A robot that runs and swims like a salamander | Auke Ijspeert

814,622 views ・ 2016-02-18

TED


Please double-click on the English subtitles below to play the video.

00:12
This is Pleurobot.
0
12760
1840
00:15
Pleurobot is a robot that we designed to closely mimic a salamander species
1
15400
3616
00:19
called Pleurodeles waltl.
2
19040
1400
00:21
Pleurobot can walk, as you can see here,
3
21240
2256
00:23
and as you'll see later, it can also swim.
4
23520
2040
00:26
So you might ask, why did we design this robot?
5
26280
2191
00:28
And in fact, this robot has been designed as a scientific tool for neuroscience.
6
28960
3762
00:33
Indeed, we designed it together with neurobiologists
7
33400
2496
00:35
to understand how animals move,
8
35920
1896
00:37
and especially how the spinal cord controls locomotion.
9
37840
2760
00:41
But the more I work in biorobotics,
10
41560
1696
00:43
the more I'm really impressed by animal locomotion.
11
43280
2381
00:45
If you think of a dolphin swimming or a cat running or jumping around,
12
45920
4296
00:50
or even us as humans,
13
50240
1576
00:51
when you go jogging or play tennis,
14
51840
1816
00:53
we do amazing things.
15
53680
1240
00:55
And in fact, our nervous system solves a very, very complex control problem.
16
55880
4136
01:00
It has to coordinate more or less 200 muscles perfectly,
17
60040
3096
01:03
because if the coordination is bad, we fall over or we do bad locomotion.
18
63160
3680
01:07
And my goal is to understand how this works.
19
67560
2720
01:11
There are four main components behind animal locomotion.
20
71160
2840
01:14
The first component is just the body,
21
74800
1936
01:16
and in fact we should never underestimate
22
76760
1976
01:18
to what extent the biomechanics already simplify locomotion in animals.
23
78760
3480
01:22
Then you have the spinal cord,
24
82920
1456
01:24
and in the spinal cord you find reflexes,
25
84400
1976
01:26
multiple reflexes that create a sensorimotor coordination loop
26
86400
3456
01:29
between neural activity in the spinal cord and mechanical activity.
27
89880
3480
01:34
A third component are central pattern generators.
28
94000
2976
01:37
These are very interesting circuits in the spinal cord of vertebrate animals
29
97000
3896
01:40
that can generate, by themselves,
30
100920
1616
01:42
very coordinated rhythmic patterns of activity
31
102560
2736
01:45
while receiving only very simple input signals.
32
105320
2376
01:47
And these input signals
33
107720
1216
01:48
coming from descending modulation from higher parts of the brain,
34
108960
3056
01:52
like the motor cortex, the cerebellum, the basal ganglia,
35
112040
2696
01:54
will all modulate activity of the spinal cord
36
114760
2136
01:56
while we do locomotion.
37
116920
1456
01:58
But what's interesting is to what extent just a low-level component,
38
118400
3216
02:01
the spinal cord, together with the body,
39
121640
1936
02:03
already solve a big part of the locomotion problem.
40
123600
2456
02:06
You probably know it by the fact that you can cut the head off a chicken,
41
126080
3422
02:09
it can still run for a while,
42
129532
1381
02:10
showing that just the lower part, spinal cord and body,
43
130937
2539
02:13
already solve a big part of locomotion.
44
133500
1873
02:15
Now, understanding how this works is very complex,
45
135397
2459
02:17
because first of all,
46
137880
1296
02:19
recording activity in the spinal cord is very difficult.
47
139200
2620
02:21
It's much easier to implant electrodes in the motor cortex
48
141844
2772
02:24
than in the spinal cord, because it's protected by the vertebrae.
49
144640
3056
02:27
Especially in humans, very hard to do.
50
147720
1816
02:29
A second difficulty is that locomotion is really due to a very complex
51
149560
3776
02:33
and very dynamic interaction between these four components.
52
153360
3056
02:36
So it's very hard to find out what's the role of each over time.
53
156440
3240
02:40
This is where biorobots like Pleurobot and mathematical models
54
160880
3736
02:44
can really help.
55
164640
1200
02:47
So what's biorobotics?
56
167480
1256
02:48
Biorobotics is a very active field of research in robotics
57
168760
2736
02:51
where people want to take inspiration from animals
58
171520
2456
02:54
to make robots to go outdoors,
59
174000
2456
02:56
like service robots or search and rescue robots
60
176480
2656
02:59
or field robots.
61
179160
1200
03:00
And the big goal here is to take inspiration from animals
62
180880
2696
03:03
to make robots that can handle complex terrain --
63
183600
2336
03:05
stairs, mountains, forests,
64
185960
1616
03:07
places where robots still have difficulties
65
187600
2016
03:09
and where animals can do a much better job.
66
189640
2056
03:11
The robot can be a wonderful scientific tool as well.
67
191720
2496
03:14
There are some very nice projects where robots are used,
68
194240
2620
03:16
like a scientific tool for neuroscience, for biomechanics or for hydrodynamics.
69
196884
3972
03:20
And this is exactly the purpose of Pleurobot.
70
200880
2120
03:23
So what we do in my lab is to collaborate with neurobiologists
71
203600
2936
03:26
like Jean-Marie Cabelguen, a neurobiologist in Bordeaux in France,
72
206560
3216
03:29
and we want to make spinal cord models and validate them on robots.
73
209800
4040
03:34
And here we want to start simple.
74
214480
1616
03:36
So it's good to start with simple animals
75
216120
1976
03:38
like lampreys, which are very primitive fish,
76
218120
2256
03:40
and then gradually go toward more complex locomotion,
77
220400
2496
03:42
like in salamanders,
78
222920
1256
03:44
but also in cats and in humans,
79
224200
1496
03:45
in mammals.
80
225720
1200
03:47
And here, a robot becomes an interesting tool
81
227880
2376
03:50
to validate our models.
82
230280
1936
03:52
And in fact, for me, Pleurobot is a kind of dream becoming true.
83
232240
3016
03:55
Like, more or less 20 years ago I was already working on a computer
84
235280
3256
03:58
making simulations of lamprey and salamander locomotion
85
238560
2656
04:01
during my PhD.
86
241240
1536
04:02
But I always knew that my simulations were just approximations.
87
242800
3376
04:06
Like, simulating the physics in water or with mud or with complex ground,
88
246200
3976
04:10
it's very hard to simulate that properly on a computer.
89
250200
2656
04:12
Why not have a real robot and real physics?
90
252880
2040
04:15
So among all these animals, one of my favorites is the salamander.
91
255600
3136
04:18
You might ask why, and it's because as an amphibian,
92
258760
3456
04:22
it's a really key animal from an evolutionary point of view.
93
262240
2856
04:25
It makes a wonderful link between swimming,
94
265120
2056
04:27
as you find it in eels or fish,
95
267200
1896
04:29
and quadruped locomotion, as you see in mammals, in cats and humans.
96
269120
4120
04:34
And in fact, the modern salamander
97
274160
1656
04:35
is very close to the first terrestrial vertebrate,
98
275840
2376
04:38
so it's almost a living fossil,
99
278240
1536
04:39
which gives us access to our ancestor,
100
279800
1936
04:41
the ancestor to all terrestrial tetrapods.
101
281760
2920
04:45
So the salamander swims
102
285240
1376
04:46
by doing what's called an anguilliform swimming gait,
103
286640
2496
04:49
so they propagate a nice traveling wave of muscle activity from head to tail.
104
289160
3640
04:53
And if you place the salamander on the ground,
105
293440
2176
04:55
it switches to what's called a walking trot gait.
106
295640
2336
04:58
In this case, you have nice periodic activation of the limbs
107
298000
2863
05:00
which are very nicely coordinated
108
300887
1609
05:02
with this standing wave undulation of the body,
109
302520
2656
05:05
and that's exactly the gait that you are seeing here on Pleurobot.
110
305200
3656
05:08
Now, one thing which is very surprising and fascinating in fact
111
308880
2976
05:11
is the fact that all this can be generated just by the spinal cord and the body.
112
311880
4136
05:16
So if you take a decerebrated salamander --
113
316040
2000
05:18
it's not so nice but you remove the head --
114
318064
2016
05:20
and if you electrically stimulate the spinal cord,
115
320104
2672
05:22
at low level of stimulation this will induce a walking-like gait.
116
322800
3256
05:26
If you stimulate a bit more, the gait accelerates.
117
326080
2456
05:28
And at some point, there's a threshold,
118
328560
1896
05:30
and automatically, the animal switches to swimming.
119
330480
2536
05:33
This is amazing.
120
333040
1376
05:34
Just changing the global drive,
121
334440
1496
05:35
as if you are pressing the gas pedal
122
335960
1736
05:37
of descending modulation to your spinal cord,
123
337720
2136
05:39
makes a complete switch between two very different gaits.
124
339880
3000
05:44
And in fact, the same has been observed in cats.
125
344440
2576
05:47
If you stimulate the spinal cord of a cat,
126
347040
2016
05:49
you can switch between walk, trot and gallop.
127
349080
2216
05:51
Or in birds, you can make a bird switch between walking,
128
351320
2736
05:54
at a low level of stimulation,
129
354080
1456
05:55
and flapping its wings at high-level stimulation.
130
355560
2816
05:58
And this really shows that the spinal cord
131
358400
2016
06:00
is a very sophisticated locomotion controller.
132
360440
2416
06:02
So we studied salamander locomotion in more detail,
133
362880
2456
06:05
and we had in fact access to a very nice X-ray video machine
134
365360
3096
06:08
from Professor Martin Fischer in Jena University in Germany.
135
368480
3576
06:12
And thanks to that, you really have an amazing machine
136
372080
2576
06:14
to record all the bone motion in great detail.
137
374680
2456
06:17
That's what we did.
138
377160
1256
06:18
So we basically figured out which bones are important for us
139
378440
3176
06:21
and collected their motion in 3D.
140
381640
3016
06:24
And what we did is collect a whole database of motions,
141
384680
2696
06:27
both on ground and in water,
142
387400
1656
06:29
to really collect a whole database of motor behaviors
143
389080
2484
06:31
that a real animal can do.
144
391589
1244
06:32
And then our job as roboticists was to replicate that in our robot.
145
392858
3150
06:36
So we did a whole optimization process to find out the right structure,
146
396033
3383
06:39
where to place the motors, how to connect them together,
147
399440
2656
06:42
to be able to replay these motions as well as possible.
148
402120
2880
06:45
And this is how Pleurobot came to life.
149
405680
2360
06:49
So let's look at how close it is to the real animal.
150
409200
2416
06:52
So what you see here is almost a direct comparison
151
412960
2496
06:55
between the walking of the real animal and the Pleurobot.
152
415480
2696
06:58
You can see that we have almost a one-to-one exact replay
153
418200
2736
07:00
of the walking gait.
154
420960
1256
07:02
If you go backwards and slowly, you see it even better.
155
422240
2600
07:07
But even better, we can do swimming.
156
427520
2376
07:09
So for that we have a dry suit that we put all over the robot --
157
429920
3016
07:12
(Laughter)
158
432960
1096
07:14
and then we can go in water and start replaying the swimming gaits.
159
434080
3176
07:17
And here, we were very happy, because this is difficult to do.
160
437280
3336
07:20
The physics of interaction are complex.
161
440640
2216
07:22
Our robot is much bigger than a small animal,
162
442880
2416
07:25
so we had to do what's called dynamic scaling of the frequencies
163
445320
3056
07:28
to make sure we had the same interaction physics.
164
448400
2336
07:30
But you see at the end, we have a very close match,
165
450760
2416
07:33
and we were very, very happy with this.
166
453200
1880
07:35
So let's go to the spinal cord.
167
455480
2216
07:37
So here what we did with Jean-Marie Cabelguen
168
457720
2296
07:40
is model the spinal cord circuits.
169
460040
2240
07:43
And what's interesting is that the salamander
170
463040
2136
07:45
has kept a very primitive circuit,
171
465200
1620
07:46
which is very similar to the one we find in the lamprey,
172
466844
2652
07:49
this primitive eel-like fish,
173
469520
1976
07:51
and it looks like during evolution,
174
471520
1736
07:53
new neural oscillators have been added to control the limbs,
175
473280
2936
07:56
to do the leg locomotion.
176
476240
1416
07:57
And we know where these neural oscillators are
177
477680
2176
07:59
but what we did was to make a mathematical model
178
479880
2256
08:02
to see how they should be coupled
179
482160
1616
08:03
to allow this transition between the two very different gaits.
180
483800
2936
08:06
And we tested that on board of a robot.
181
486760
2560
08:09
And this is how it looks.
182
489680
1200
08:18
So what you see here is a previous version of Pleurobot
183
498920
3016
08:21
that's completely controlled by our spinal cord model
184
501960
3096
08:25
programmed on board of the robot.
185
505080
1600
08:27
And the only thing we do
186
507280
1216
08:28
is send to the robot through a remote control
187
508520
2176
08:30
the two descending signals it normally should receive
188
510720
2496
08:33
from the upper part of the brain.
189
513240
1600
08:35
And what's interesting is, by playing with these signals,
190
515480
2696
08:38
we can completely control speed, heading and type of gait.
191
518200
2800
08:41
For instance,
192
521600
1216
08:42
when we stimulate at a low level, we have the walking gait,
193
522840
3576
08:46
and at some point, if we stimulate a lot,
194
526440
1976
08:48
very rapidly it switches to the swimming gait.
195
528440
2160
08:51
And finally, we can also do turning very nicely
196
531480
2216
08:53
by just stimulating more one side of the spinal cord than the other.
197
533720
3520
08:58
And I think it's really beautiful
198
538200
1616
08:59
how nature has distributed control
199
539840
2256
09:02
to really give a lot of responsibility to the spinal cord
200
542120
2856
09:05
so that the upper part of the brain doesn't need to worry about every muscle.
201
545000
3656
09:08
It just has to worry about this high-level modulation,
202
548680
2536
09:11
and it's really the job of the spinal cord to coordinate all the muscles.
203
551240
3576
09:14
So now let's go to cat locomotion and the importance of biomechanics.
204
554840
3520
09:19
So this is another project
205
559080
1256
09:20
where we studied cat biomechanics,
206
560360
2416
09:22
and we wanted to see how much the morphology helps locomotion.
207
562800
3896
09:26
And we found three important criteria in the properties,
208
566720
3616
09:30
basically, of the limbs.
209
570360
1320
09:32
The first one is that a cat limb
210
572320
1976
09:34
more or less looks like a pantograph-like structure.
211
574320
2696
09:37
So a pantograph is a mechanical structure
212
577040
2216
09:39
which keeps the upper segment and the lower segments always parallel.
213
579280
3400
09:43
So a simple geometrical system that kind of coordinates a bit
214
583600
3096
09:46
the internal movement of the segments.
215
586720
1816
09:48
A second property of cat limbs is that they are very lightweight.
216
588560
3056
09:51
Most of the muscles are in the trunk,
217
591640
1856
09:53
which is a good idea, because then the limbs have low inertia
218
593520
2896
09:56
and can be moved very rapidly.
219
596440
1776
09:58
The last final important property is this very elastic behavior of the cat limb,
220
598240
3816
10:02
so to handle impacts and forces.
221
602080
2656
10:04
And this is how we designed Cheetah-Cub.
222
604760
2336
10:07
So let's invite Cheetah-Cub onstage.
223
607120
2200
10:14
So this is Peter Eckert, who does his PhD on this robot,
224
614160
3656
10:17
and as you see, it's a cute little robot.
225
617840
2056
10:19
It looks a bit like a toy,
226
619920
1256
10:21
but it was really used as a scientific tool
227
621200
2056
10:23
to investigate these properties of the legs of the cat.
228
623280
3296
10:26
So you see, it's very compliant, very lightweight,
229
626600
2616
10:29
and also very elastic,
230
629240
1256
10:30
so you can easily press it down and it will not break.
231
630520
2776
10:33
It will just jump, in fact.
232
633320
1456
10:34
And this very elastic property is also very important.
233
634800
2880
10:39
And you also see a bit these properties
234
639160
1896
10:41
of these three segments of the leg as pantograph.
235
641080
2400
10:44
Now, what's interesting is that this quite dynamic gait
236
644280
2776
10:47
is obtained purely in open loop,
237
647080
1896
10:49
meaning no sensors, no complex feedback loops.
238
649000
3136
10:52
And that's interesting, because it means
239
652160
2416
10:54
that just the mechanics already stabilized this quite rapid gait,
240
654600
4016
10:58
and that really good mechanics already basically simplify locomotion.
241
658640
4176
11:02
To the extent that we can even disturb a bit locomotion,
242
662840
3296
11:06
as you will see in the next video,
243
666160
1656
11:07
where we can for instance do some exercise where we have the robot go down a step,
244
667840
3896
11:11
and the robot will not fall over,
245
671760
1616
11:13
which was a surprise for us.
246
673400
1576
11:15
This is a small perturbation.
247
675000
1416
11:16
I was expecting the robot to immediately fall over,
248
676440
2416
11:18
because there are no sensors, no fast feedback loop.
249
678880
2436
11:21
But no, just the mechanics stabilized the gait,
250
681340
2196
11:23
and the robot doesn't fall over.
251
683560
1576
11:25
Obviously, if you make the step bigger, and if you have obstacles,
252
685160
3136
11:28
you need the full control loops and reflexes and everything.
253
688320
3656
11:32
But what's important here is that just for small perturbation,
254
692000
2936
11:34
the mechanics are right.
255
694960
1496
11:36
And I think this is a very important message
256
696480
2096
11:38
from biomechanics and robotics to neuroscience,
257
698600
2191
11:40
saying don't underestimate to what extent the body already helps locomotion.
258
700815
4680
11:47
Now, how does this relate to human locomotion?
259
707440
2160
11:49
Clearly, human locomotion is more complex than cat and salamander locomotion,
260
709960
3640
11:54
but at the same time, the nervous system of humans is very similar
261
714360
3136
11:57
to that of other vertebrates.
262
717520
1576
11:59
And especially the spinal cord
263
719120
1456
12:00
is also the key controller for locomotion in humans.
264
720600
2640
12:03
That's why, if there's a lesion of the spinal cord,
265
723760
2416
12:06
this has dramatic effects.
266
726200
1496
12:07
The person can become paraplegic or tetraplegic.
267
727720
2776
12:10
This is because the brain loses this communication
268
730520
2376
12:12
with the spinal cord.
269
732920
1256
12:14
Especially, it loses this descending modulation
270
734200
2216
12:16
to initiate and modulate locomotion.
271
736440
1920
12:19
So a big goal of neuroprosthetics
272
739640
1696
12:21
is to be able to reactivate that communication
273
741360
2376
12:23
using electrical or chemical stimulations.
274
743760
2440
12:26
And there are several teams in the world that do exactly that,
275
746840
2936
12:29
especially at EPFL.
276
749800
1216
12:31
My colleagues Grégoire Courtine and Silvestro Micera,
277
751040
2496
12:33
with whom I collaborate.
278
753560
1240
12:35
But to do this properly, it's very important to understand
279
755960
3096
12:39
how the spinal cord works,
280
759080
1736
12:40
how it interacts with the body,
281
760840
1696
12:42
and how the brain communicates with the spinal cord.
282
762560
2480
12:45
This is where the robots and models that I've presented today
283
765800
2896
12:48
will hopefully play a key role
284
768720
1896
12:50
towards these very important goals.
285
770640
2656
12:53
Thank you.
286
773320
1216
12:54
(Applause)
287
774560
4560
13:04
Bruno Giussani: Auke, I've seen in your lab other robots
288
784100
2636
13:06
that do things like swim in pollution
289
786760
2456
13:09
and measure the pollution while they swim.
290
789240
2456
13:11
But for this one,
291
791720
1216
13:12
you mentioned in your talk, like a side project,
292
792960
3480
13:17
search and rescue,
293
797640
1216
13:18
and it does have a camera on its nose.
294
798880
2176
13:21
Auke Ijspeert: Absolutely. So the robot --
295
801080
2496
13:23
We have some spin-off projects
296
803600
1429
13:25
where we would like to use the robots to do search and rescue inspection,
297
805053
3443
13:28
so this robot is now seeing you.
298
808520
1576
13:30
And the big dream is to, if you have a difficult situation
299
810120
3176
13:33
like a collapsed building or a building that is flooded,
300
813320
3616
13:36
and this is very dangerous for a rescue team or even rescue dogs,
301
816960
3336
13:40
why not send in a robot that can crawl around, swim, walk,
302
820320
2896
13:43
with a camera onboard to do inspection and identify survivors
303
823240
3176
13:46
and possibly create a communication link with the survivor.
304
826440
2776
13:49
BG: Of course, assuming the survivors don't get scared by the shape of this.
305
829240
3576
13:52
AI: Yeah, we should probably change the appearance quite a bit,
306
832840
3296
13:56
because here I guess a survivor might die of a heart attack
307
836160
2816
13:59
just of being worried that this would feed on you.
308
839000
2536
14:01
But by changing the appearance and it making it more robust,
309
841560
2856
14:04
I'm sure we can make a good tool out of it.
310
844440
2056
14:06
BG: Thank you very much. Thank you and your team.
311
846520
2286
About this website

This site will introduce you to YouTube videos that are useful for learning English. You will see English lessons taught by top-notch teachers from around the world. Double-click on the English subtitles displayed on each video page to play the video from there. The subtitles scroll in sync with the video playback. If you have any comments or requests, please contact us using this contact form.

https://forms.gle/WvT1wiN1qDtmnspy7