TED2007
Hod Lipson: Building "self-aware" robots
Filmed:
Readability: 3.1
1,460,460 views
Hod Lipson demonstrates a few of his cool little robots, which have the ability to learn, understand themselves and even self-replicate.
Hod Lipson - Roboticist
Hod Lipson works at the intersection of engineering and biology, studying robots and the way they "behave" and evolve. His work has exciting implications for design and manufacturing -- and serves as a window to understand our own behavior and evolution. Full bio
Hod Lipson works at the intersection of engineering and biology, studying robots and the way they "behave" and evolve. His work has exciting implications for design and manufacturing -- and serves as a window to understand our own behavior and evolution. Full bio
Double-click the English transcript below to play the video.
00:25
So, where are the robots?
0
0
2000
00:27
We've been told for 40 years already that they're coming soon.
1
2000
3000
00:30
Very soon they'll be doing everything for us.
2
5000
3000
00:33
They'll be cooking, cleaning, buying things, shopping, building. But they aren't here.
3
8000
5000
00:38
Meanwhile, we have illegal immigrants doing all the work,
4
13000
4000
00:42
but we don't have any robots.
5
17000
2000
00:44
So what can we do about that? What can we say?
6
19000
4000
00:48
So I want to give a little bit of a different perspective
7
23000
4000
00:52
of how we can perhaps look at these things in a little bit of a different way.
8
27000
6000
00:58
And this is an x-ray picture
9
33000
2000
01:00
of a real beetle, and a Swiss watch, back from '88. You look at that --
10
35000
5000
01:05
what was true then is certainly true today.
11
40000
2000
01:07
We can still make the pieces. We can make the right pieces.
12
42000
3000
01:10
We can make the circuitry of the right computational power,
13
45000
3000
01:13
but we can't actually put them together to make something
14
48000
3000
01:16
that will actually work and be as adaptive as these systems.
15
51000
5000
01:21
So let's try to look at it from a different perspective.
16
56000
2000
01:23
Let's summon the best designer, the mother of all designers.
17
58000
4000
01:27
Let's see what evolution can do for us.
18
62000
3000
01:30
So we threw in -- we created a primordial soup
19
65000
4000
01:34
with lots of pieces of robots -- with bars, with motors, with neurons.
20
69000
4000
01:38
Put them all together, and put all this under kind of natural selection,
21
73000
4000
01:42
under mutation, and rewarded things for how well they can move forward.
22
77000
4000
01:46
A very simple task, and it's interesting to see what kind of things came out of that.
23
81000
6000
01:52
So if you look, you can see a lot of different machines
24
87000
3000
01:55
come out of this. They all move around.
25
90000
2000
01:57
They all crawl in different ways, and you can see on the right,
26
92000
4000
02:01
that we actually made a couple of these things,
27
96000
2000
02:03
and they work in reality. These are not very fantastic robots,
28
98000
3000
02:06
but they evolved to do exactly what we reward them for:
29
101000
4000
02:10
for moving forward. So that was all done in simulation,
30
105000
3000
02:13
but we can also do that on a real machine.
31
108000
2000
02:15
Here's a physical robot that we actually
32
110000
5000
02:20
have a population of brains,
33
115000
3000
02:23
competing, or evolving on the machine.
34
118000
2000
02:25
It's like a rodeo show. They all get a ride on the machine,
35
120000
3000
02:28
and they get rewarded for how fast or how far
36
123000
3000
02:31
they can make the machine move forward.
37
126000
2000
02:33
And you can see these robots are not ready
38
128000
2000
02:35
to take over the world yet, but
39
130000
3000
02:38
they gradually learn how to move forward,
40
133000
2000
02:40
and they do this autonomously.
41
135000
3000
02:43
So in these two examples, we had basically
42
138000
4000
02:47
machines that learned how to walk in simulation,
43
142000
3000
02:50
and also machines that learned how to walk in reality.
44
145000
2000
02:52
But I want to show you a different approach,
45
147000
2000
02:54
and this is this robot over here, which has four legs.
46
149000
6000
03:00
It has eight motors, four on the knees and four on the hip.
47
155000
2000
03:02
It has also two tilt sensors that tell the machine
48
157000
3000
03:05
which way it's tilting.
49
160000
3000
03:08
But this machine doesn't know what it looks like.
50
163000
2000
03:10
You look at it and you see it has four legs,
51
165000
2000
03:12
the machine doesn't know if it's a snake, if it's a tree,
52
167000
2000
03:14
it doesn't have any idea what it looks like,
53
169000
3000
03:17
but it's going to try to find that out.
54
172000
2000
03:19
Initially, it does some random motion,
55
174000
2000
03:21
and then it tries to figure out what it might look like.
56
176000
3000
03:24
And you're seeing a lot of things passing through its minds,
57
179000
2000
03:26
a lot of self-models that try to explain the relationship
58
181000
4000
03:30
between actuation and sensing. It then tries to do
59
185000
3000
03:33
a second action that creates the most disagreement
60
188000
4000
03:37
among predictions of these alternative models,
61
192000
2000
03:39
like a scientist in a lab. Then it does that
62
194000
2000
03:41
and tries to explain that, and prune out its self-models.
63
196000
4000
03:45
This is the last cycle, and you can see it's pretty much
64
200000
3000
03:48
figured out what its self looks like. And once it has a self-model,
65
203000
4000
03:52
it can use that to derive a pattern of locomotion.
66
207000
4000
03:56
So what you're seeing here are a couple of machines --
67
211000
2000
03:58
a pattern of locomotion.
68
213000
2000
04:00
We were hoping that it wass going to have a kind of evil, spidery walk,
69
215000
4000
04:04
but instead it created this pretty lame way of moving forward.
70
219000
4000
04:08
But when you look at that, you have to remember
71
223000
3000
04:11
that this machine did not do any physical trials on how to move forward,
72
226000
6000
04:17
nor did it have a model of itself.
73
232000
2000
04:19
It kind of figured out what it looks like, and how to move forward,
74
234000
3000
04:22
and then actually tried that out.
75
237000
4000
04:26
(Applause)
76
241000
5000
04:31
So, we'll move forward to a different idea.
77
246000
4000
04:35
So that was what happened when we had a couple of --
78
250000
5000
04:40
that's what happened when you had a couple of -- OK, OK, OK --
79
255000
4000
04:44
(Laughter)
80
259000
2000
04:46
-- they don't like each other. So
81
261000
2000
04:48
there's a different robot.
82
263000
3000
04:51
That's what happened when the robots actually
83
266000
2000
04:53
are rewarded for doing something.
84
268000
2000
04:55
What happens if you don't reward them for anything, you just throw them in?
85
270000
3000
04:58
So we have these cubes, like the diagram showed here.
86
273000
3000
05:01
The cube can swivel, or flip on its side,
87
276000
2000
05:04
and we just throw 1,000 of these cubes into a soup --
88
279000
4000
05:08
this is in simulation --and don't reward them for anything,
89
283000
2000
05:10
we just let them flip. We pump energy into this
90
285000
3000
05:13
and see what happens in a couple of mutations.
91
288000
3000
05:16
So, initially nothing happens, they're just flipping around there.
92
291000
3000
05:19
But after a very short while, you can see these blue things
93
294000
4000
05:23
on the right there begin to take over.
94
298000
2000
05:25
They begin to self-replicate. So in absence of any reward,
95
300000
4000
05:29
the intrinsic reward is self-replication.
96
304000
3000
05:32
And we've actually built a couple of these,
97
307000
1000
05:33
and this is part of a larger robot made out of these cubes.
98
308000
4000
05:37
It's an accelerated view, where you can see the robot actually
99
312000
3000
05:40
carrying out some of its replication process.
100
315000
2000
05:42
So you're feeding it with more material -- cubes in this case --
101
317000
4000
05:46
and more energy, and it can make another robot.
102
321000
3000
05:49
So of course, this is a very crude machine,
103
324000
3000
05:52
but we're working on a micro-scale version of these,
104
327000
2000
05:54
and hopefully the cubes will be like a powder that you pour in.
105
329000
3000
05:57
OK, so what can we learn? These robots are of course
106
332000
5000
06:02
not very useful in themselves, but they might teach us something
107
337000
3000
06:05
about how we can build better robots,
108
340000
3000
06:08
and perhaps how humans, animals, create self-models and learn.
109
343000
5000
06:13
And one of the things that I think is important
110
348000
2000
06:15
is that we have to get away from this idea
111
350000
2000
06:17
of designing the machines manually,
112
352000
2000
06:19
but actually let them evolve and learn, like children,
113
354000
3000
06:22
and perhaps that's the way we'll get there. Thank you.
114
357000
2000
06:24
(Applause)
115
359000
2000
ABOUT THE SPEAKER
Hod Lipson - RoboticistHod Lipson works at the intersection of engineering and biology, studying robots and the way they "behave" and evolve. His work has exciting implications for design and manufacturing -- and serves as a window to understand our own behavior and evolution.
Why you should listen
To say that Hod Lipson and his team at Cornell build robots is not completely accurate: They may simply set out a pile of virtual robot parts, devise some rules for assembly, and see what the parts build themselves into. They've created robots that decide for themselves how they want to walk; robots that develop a sense of what they look like; even robots that can, through trial and error, construct other robots just like themselves.
Working across disciplines -- physics, computer science, math, biology and several flavors of engineer -- the team studies techniques for self-assembly and evolution that have great implications for fields such as micro-manufacturing -- allowing tiny pieces to assemble themselves at scales heretofore impossible -- and extreme custom manufacturing (in other words, 3-D printers for the home).
His lab's Outreach page is a funhouse of tools and instructions, including the amazing Golem@Home -- a self-assembling virtual robot who lives in your screensaver.
Hod Lipson | Speaker | TED.com