TED2011
Skylar Tibbits: Can we make things that make themselves?
Filmed:
Readability: 4.4
1,072,366 views
MIT researcher Skylar Tibbits works on self-assembly -- the idea that instead of building something (a chair, a skyscraper), we can create materials that build themselves, much the way a strand of DNA zips itself together. It's a big concept at early stages; Tibbits shows us three in-the-lab projects that hint at what a self-assembling future might look like.
Skylar Tibbits - Inventor
Skylar Tibbits, a TED Fellow, is an artist and computational architect working on "smart" components that can assemble themselves. Full bio
Skylar Tibbits, a TED Fellow, is an artist and computational architect working on "smart" components that can assemble themselves. Full bio
Double-click the English transcript below to play the video.
00:15
Today I'd like to show you
0
0
2000
00:17
the future of the way we make things.
1
2000
2000
00:19
I believe that soon our buildings and machines
2
4000
2000
00:21
will be self-assembling,
3
6000
2000
00:23
replicating and repairing themselves.
4
8000
2000
00:25
So I'm going to show you
5
10000
2000
00:27
what I believe is the current state of manufacturing,
6
12000
2000
00:29
and then compare that to some natural systems.
7
14000
3000
00:32
So in the current state of manufacturing, we have skyscrapers --
8
17000
3000
00:35
two and a half years [of assembly time],
9
20000
2000
00:37
500,000 to a million parts,
10
22000
2000
00:39
fairly complex,
11
24000
2000
00:41
new, exciting technologies in steel, concrete, glass.
12
26000
3000
00:44
We have exciting machines
13
29000
2000
00:46
that can take us into space --
14
31000
2000
00:48
five years [of assembly time], 2.5 million parts.
15
33000
3000
00:51
But on the other side, if you look at the natural systems,
16
36000
3000
00:54
we have proteins
17
39000
2000
00:56
that have two million types,
18
41000
2000
00:58
can fold in 10,000 nanoseconds,
19
43000
2000
01:00
or DNA with three billion base pairs
20
45000
2000
01:02
we can replicate in roughly an hour.
21
47000
3000
01:05
So there's all of this complexity
22
50000
2000
01:07
in our natural systems,
23
52000
2000
01:09
but they're extremely efficient,
24
54000
2000
01:11
far more efficient than anything we can build,
25
56000
2000
01:13
far more complex than anything we can build.
26
58000
2000
01:15
They're far more efficient in terms of energy.
27
60000
2000
01:17
They hardly ever make mistakes.
28
62000
3000
01:20
And they can repair themselves for longevity.
29
65000
2000
01:22
So there's something super interesting about natural systems.
30
67000
3000
01:25
And if we can translate that
31
70000
2000
01:27
into our built environment,
32
72000
2000
01:29
then there's some exciting potential for the way that we build things.
33
74000
2000
01:31
And I think the key to that is self-assembly.
34
76000
3000
01:34
So if we want to utilize self-assembly in our physical environment,
35
79000
3000
01:37
I think there's four key factors.
36
82000
2000
01:39
The first is that we need to decode
37
84000
2000
01:41
all of the complexity of what we want to build --
38
86000
2000
01:43
so our buildings and machines.
39
88000
2000
01:45
And we need to decode that into simple sequences --
40
90000
2000
01:47
basically the DNA of how our buildings work.
41
92000
2000
01:49
Then we need programmable parts
42
94000
2000
01:51
that can take that sequence
43
96000
2000
01:53
and use that to fold up, or reconfigure.
44
98000
3000
01:56
We need some energy that's going to allow that to activate,
45
101000
3000
01:59
allow our parts to be able to fold up from the program.
46
104000
3000
02:02
And we need some type of error correction redundancy
47
107000
2000
02:04
to guarantee that we have successfully built what we want.
48
109000
3000
02:07
So I'm going to show you a number of projects
49
112000
2000
02:09
that my colleagues and I at MIT are working on
50
114000
2000
02:11
to achieve this self-assembling future.
51
116000
2000
02:13
The first two are the MacroBot and DeciBot.
52
118000
3000
02:16
So these projects are large-scale reconfigurable robots --
53
121000
4000
02:20
8 ft., 12 ft. long proteins.
54
125000
3000
02:23
They're embedded with mechanical electrical devices, sensors.
55
128000
3000
02:26
You decode what you want to fold up into,
56
131000
2000
02:28
into a sequence of angles --
57
133000
2000
02:30
so negative 120, negative 120, 0, 0,
58
135000
2000
02:32
120, negative 120 -- something like that;
59
137000
3000
02:35
so a sequence of angles, or turns,
60
140000
2000
02:37
and you send that sequence through the string.
61
142000
3000
02:40
Each unit takes its message -- so negative 120 --
62
145000
3000
02:43
it rotates to that, checks if it got there
63
148000
2000
02:45
and then passes it to its neighbor.
64
150000
3000
02:48
So these are the brilliant scientists,
65
153000
2000
02:50
engineers, designers that worked on this project.
66
155000
2000
02:52
And I think it really brings to light:
67
157000
2000
02:54
Is this really scalable?
68
159000
2000
02:56
I mean, thousands of dollars, lots of man hours
69
161000
2000
02:58
made to make this eight-foot robot.
70
163000
3000
03:01
Can we really scale this up? Can we really embed robotics into every part?
71
166000
3000
03:04
The next one questions that
72
169000
2000
03:06
and looks at passive nature,
73
171000
2000
03:08
or passively trying to have reconfiguration programmability.
74
173000
3000
03:11
But it goes a step further,
75
176000
2000
03:13
and it tries to have actual computation.
76
178000
2000
03:15
It basically embeds the most fundamental building block of computing,
77
180000
2000
03:17
the digital logic gate,
78
182000
2000
03:19
directly into your parts.
79
184000
2000
03:21
So this is a NAND gate.
80
186000
2000
03:23
You have one tetrahedron which is the gate
81
188000
2000
03:25
that's going to do your computing,
82
190000
2000
03:27
and you have two input tetrahedrons.
83
192000
2000
03:29
One of them is the input from the user, as you're building your bricks.
84
194000
3000
03:32
The other one is from the previous brick that was placed.
85
197000
3000
03:35
And then it gives you an output in 3D space.
86
200000
3000
03:38
So what this means
87
203000
2000
03:40
is that the user can start plugging in what they want the bricks to do.
88
205000
3000
03:43
It computes on what it was doing before
89
208000
2000
03:45
and what you said you wanted it to do.
90
210000
2000
03:47
And now it starts moving in three-dimensional space --
91
212000
2000
03:49
so up or down.
92
214000
2000
03:51
So on the left-hand side, [1,1] input equals 0 output, which goes down.
93
216000
3000
03:54
On the right-hand side,
94
219000
2000
03:56
[0,0] input is a 1 output, which goes up.
95
221000
3000
03:59
And so what that really means
96
224000
2000
04:01
is that our structures now contain the blueprints
97
226000
2000
04:03
of what we want to build.
98
228000
2000
04:05
So they have all of the information embedded in them of what was constructed.
99
230000
3000
04:08
So that means that we can have some form of self-replication.
100
233000
3000
04:11
In this case I call it self-guided replication,
101
236000
3000
04:14
because your structure contains the exact blueprints.
102
239000
2000
04:16
If you have errors, you can replace a part.
103
241000
2000
04:18
All the local information is embedded to tell you how to fix it.
104
243000
3000
04:21
So you could have something that climbs along and reads it
105
246000
2000
04:23
and can output at one to one.
106
248000
2000
04:25
It's directly embedded; there's no external instructions.
107
250000
2000
04:27
So the last project I'll show is called Biased Chains,
108
252000
3000
04:30
and it's probably the most exciting example that we have right now
109
255000
3000
04:33
of passive self-assembly systems.
110
258000
2000
04:35
So it takes the reconfigurability
111
260000
2000
04:37
and programmability
112
262000
2000
04:39
and makes it a completely passive system.
113
264000
3000
04:43
So basically you have a chain of elements.
114
268000
2000
04:45
Each element is completely identical,
115
270000
2000
04:47
and they're biased.
116
272000
2000
04:49
So each chain, or each element, wants to turn right or left.
117
274000
3000
04:52
So as you assemble the chain, you're basically programming it.
118
277000
3000
04:55
You're telling each unit if it should turn right or left.
119
280000
3000
04:58
So when you shake the chain,
120
283000
3000
05:01
it then folds up
121
286000
2000
05:03
into any configuration that you've programmed in --
122
288000
3000
05:06
so in this case, a spiral,
123
291000
2000
05:08
or in this case,
124
293000
3000
05:11
two cubes next to each other.
125
296000
3000
05:14
So you can basically program
126
299000
2000
05:16
any three-dimensional shape --
127
301000
2000
05:18
or one-dimensional, two-dimensional -- up into this chain completely passively.
128
303000
3000
05:21
So what does this tell us about the future?
129
306000
2000
05:23
I think that it's telling us
130
308000
2000
05:25
that there's new possibilities for self-assembly, replication, repair
131
310000
3000
05:28
in our physical structures, our buildings, machines.
132
313000
3000
05:31
There's new programmability in these parts.
133
316000
2000
05:33
And from that you have new possibilities for computing.
134
318000
2000
05:35
We'll have spatial computing.
135
320000
2000
05:37
Imagine if our buildings, our bridges, machines,
136
322000
2000
05:39
all of our bricks could actually compute.
137
324000
2000
05:41
That's amazing parallel and distributed computing power,
138
326000
2000
05:43
new design possibilities.
139
328000
2000
05:45
So it's exciting potential for this.
140
330000
2000
05:47
So I think these projects I've showed here
141
332000
2000
05:49
are just a tiny step towards this future,
142
334000
2000
05:51
if we implement these new technologies
143
336000
2000
05:53
for a new self-assembling world.
144
338000
2000
05:55
Thank you.
145
340000
2000
05:57
(Applause)
146
342000
2000
ABOUT THE SPEAKER
Skylar Tibbits - InventorSkylar Tibbits, a TED Fellow, is an artist and computational architect working on "smart" components that can assemble themselves.
Why you should listen
Can we create objects that assemble themselves -- that zip together like a strand of DNA or that have the ability for transformation embedded into them? These are the questions that Skylar Tibbits investigates in his Self-Assembly Lab at MIT, a cross-disciplinary research space where designers, scientists and engineers come together to find ways for disordered parts to become ordered structures.
A trained architect, designer and computer scientist, Tibbits teaches design studios at MIT’s Department of Architecture and co-teaches the seminar “How to Make (Almost) Anything” at MIT’s Media Lab. Before that, he worked at a number of design offices including Zaha Hadid Architects, Asymptote Architecture, SKIII Space Variations and Point b Design. His work has been shown at the Guggenheim Museum and the Beijing Biennale.
Tibbits has collaborated with a number of influential people over the years, including Neil Gershenfeld and The Center for Bits and Atoms, Erik and Marty Demaine at MIT, Adam Bly at SEED Media Group and Marc Fornes of THEVERYMANY. In 2007, he and Marc Fornes co-curated Scriptedbypurpose, the first exhibition focused exclusively on scripted processes within design. Also in 2007, he founded SJET, a multifaceted practice and research platform for experimental computation and design. SJET crosses disciplines from architecture and design, fabrication, computer science and robotics.
Skylar Tibbits | Speaker | TED.com