ABOUT THE SPEAKER
Raffaello D'Andrea - Autonomous systems pioneer
Raffaello D'Andrea explores the possibilities of autonomous technology by collaborating with artists, engineers and entrepreneurs.

Why you should listen

Raffaello D'Andrea combines academics, business, and the arts to explore the capabilities of autonomous systems. As part of his research as professor of dynamic systems and control at the Swiss Federal Institute of Technology (ETH Zürich), he and his collaborators enchant viewers with works like the self-destructing, self-assembling Robotic Chair, or the Balancing Cube that can perch itself on its corners.

D’Andrea and his team created the Flying Machine Arena to test the gravity-defying abilities of their athletic flying robots. Building on research in the Flying Machine Arena, ETH Zürich partnered with its spin-off company Verity Studios and with Cirque du Soleil to create “Sparked,” a short film showcasing the unexpected airborne dexterity of quadcopters. D’Andrea is the co-founder of Kiva Systems, a robotics company that develops intelligent automated warehouse systems and that was acquired by Amazon in 2012.

More profile about the speaker
Raffaello D'Andrea | Speaker | TED.com
TEDGlobal 2013

Raffaello D'Andrea: The astounding athletic power of quadcopters

Filmed:
12,409,728 views

In a robot lab at TEDGlobal, Raffaello D'Andrea demos his flying quadcopters: robots that think like athletes, solving physical problems with algorithms that help them learn. In a series of nifty demos, D'Andrea show drones that play catch, balance and make decisions together -- and watch out for an I-want-this-now demo of Kinect-controlled quads.
- Autonomous systems pioneer
Raffaello D'Andrea explores the possibilities of autonomous technology by collaborating with artists, engineers and entrepreneurs. Full bio

Double-click the English transcript below to play the video.

00:23
So what does it mean for a machine to be athletic?
0
11307
3184
00:26
We will demonstrate the concept of machine athleticism
1
14491
3417
00:29
and the research to achieve it
2
17908
1720
00:31
with the help of these flying machines called quadrocopters,
3
19628
2511
00:34
or quads, for short.
4
22139
1887
00:38
Quads have been around for a long time.
5
26034
2529
00:40
The reason that they're so popular these days
6
28563
1599
00:42
is because they're mechanically simple.
7
30162
1965
00:44
By controlling the speeds of these four propellers,
8
32127
2108
00:46
these machines can roll, pitch, yaw,
9
34235
3048
00:49
and accelerate along their common orientation.
10
37283
2825
00:52
On board are also a battery, a computer,
11
40108
3071
00:55
various sensors and wireless radios.
12
43179
3954
00:59
Quads are extremely agile, but this agility comes at a cost.
13
47133
4629
01:03
They are inherently unstable, and they need some form
14
51762
2992
01:06
of automatic feedback control in order to be able to fly.
15
54754
3972
01:16
So, how did it just do that?
16
64259
3238
01:19
Cameras on the ceiling and a laptop
17
67497
1818
01:21
serve as an indoor global positioning system.
18
69315
2576
01:23
It's used to locate objects in the space
19
71891
2185
01:26
that have these reflective markers on them.
20
74076
2527
01:28
This data is then sent to another laptop
21
76603
2061
01:30
that is running estimation and control algorithms,
22
78664
2211
01:32
which in turn sends commands to the quad,
23
80875
1886
01:34
which is also running estimation and control algorithms.
24
82761
3485
01:41
The bulk of our research is algorithms.
25
89620
2510
01:44
It's the magic that brings these machines to life.
26
92130
4161
01:48
So how does one design the algorithms
27
96291
2043
01:50
that create a machine athlete?
28
98334
2277
01:52
We use something broadly called model-based design.
29
100611
2880
01:55
We first capture the physics with a mathematical model
30
103491
3154
01:58
of how the machines behave.
31
106645
2134
02:00
We then use a branch of mathematics
32
108779
2297
02:03
called control theory to analyze these models
33
111076
2831
02:05
and also to synthesize algorithms for controlling them.
34
113907
3913
02:09
For example, that's how we can make the quad hover.
35
117820
3048
02:12
We first captured the dynamics
36
120868
1408
02:14
with a set of differential equations.
37
122276
1906
02:16
We then manipulate these equations with the help
38
124182
2349
02:18
of control theory to create algorithms that stabilize the quad.
39
126531
4640
02:23
Let me demonstrate the strength of this approach.
40
131171
2946
02:29
Suppose that we want this quad to not only hover
41
137198
2549
02:31
but to also balance this pole.
42
139747
2782
02:34
With a little bit of practice,
43
142529
1844
02:36
it's pretty straightforward for a human being to do this,
44
144373
2391
02:38
although we do have the advantage of having
45
146764
1755
02:40
two feet on the ground
46
148519
1328
02:41
and the use of our very versatile hands.
47
149847
2957
02:44
It becomes a little bit more difficult
48
152804
2469
02:47
when I only have one foot on the ground
49
155273
2296
02:49
and when I don't use my hands.
50
157569
2799
02:52
Notice how this pole has a reflective marker on top,
51
160368
2951
02:55
which means that it can be located in the space.
52
163319
3894
03:04
(Applause)
53
172971
5976
03:10
You can notice that this quad is making fine adjustments
54
178947
2826
03:13
to keep the pole balanced.
55
181773
2195
03:15
How did we design the algorithms to do this?
56
183968
3059
03:19
We added the mathematical model of the pole
57
187027
2105
03:21
to that of the quad.
58
189132
1519
03:22
Once we have a model of the combined quad-pole system,
59
190651
2954
03:25
we can use control theory to create algorithms for controlling it.
60
193605
4933
03:30
Here, you see that it's stable,
61
198538
1588
03:32
and even if I give it little nudges,
62
200126
2693
03:34
it goes back to the nice, balanced position.
63
202819
5403
03:40
We can also augment the model to include
64
208222
1663
03:41
where we want the quad to be in space.
65
209885
2441
03:44
Using this pointer, made out of reflective markers,
66
212326
3069
03:47
I can point to where I want the quad to be in space
67
215395
2224
03:49
a fixed distance away from me.
68
217619
3367
04:07
The key to these acrobatic maneuvers is algorithms,
69
235763
2902
04:10
designed with the help of mathematical models
70
238665
2426
04:13
and control theory.
71
241091
1950
04:15
Let's tell the quad to come back here
72
243041
2214
04:17
and let the pole drop,
73
245255
1819
04:19
and I will next demonstrate the importance
74
247074
1918
04:20
of understanding physical models
75
248992
2119
04:23
and the workings of the physical world.
76
251111
4120
04:37
Notice how the quad lost altitude
77
265462
1770
04:39
when I put this glass of water on it.
78
267232
2082
04:41
Unlike the balancing pole, I did not include
79
269314
2801
04:44
the mathematical model of the glass in the system.
80
272115
2584
04:46
In fact, the system doesn't even know that the glass of water is there.
81
274699
2919
04:49
Like before, I could use the pointer to tell the quad
82
277618
3292
04:52
where I want it to be in space.
83
280910
2589
04:55
(Applause)
84
283499
9597
05:05
Okay, you should be asking yourself,
85
293096
2110
05:07
why doesn't the water fall out of the glass?
86
295206
2509
05:09
Two facts: The first is that gravity acts
87
297715
3211
05:12
on all objects in the same way.
88
300926
2125
05:15
The second is that the propellers are all pointing
89
303051
2464
05:17
in the same direction of the glass, pointing up.
90
305515
3127
05:20
You put these two things together, the net result
91
308642
2377
05:23
is that all side forces on the glass are small
92
311019
2287
05:25
and are mainly dominated by aerodynamic effects,
93
313306
2785
05:28
which as these speeds are negligible.
94
316091
3841
05:35
And that's why you don't need to model the glass.
95
323336
2071
05:37
It naturally doesn't spill no matter what the quad does.
96
325407
3848
05:50
(Applause)
97
338651
7051
05:57
The lesson here is that some high-performance tasks
98
345702
4016
06:01
are easier than others,
99
349718
1397
06:03
and that understanding the physics of the problem
100
351115
2296
06:05
tells you which ones are easy and which ones are hard.
101
353411
2616
06:08
In this instance, carrying a glass of water is easy.
102
356027
2359
06:10
Balancing a pole is hard.
103
358386
3785
06:14
We've all heard stories of athletes
104
362171
1802
06:15
performing feats while physically injured.
105
363973
2493
06:18
Can a machine also perform
106
366466
1519
06:19
with extreme physical damage?
107
367985
2651
06:22
Conventional wisdom says that you need
108
370636
1726
06:24
at least four fixed motor propeller pairs in order to fly,
109
372362
3690
06:28
because there are four degrees of freedom to control:
110
376052
2112
06:30
roll, pitch, yaw and acceleration.
111
378164
3028
06:33
Hexacopters and octocopters, with six and eight propellers,
112
381192
3245
06:36
can provide redundancy,
113
384437
1483
06:37
but quadrocopters are much more popular
114
385920
1989
06:39
because they have the minimum number
115
387909
1815
06:41
of fixed motor propeller pairs: four.
116
389724
2476
06:44
Or do they?
117
392200
1965
07:01
If we analyze the mathematical model of this machine
118
409476
2600
07:04
with only two working propellers,
119
412076
2176
07:06
we discover that there's an unconventional way to fly it.
120
414252
6756
07:19
We relinquish control of yaw,
121
427980
1706
07:21
but roll, pitch and acceleration can still be controlled
122
429686
3070
07:24
with algorithms that exploit this new configuration.
123
432756
5452
07:33
Mathematical models tell us exactly when
124
441668
2432
07:36
and why this is possible.
125
444100
2072
07:38
In this instance, this knowledge allows us to design
126
446172
2368
07:40
novel machine architectures
127
448540
2256
07:42
or to design clever algorithms that gracefully handle damage,
128
450796
4176
07:46
just like human athletes do,
129
454972
1724
07:48
instead of building machines with redundancy.
130
456696
3823
07:52
We can't help but hold our breath when we watch
131
460519
2112
07:54
a diver somersaulting into the water,
132
462631
2437
07:57
or when a vaulter is twisting in the air,
133
465068
1790
07:58
the ground fast approaching.
134
466858
1718
08:00
Will the diver be able to pull off a rip entry?
135
468576
2568
08:03
Will the vaulter stick the landing?
136
471144
2000
08:05
Suppose we want this quad here
137
473144
1668
08:06
to perform a triple flip and finish off
138
474812
2451
08:09
at the exact same spot that it started.
139
477263
2269
08:11
This maneuver is going to happen so quickly
140
479532
2347
08:13
that we can't use position feedback to correct the motion during execution.
141
481879
3629
08:17
There simply isn't enough time.
142
485508
2274
08:19
Instead, what the quad can do is perform the maneuver blindly,
143
487782
3694
08:23
observe how it finishes the maneuver,
144
491476
2232
08:25
and then use that information to modify its behavior
145
493708
2424
08:28
so that the next flip is better.
146
496132
2289
08:30
Similar to the diver and the vaulter,
147
498421
1812
08:32
it is only through repeated practice
148
500233
1919
08:34
that the maneuver can be learned and executed
149
502152
1925
08:36
to the highest standard.
150
504077
2133
08:46
(Applause)
151
514412
4752
08:51
Striking a moving ball is a necessary skill in many sports.
152
519164
3472
08:54
How do we make a machine do
153
522636
1795
08:56
what an athlete does seemingly without effort?
154
524431
3753
09:15
(Applause)
155
543943
6599
09:22
This quad has a racket strapped onto its head
156
550542
2650
09:25
with a sweet spot roughly the size of an apple, so not too large.
157
553192
3487
09:28
The following calculations are made every 20 milliseconds,
158
556679
3167
09:31
or 50 times per second.
159
559846
1756
09:33
We first figure out where the ball is going.
160
561602
2812
09:36
We then next calculate how the quad should hit the ball
161
564414
2437
09:38
so that it flies to where it was thrown from.
162
566851
3099
09:41
Third, a trajectory is planned that carries the quad
163
569950
4456
09:46
from its current state to the impact point with the ball.
164
574406
3024
09:49
Fourth, we only execute 20 milliseconds' worth of that strategy.
165
577430
3809
09:53
Twenty milliseconds later, the whole process is repeated
166
581239
2807
09:56
until the quad strikes the ball.
167
584046
2243
10:07
(Applause)
168
595560
2646
10:10
Machines can not only perform dynamic maneuvers on their own,
169
598206
3372
10:13
they can do it collectively.
170
601578
1863
10:15
These three quads are cooperatively carrying a sky net.
171
603441
3587
10:28
(Applause)
172
616574
5411
10:33
They perform an extremely dynamic
173
621985
2197
10:36
and collective maneuver
174
624182
2168
10:38
to launch the ball back to me.
175
626350
1692
10:40
Notice that, at full extension, these quads are vertical.
176
628042
3511
10:48
(Applause)
177
636486
1772
10:50
In fact, when fully extended,
178
638258
2287
10:52
this is roughly five times greater than what a bungee jumper feels
179
640545
2685
10:55
at the end of their launch.
180
643230
4588
11:03
The algorithms to do this are very similar
181
651009
2677
11:05
to what the single quad used to hit the ball back to me.
182
653686
3384
11:09
Mathematical models are used to continuously re-plan
183
657070
2864
11:11
a cooperative strategy 50 times per second.
184
659934
4515
11:16
Everything we have seen so far has been
185
664449
1877
11:18
about the machines and their capabilities.
186
666326
2463
11:20
What happens when we couple this machine athleticism
187
668789
3015
11:23
with that of a human being?
188
671804
1776
11:25
What I have in front of me is a commercial gesture sensor
189
673580
3805
11:29
mainly used in gaming.
190
677385
1309
11:30
It can recognize what my various body parts
191
678694
1804
11:32
are doing in real time.
192
680498
2048
11:34
Similar to the pointer that I used earlier,
193
682546
2116
11:36
we can use this as inputs to the system.
194
684662
2792
11:39
We now have a natural way of interacting
195
687454
2396
11:41
with the raw athleticism of these quads with my gestures.
196
689850
4944
12:22
(Applause)
197
730453
4400
12:35
Interaction doesn't have to be virtual. It can be physical.
198
743982
4007
12:39
Take this quad, for example.
199
747989
1725
12:41
It's trying to stay at a fixed point in space.
200
749714
2628
12:44
If I try to move it out of the way, it fights me,
201
752342
3837
12:48
and moves back to where it wants to be.
202
756179
4155
12:52
We can change this behavior, however.
203
760334
3039
12:55
We can use mathematical models
204
763373
1699
12:57
to estimate the force that I'm applying to the quad.
205
765072
3133
13:00
Once we know this force, we can also change the laws of physics,
206
768205
3238
13:03
as far as the quad is concerned, of course.
207
771443
4499
13:07
Here the quad is behaving as if it were
208
775942
2263
13:10
in a viscous fluid.
209
778205
4365
13:14
We now have an intimate way
210
782570
2174
13:16
of interacting with a machine.
211
784744
1849
13:18
I will use this new capability to position
212
786593
2506
13:21
this camera-carrying quad to the appropriate location
213
789099
2577
13:23
for filming the remainder of this demonstration.
214
791676
3186
13:36
So we can physically interact with these quads
215
804222
2768
13:38
and we can change the laws of physics.
216
806990
2403
13:41
Let's have a little bit of fun with this.
217
809393
2313
13:43
For what you will see next, these quads
218
811706
1704
13:45
will initially behave as if they were on Pluto.
219
813410
3496
13:48
As time goes on, gravity will be increased
220
816906
2312
13:51
until we're all back on planet Earth,
221
819218
2120
13:53
but I assure you we won't get there.
222
821338
1719
13:55
Okay, here goes.
223
823057
3501
14:05
(Laughter)
224
833667
3412
14:34
(Laughter)
225
862778
3208
14:37
(Applause)
226
865986
3386
14:41
Whew!
227
869372
1207
14:46
You're all thinking now,
228
874779
1517
14:48
these guys are having way too much fun,
229
876296
1975
14:50
and you're probably also asking yourself,
230
878271
2079
14:52
why exactly are they building machine athletes?
231
880350
3936
14:56
Some conjecture that the role of play in the animal kingdom
232
884286
2652
14:58
is to hone skills and develop capabilities.
233
886938
2752
15:01
Others think that it has more of a social role,
234
889690
2066
15:03
that it's used to bind the group.
235
891756
1674
15:05
Similarly, we use the analogy of sports and athleticism
236
893430
3533
15:08
to create new algorithms for machines
237
896963
2079
15:11
to push them to their limits.
238
899042
2147
15:13
What impact will the speed of machines have on our way of life?
239
901189
3677
15:16
Like all our past creations and innovations,
240
904866
2389
15:19
they may be used to improve the human condition
241
907255
2827
15:22
or they may be misused and abused.
242
910082
2501
15:24
This is not a technical choice we are faced with;
243
912583
1923
15:26
it's a social one.
244
914506
1755
15:28
Let's make the right choice,
245
916261
1390
15:29
the choice that brings out the best in the future of machines,
246
917651
2405
15:32
just like athleticism in sports
247
920056
1812
15:33
can bring out the best in us.
248
921868
2161
15:36
Let me introduce you to the wizards behind the green curtain.
249
924029
3450
15:39
They're the current members of the Flying Machine Arena research team.
250
927479
2900
15:42
(Applause)
251
930379
4769
15:47
Federico Augugliaro, Dario Brescianini, Markus Hehn,
252
935148
3046
15:50
Sergei Lupashin, Mark Muller and Robin Ritz.
253
938194
2824
15:53
Look out for them. They're destined for great things.
254
941018
1932
15:54
Thank you.
255
942950
1178
15:56
(Applause)
256
944128
6354
Translated by Joseph Geni
Reviewed by Morton Bast

▲Back to top

ABOUT THE SPEAKER
Raffaello D'Andrea - Autonomous systems pioneer
Raffaello D'Andrea explores the possibilities of autonomous technology by collaborating with artists, engineers and entrepreneurs.

Why you should listen

Raffaello D'Andrea combines academics, business, and the arts to explore the capabilities of autonomous systems. As part of his research as professor of dynamic systems and control at the Swiss Federal Institute of Technology (ETH Zürich), he and his collaborators enchant viewers with works like the self-destructing, self-assembling Robotic Chair, or the Balancing Cube that can perch itself on its corners.

D’Andrea and his team created the Flying Machine Arena to test the gravity-defying abilities of their athletic flying robots. Building on research in the Flying Machine Arena, ETH Zürich partnered with its spin-off company Verity Studios and with Cirque du Soleil to create “Sparked,” a short film showcasing the unexpected airborne dexterity of quadcopters. D’Andrea is the co-founder of Kiva Systems, a robotics company that develops intelligent automated warehouse systems and that was acquired by Amazon in 2012.

More profile about the speaker
Raffaello D'Andrea | Speaker | TED.com