1
00:00:11,544 --> 00:00:14,067
So what does it mean
for a machine to be athletic?

2
00:00:15,059 --> 00:00:17,884
We will demonstrate the concept
of machine athleticism

3
00:00:17,908 --> 00:00:19,604
and the research to achieve it

4
00:00:19,628 --> 00:00:22,486
with the help of these flying machines
called quadrocopters,

5
00:00:22,510 --> 00:00:23,557
or quads, for short.

6
00:00:26,291 --> 00:00:28,184
Quads have been around for a long time.

7
00:00:28,501 --> 00:00:31,778
They're so popular these days
because they're mechanically simple.

8
00:00:32,255 --> 00:00:34,644
By controlling
the speeds of these four propellers,

9
00:00:34,668 --> 00:00:37,413
these machines can roll, pitch, yaw,

10
00:00:37,437 --> 00:00:39,608
and accelerate
along their common orientation.

11
00:00:40,179 --> 00:00:43,377
On board are also a battery, a computer,

12
00:00:43,401 --> 00:00:45,354
various sensors and wireless radios.

13
00:00:47,298 --> 00:00:50,607
Quads are extremely agile,
but this agility comes at a cost.

14
00:00:51,762 --> 00:00:53,850
They are inherently unstable,

15
00:00:53,874 --> 00:00:56,445
and they need some form
of automatic feedback control

16
00:00:56,469 --> 00:00:57,842
in order to be able to fly.

17
00:01:04,259 --> 00:01:06,512
So, how did it just do that?

18
00:01:07,497 --> 00:01:09,291
Cameras on the ceiling and a laptop

19
00:01:09,315 --> 00:01:11,745
serve as an indoor
global positioning system.

20
00:01:12,301 --> 00:01:14,340
It's used to locate objects in the space

21
00:01:14,364 --> 00:01:16,452
that have these reflective
markers on them.

22
00:01:16,476 --> 00:01:18,514
This data is then sent to another laptop

23
00:01:18,538 --> 00:01:20,896
that is running estimation
and control algorithms,

24
00:01:20,920 --> 00:01:22,874
which in turn sends commands to the quad,

25
00:01:22,898 --> 00:01:25,531
which is also running estimation
and control algorithms.

26
00:01:29,864 --> 00:01:32,106
The bulk of our research is algorithms.

27
00:01:32,130 --> 00:01:34,706
It's the magic that brings
these machines to life.

28
00:01:36,291 --> 00:01:40,324
So how does one design the algorithms
that create a machine athlete?

29
00:01:40,792 --> 00:01:43,379
We use something broadly
called model-based design.

30
00:01:43,839 --> 00:01:45,205
We first capture the physics

31
00:01:45,229 --> 00:01:48,212
with a mathematical model
of how the machines behave.

32
00:01:48,636 --> 00:01:52,586
We then use a branch of mathematics
called control theory

33
00:01:52,610 --> 00:01:54,267
to analyze these models

34
00:01:54,291 --> 00:01:57,061
and also to synthesize
algorithms for controlling them.

35
00:01:57,820 --> 00:02:00,844
For example, that's how we can
make the quad hover.

36
00:02:00,868 --> 00:02:04,158
We first captured the dynamics
with a set of differential equations.

37
00:02:04,182 --> 00:02:07,634
We then manipulate these equations
with the help of control theory

38
00:02:07,658 --> 00:02:10,595
to create algorithms
that stabilize the quad.

39
00:02:11,130 --> 00:02:13,423
Let me demonstrate
the strength of this approach.

40
00:02:17,198 --> 00:02:19,831
Suppose that we want
this quad to not only hover

41
00:02:19,855 --> 00:02:21,514
but to also balance this pole.

42
00:02:22,520 --> 00:02:24,294
With a little bit of practice,

43
00:02:24,318 --> 00:02:27,001
it's pretty straightforward
for a human being to do this,

44
00:02:27,025 --> 00:02:30,216
although we do have the advantage
of having two feet on the ground

45
00:02:30,240 --> 00:02:32,299
and the use of our very versatile hands.

46
00:02:32,804 --> 00:02:35,542
It becomes a little bit more difficult

47
00:02:35,566 --> 00:02:37,545
when I only have one foot on the ground

48
00:02:37,569 --> 00:02:39,171
and when I don't use my hands.

49
00:02:40,733 --> 00:02:43,295
Notice how this pole has
a reflective marker on top,

50
00:02:43,319 --> 00:02:45,980
which means that it can
be located in the space.

51
00:02:52,276 --> 00:02:53,343
(Audience) Oh!

52
00:02:53,367 --> 00:02:54,626
(Applause)

53
00:02:57,695 --> 00:02:59,363
(Applause ends)

54
00:02:59,387 --> 00:03:02,028
You can notice that this quad
is making fine adjustments

55
00:03:02,052 --> 00:03:03,426
to keep the pole balanced.

56
00:03:04,417 --> 00:03:06,489
How did we design
the algorithms to do this?

57
00:03:07,027 --> 00:03:09,108
We added the mathematical
model of the pole

58
00:03:09,132 --> 00:03:10,579
to that of the quad.

59
00:03:10,603 --> 00:03:13,581
Once we have a model
of the combined quad-pole system,

60
00:03:13,605 --> 00:03:16,924
we can use control theory to create
algorithms for controlling it.

61
00:03:18,538 --> 00:03:20,102
Here, you see that it's stable,

62
00:03:20,126 --> 00:03:22,795
and even if I give it little nudges,

63
00:03:22,819 --> 00:03:23,985
it goes back --

64
00:03:25,589 --> 00:03:27,358
to the nice, balanced position.

65
00:03:28,087 --> 00:03:29,477
We can also augment the model

66
00:03:29,501 --> 00:03:31,872
to include where we want
the quad to be in space.

67
00:03:32,428 --> 00:03:35,371
Using this pointer,
made out of reflective markers,

68
00:03:35,395 --> 00:03:37,824
I can point to where I want
the quad to be in space

69
00:03:37,848 --> 00:03:39,445
a fixed distance away from me.

70
00:03:47,785 --> 00:03:49,665
(Laughter)

71
00:03:55,763 --> 00:03:58,889
The key to these acrobatic
maneuvers is algorithms,

72
00:03:58,913 --> 00:04:01,206
designed with the help
of mathematical models

73
00:04:01,230 --> 00:04:02,415
and control theory.

74
00:04:03,246 --> 00:04:05,231
Let's tell the quad to come back here

75
00:04:05,255 --> 00:04:06,919
and let the pole drop,

76
00:04:06,943 --> 00:04:09,074
and I will next demonstrate the importance

77
00:04:09,098 --> 00:04:11,087
of understanding physical models

78
00:04:11,111 --> 00:04:13,156
and the workings of the physical world.

79
00:04:25,462 --> 00:04:29,290
Notice how the quad lost altitude
when I put this glass of water on it.

80
00:04:29,314 --> 00:04:30,717
Unlike the balancing pole,

81
00:04:30,741 --> 00:04:33,756
I did not include the mathematical
model of the glass

82
00:04:33,780 --> 00:04:35,087
in the system.

83
00:04:35,111 --> 00:04:38,124
In fact, the system doesn't even know
that the glass is there.

84
00:04:38,148 --> 00:04:40,886
Like before, I could use
the pointer to tell the quad

85
00:04:40,910 --> 00:04:42,481
where I want it to be in space.

86
00:04:45,684 --> 00:04:47,483
(Applause)

87
00:04:51,531 --> 00:04:53,071
(Applause ends)

88
00:04:53,515 --> 00:04:55,254
Okay, you should be asking yourself,

89
00:04:55,278 --> 00:04:57,340
why doesn't the water
fall out of the glass?

90
00:04:57,888 --> 00:04:58,918
Two facts.

91
00:04:58,942 --> 00:05:02,575
The first is that gravity acts
on all objects in the same way.

92
00:05:03,051 --> 00:05:04,830
The second is that the propellers

93
00:05:04,854 --> 00:05:07,927
are all pointing in the same direction
of the glass, pointing up.

94
00:05:08,642 --> 00:05:10,289
You put these two things together,

95
00:05:10,313 --> 00:05:13,868
the net result is that all side forces
on the glass are small

96
00:05:13,892 --> 00:05:16,169
and are mainly dominated
by aerodynamic effects,

97
00:05:16,193 --> 00:05:18,154
which at these speeds are negligible.

98
00:05:23,336 --> 00:05:25,670
And that's why you don't need
to model the glass.

99
00:05:25,694 --> 00:05:28,435
It naturally doesn't spill,
no matter what the quad does.

100
00:05:32,500 --> 00:05:35,063
(Audience) Oh!

101
00:05:38,340 --> 00:05:40,198
(Applause)

102
00:05:43,159 --> 00:05:44,414
(Applause ends)

103
00:05:45,940 --> 00:05:47,054
The lesson here

104
00:05:47,078 --> 00:05:51,091
is that some high-performance tasks
are easier than others,

105
00:05:51,115 --> 00:05:53,452
and that understanding
the physics of the problem

106
00:05:53,476 --> 00:05:56,170
tells you which ones are easy
and which ones are hard.

107
00:05:56,194 --> 00:05:58,704
In this instance, carrying
a glass of water is easy.

108
00:05:58,728 --> 00:06:00,358
Balancing a pole is hard.

109
00:06:02,171 --> 00:06:06,442
We've all heard stories of athletes
performing feats while physically injured.

110
00:06:06,466 --> 00:06:09,651
Can a machine also perform
with extreme physical damage?

111
00:06:10,636 --> 00:06:12,121
Conventional wisdom says

112
00:06:12,145 --> 00:06:16,028
that you need at least four fixed motor
propeller pairs in order to fly,

113
00:06:16,052 --> 00:06:18,576
because there are four degrees
of freedom to control:

114
00:06:18,600 --> 00:06:20,565
roll, pitch, yaw and acceleration.

115
00:06:21,192 --> 00:06:24,413
Hexacopters and octocopters,
with six and eight propellers,

116
00:06:24,437 --> 00:06:25,896
can provide redundancy,

117
00:06:25,920 --> 00:06:27,885
but quadrocopters are much more popular

118
00:06:27,909 --> 00:06:31,595
because they have the minimum number
of fixed motor propeller pairs: four.

119
00:06:32,168 --> 00:06:33,184
Or do they?

120
00:06:40,827 --> 00:06:41,827
(Audience) Oh!

121
00:06:42,407 --> 00:06:43,407
(Laughter)

122
00:06:49,334 --> 00:06:52,052
If we analyze the mathematical
model of this machine

123
00:06:52,076 --> 00:06:54,228
with only two working propellers,

124
00:06:54,252 --> 00:06:56,992
we discover that there's
an unconventional way to fly it.

125
00:07:07,980 --> 00:07:09,662
We relinquish control of yaw,

126
00:07:09,686 --> 00:07:12,732
but roll, pitch and acceleration
can still be controlled

127
00:07:12,756 --> 00:07:16,102
with algorithms that exploit
this new configuration.

128
00:07:21,668 --> 00:07:25,957
Mathematical models tell us
exactly when and why this is possible.

129
00:07:25,981 --> 00:07:28,759
In this instance, this knowledge
allows us to design

130
00:07:28,783 --> 00:07:31,256
novel machine architectures

131
00:07:31,280 --> 00:07:34,948
or to design clever algorithms
that gracefully handle damage,

132
00:07:34,972 --> 00:07:36,672
just like human athletes do,

133
00:07:36,696 --> 00:07:38,960
instead of building
machines with redundancy.

134
00:07:40,688 --> 00:07:42,345
We can't help but hold our breath

135
00:07:42,369 --> 00:07:45,044
when we watch a diver
somersaulting into the water,

136
00:07:45,068 --> 00:07:47,021
or when a vaulter is twisting in the air,

137
00:07:47,045 --> 00:07:48,552
the ground fast approaching.

138
00:07:48,576 --> 00:07:51,120
Will the diver be able
to pull off a rip entry?

139
00:07:51,144 --> 00:07:52,951
Will the vaulter stick the landing?

140
00:07:53,292 --> 00:07:56,472
Suppose we want this quad here
to perform a triple flip

141
00:07:56,496 --> 00:07:59,322
and finish off at the exact same
spot that it started.

142
00:07:59,736 --> 00:08:01,855
This maneuver is going
to happen so quickly

143
00:08:01,879 --> 00:08:05,724
that we can't use position feedback
to correct the motion during execution.

144
00:08:05,748 --> 00:08:07,345
There simply isn't enough time.

145
00:08:07,782 --> 00:08:11,452
Instead, what the quad can do
is perform the maneuver blindly,

146
00:08:11,476 --> 00:08:13,684
observe how it finishes the maneuver,

147
00:08:13,708 --> 00:08:16,185
and then use that information
to modify its behavior

148
00:08:16,209 --> 00:08:17,847
so that the next flip is better.

149
00:08:18,421 --> 00:08:20,209
Similar to the diver and the vaulter,

150
00:08:20,233 --> 00:08:22,128
it is only through repeated practice

151
00:08:22,152 --> 00:08:24,295
that the maneuver can
be learned and executed

152
00:08:24,319 --> 00:08:25,561
to the highest standard.

153
00:08:33,100 --> 00:08:34,369
(Laughter)

154
00:08:34,393 --> 00:08:38,178
(Applause)

155
00:08:39,164 --> 00:08:42,612
Striking a moving ball
is a necessary skill in many sports.

156
00:08:43,127 --> 00:08:44,619
How do we make a machine do

157
00:08:44,643 --> 00:08:47,231
what an athlete does
seemingly without effort?

158
00:08:57,547 --> 00:08:59,054
(Laughter)

159
00:09:04,143 --> 00:09:06,000
(Applause)

160
00:09:08,500 --> 00:09:09,798
(Applause ends)

161
00:09:10,542 --> 00:09:13,168
This quad has a racket
strapped onto its head

162
00:09:13,192 --> 00:09:16,655
with a sweet spot roughly the size
of an apple, so not too large.

163
00:09:17,017 --> 00:09:19,822
The following calculations
are made every 20 milliseconds,

164
00:09:19,846 --> 00:09:21,163
or 50 times per second.

165
00:09:21,602 --> 00:09:23,877
We first figure out where
the ball is going.

166
00:09:24,414 --> 00:09:27,035
We then next calculate
how the quad should hit the ball

167
00:09:27,059 --> 00:09:29,544
so that it flies
to where it was thrown from.

168
00:09:29,950 --> 00:09:34,382
Third, a trajectory is planned
that carries the quad

169
00:09:34,406 --> 00:09:37,406
from its current state
to the impact point with the ball.

170
00:09:37,430 --> 00:09:41,009
Fourth, we only execute 20 milliseconds'
worth of that strategy.

171
00:09:41,366 --> 00:09:44,159
Twenty milliseconds later,
the whole process is repeated

172
00:09:44,183 --> 00:09:45,950
until the quad strikes the ball.

173
00:09:55,290 --> 00:09:57,694
(Applause)

174
00:09:58,131 --> 00:10:01,554
Machines can not only perform
dynamic maneuvers on their own,

175
00:10:01,578 --> 00:10:03,417
they can do it collectively.

176
00:10:03,441 --> 00:10:07,028
These three quads are cooperatively
carrying a sky net.

177
00:10:15,589 --> 00:10:17,268
(Applause)

178
00:10:20,372 --> 00:10:21,961
(Applause ends)

179
00:10:21,985 --> 00:10:26,326
They perform an extremely dynamic
and collective maneuver

180
00:10:26,350 --> 00:10:28,018
to launch the ball back to me.

181
00:10:28,042 --> 00:10:31,188
Notice that, at full extension,
these quads are vertical.

182
00:10:36,486 --> 00:10:38,234
(Applause)

183
00:10:38,258 --> 00:10:40,401
In fact, when fully extended,

184
00:10:40,425 --> 00:10:43,688
this is roughly five times greater
than what a bungee jumper feels

185
00:10:43,712 --> 00:10:45,682
at the end of their launch.

186
00:10:51,203 --> 00:10:53,662
The algorithms to do this are very similar

187
00:10:53,686 --> 00:10:56,601
to what the single quad used
to hit the ball back to me.

188
00:10:57,070 --> 00:10:59,910
Mathematical models are used
to continuously re-plan

189
00:10:59,934 --> 00:11:02,765
a cooperative strategy
50 times per second.

190
00:11:04,506 --> 00:11:08,765
Everything we have seen so far has been
about the machines and their capabilities.

191
00:11:08,789 --> 00:11:11,781
What happens when we couple
this machine athleticism

192
00:11:11,805 --> 00:11:13,288
with that of a human being?

193
00:11:13,765 --> 00:11:17,361
What I have in front of me
is a commercial gesture sensor

194
00:11:17,385 --> 00:11:18,670
mainly used in gaming.

195
00:11:18,694 --> 00:11:20,742
It can recognize
what my various body parts

196
00:11:20,766 --> 00:11:22,083
are doing in real time.

197
00:11:22,710 --> 00:11:24,789
Similar to the pointer
that I used earlier,

198
00:11:24,813 --> 00:11:27,035
we can use this as inputs to the system.

199
00:11:27,454 --> 00:11:29,826
We now have a natural way of interacting

200
00:11:29,850 --> 00:11:33,111
with the raw athleticism
of these quads with my gestures.

201
00:12:10,715 --> 00:12:14,572
(Applause)

202
00:12:24,354 --> 00:12:26,370
Interaction doesn't have to be virtual.

203
00:12:26,394 --> 00:12:27,584
It can be physical.

204
00:12:27,989 --> 00:12:29,504
Take this quad, for example.

205
00:12:29,893 --> 00:12:32,116
It's trying to stay
at a fixed point in space.

206
00:12:32,933 --> 00:12:36,622
If I try to move it
out of the way, it fights me,

207
00:12:36,646 --> 00:12:38,520
and moves back to where it wants to be.

208
00:12:40,254 --> 00:12:42,341
We can change this behavior, however.

209
00:12:43,373 --> 00:12:45,048
We can use mathematical models

210
00:12:45,072 --> 00:12:47,736
to estimate the force
that I'm applying to the quad.

211
00:12:48,205 --> 00:12:51,419
Once we know this force,
we can also change the laws of physics,

212
00:12:51,443 --> 00:12:53,530
as far as the quad
is concerned, of course.

213
00:12:55,942 --> 00:12:59,251
Here, the quad is behaving
as if it were in a viscous fluid.

214
00:13:02,570 --> 00:13:06,289
We now have an intimate way
of interacting with a machine.

215
00:13:06,593 --> 00:13:09,075
I will use this new capability to position

216
00:13:09,099 --> 00:13:11,652
this camera-carrying quad
to the appropriate location

217
00:13:11,676 --> 00:13:14,104
for filming the remainder
of this demonstration.

218
00:13:24,583 --> 00:13:26,966
So we can physically interact
with these quads

219
00:13:26,990 --> 00:13:28,950
and we can change the laws of physics.

220
00:13:29,393 --> 00:13:31,370
Let's have a little bit of fun with this.

221
00:13:31,539 --> 00:13:32,908
For what you will see next,

222
00:13:32,932 --> 00:13:36,226
these quads will initially behave
as if they were on Pluto.

223
00:13:36,906 --> 00:13:39,194
As time goes on, gravity will be increased

224
00:13:39,218 --> 00:13:41,243
until we're all back on planet Earth,

225
00:13:41,267 --> 00:13:43,033
but I assure you we won't get there.

226
00:13:43,462 --> 00:13:44,486
Okay, here goes.

227
00:13:52,897 --> 00:13:54,448
(Laughter)

228
00:14:22,778 --> 00:14:25,738
(Laughter)

229
00:14:25,858 --> 00:14:29,619
(Applause)

230
00:14:29,760 --> 00:14:30,826
Whew!

231
00:14:35,045 --> 00:14:36,272
You're all thinking now,

232
00:14:36,296 --> 00:14:38,247
these guys are having way too much fun,

233
00:14:38,271 --> 00:14:40,326
and you're probably also asking yourself,

234
00:14:40,350 --> 00:14:43,151
why exactly are they building
machine athletes?

235
00:14:44,286 --> 00:14:47,096
Some conjecture that the role
of play in the animal kingdom

236
00:14:47,120 --> 00:14:49,257
is to hone skills
and develop capabilities.

237
00:14:49,690 --> 00:14:51,905
Others think that it has
more of a social role,

238
00:14:51,929 --> 00:14:53,579
that it's used to bind the group.

239
00:14:53,603 --> 00:14:56,939
Similarly, we use the analogy
of sports and athleticism

240
00:14:56,963 --> 00:14:59,018
to create new algorithms for machines

241
00:14:59,042 --> 00:15:00,485
to push them to their limits.

242
00:15:01,350 --> 00:15:04,398
What impact will the speed
of machines have on our way of life?

243
00:15:04,866 --> 00:15:07,231
Like all our past creations
and innovations,

244
00:15:07,255 --> 00:15:10,058
they may be used to improve
the human condition

245
00:15:10,082 --> 00:15:12,056
or they may be misused and abused.

246
00:15:12,595 --> 00:15:14,897
This is not a technical choice
we are faced with;

247
00:15:14,921 --> 00:15:15,936
it's a social one.

248
00:15:16,332 --> 00:15:17,683
Let's make the right choice,

249
00:15:17,707 --> 00:15:20,619
the choice that brings out the best
in the future of machines,

250
00:15:20,643 --> 00:15:23,508
just like athleticism in sports
can bring out the best in us.

251
00:15:24,380 --> 00:15:27,455
Let me introduce you to the wizards
behind the green curtain.

252
00:15:27,479 --> 00:15:30,813
They're the current members
of the Flying Machine Arena research team.

253
00:15:30,837 --> 00:15:35,054
(Applause)

254
00:15:35,267 --> 00:15:37,545
Federico Augugliaro, Dario Brescianini,

255
00:15:37,569 --> 00:15:40,994
Markus Hehn, Sergei Lupashin,
Mark Muller and Robin Ritz.

256
00:15:41,018 --> 00:15:43,542
Look out for them.
They're destined for great things.

257
00:15:43,566 --> 00:15:44,620
Thank you.

258
00:15:44,644 --> 00:15:48,100
(Applause)