:21.170 --> 00:08:22.827
A few short weeks ago,

00:08:22.827 --> 00:08:24.678
we packed up SkySat 1,

00:08:24.678 --> 00:08:26.451
put our signatures on it,

00:08:26.451 --> 00:08:28.674
and waved goodbye for the last time on Earth.

00:08:28.674 --> 00:08:32.155
Today, it's sitting in its final launch configuration

00:08:32.155 --> 00:08:35.216
ready to blast off in a few short weeks.

00:08:35.216 --> 00:08:37.558
And soon, we'll turn our attention to launching

00:08:37.558 --> 00:08:40.589
a constellation of 24 or more of these satellites

00:08:40.589 --> 00:08:43.139
and beginning to build the scalable analytics

00:08:43.139 --> 00:08:45.478
that will allow us to unearth the insights

00:08:45.478 --> 00:08:48.667
in the petabytes of data we will collect.

NOTE Paragraph

00:08:48.667 --> 00:08:52.640
So why do all of this? Why build these satellites?

00:08:52.640 --> 00:08:55.251
Well, it turns out imaging satellites

00:08:55.251 --> 00:08:58.961
have a unique ability to provide global transparency,

00:08:58.961 --> 00:09:01.587
and providing that transparency on a timely basis

00:09:01.587 --> 00:09:04.998
is simply an idea whose time has come.

00:09:04.998 --> 00:09:08.521
We see ourselves as pioneers of a new frontier,

00:09:08.521 --> 00:09:10.285
and beyond economic data,

00:09:10.285 --> 00:09:13.886
unlocking the human story, moment by moment.

00:09:13.886 --> 00:09:15.449
For a data scientist

00:09:15.449 --> 00:09:18.199
that just happened to go to space camp as a kid,

00:09:18.199 --> 00:09:20.914
it just doesn't get much better than that.

NOTE Paragraph

00:09:20.914 --> 00:09:23.129
Thank you.

NOTE Paragraph

00:09:23.129 --> 00:09:27.263
(Applause)