TEDGlobal 2012
Michael Hansmeyer: Building unimaginable shapes
Filmed:
Readability: 4.1
927,090 views
Inspired by cell division, Michael Hansmeyer writes algorithms that design outrageously fascinating shapes and forms with millions of facets. No person could draft them by hand, but they're buildable -- and they could revolutionize the way we think of architectural form.
Michael Hansmeyer - Computational architect
Michael Hansmeyer is an architect and programmer who explores the use of algorithms and computation to generate architectural form. Full bio
Michael Hansmeyer is an architect and programmer who explores the use of algorithms and computation to generate architectural form. Full bio
Double-click the English transcript below to play the video.
00:16
As an architect, I often ask myself,
0
470
2449
00:18
what is the origin of the forms that we design?
1
2919
2840
00:21
What kind of forms could we design
2
5759
2905
00:24
if we wouldn't work with references anymore?
3
8664
2823
00:27
If we had no bias, if we had no preconceptions,
4
11487
3385
00:30
what kind of forms could we design
5
14872
2207
00:32
if we could free ourselves from
6
17079
1600
00:34
our experience?
7
18679
2136
00:36
If we could free ourselves from our education?
8
20815
4734
00:41
What would these unseen forms look like?
9
25549
3027
00:44
Would they surprise us? Would they intrigue us?
10
28576
3711
00:48
Would they delight us?
11
32287
2415
00:50
If so, then how can we go about creating something that is truly new?
12
34702
3921
00:54
I propose we look to nature.
13
38623
2392
00:56
Nature has been called the greatest architect of forms.
14
41015
4264
01:01
And I'm not saying that we should copy nature,
15
45279
3588
01:04
I'm not saying we should mimic biology,
16
48867
2380
01:07
instead I propose that we can borrow nature's processes.
17
51247
3448
01:10
We can abstract them and to create something that is new.
18
54695
4454
01:15
Nature's main process of creation, morphogenesis,
19
59149
3878
01:18
is the splitting of one cell into two cells.
20
63027
3829
01:22
And these cells can either be identical,
21
66856
2057
01:24
or they can be distinct from each other
22
68913
2175
01:26
through asymmetric cell division.
23
71088
2064
01:29
If we abstract this process, and simplify it as much as possible,
24
73152
3799
01:32
then we could start with a single sheet of paper,
25
76951
2233
01:35
one surface, and we could make a fold
26
79184
2496
01:37
and divide the surface into two surfaces.
27
81680
2936
01:40
We're free to choose where we make the fold.
28
84616
2263
01:42
And by doing so, we can differentiate the surfaces.
29
86879
4910
01:47
Through this very simple process,
30
91789
1894
01:49
we can create an astounding variety of forms.
31
93683
3612
01:53
Now, we can take this form and use the same process
32
97295
2725
01:55
to generate three-dimensional structures,
33
100020
1992
01:57
but rather than folding things by hand,
34
102012
2408
02:00
we'll bring the structure into the computer,
35
104420
2585
02:02
and code it as an algorithm.
36
107005
2975
02:05
And in doing so, we can suddenly fold anything.
37
109980
3008
02:08
We can fold a million times faster,
38
112988
2159
02:11
we can fold in hundreds and hundreds of variations.
39
115147
3161
02:14
And as we're seeking to make something three-dimensional,
40
118308
2480
02:16
we start not with a single surface, but with a volume.
41
120788
3216
02:19
A simple volume, the cube.
42
124004
1353
02:21
If we take its surfaces and fold them
43
125357
1720
02:22
again and again and again and again,
44
127077
2056
02:25
then after 16 iterations, 16 steps,
45
129133
2863
02:27
we end up with 400,000 surfaces and a shape that looks,
46
131996
3849
02:31
for instance, like this.
47
135845
2535
02:34
And if we change where we make the folds,
48
138380
2712
02:36
if we change the folding ratio,
49
141092
1641
02:38
then this cube turns into this one.
50
142733
3423
02:42
We can change the folding ratio again to produce this shape,
51
146156
4139
02:46
or this shape.
52
150295
1827
02:48
So we exert control over the form
53
152122
1623
02:49
by specifying the position of where we're making the fold,
54
153745
3504
02:53
but essentially you're looking at a folded cube.
55
157249
4706
02:57
And we can play with this.
56
161955
1064
02:58
We can apply different folding ratios to different parts
57
163019
2616
03:01
of the form to create local conditions.
58
165635
2745
03:04
We can begin to sculpt the form.
59
168380
1943
03:06
And because we're doing the folding on the computer,
60
170323
2888
03:09
we are completely free of any physical constraints.
61
173211
3696
03:12
So that means that surfaces can intersect themselves,
62
176907
2951
03:15
they can become impossibly small.
63
179858
1489
03:17
We can make folds that we otherwise could not make.
64
181347
3601
03:20
Surfaces can become porous.
65
184948
1878
03:22
They can stretch. They can tear.
66
186826
2729
03:25
And all of this expounds the scope of forms that we can produce.
67
189555
4635
03:30
But in each case, I didn't design the form.
68
194190
2750
03:32
I designed the process that generated the form.
69
196940
4601
03:37
In general, if we make a small change to the folding ratio,
70
201541
4488
03:41
which is what you're seeing here,
71
206029
1595
03:43
then the form changes correspondingly.
72
207624
3773
03:47
But that's only half of the story --
73
211397
2256
03:49
99.9 percent of the folding ratios produce not this,
74
213653
4323
03:53
but this, the geometric equivalent of noise.
75
217976
4896
03:58
The forms that I showed before were made actually
76
222872
2120
04:00
through very long trial and error.
77
224992
1648
04:02
A far more effective way to create forms, I have found,
78
226640
3216
04:05
is to use information that is already contained in forms.
79
229856
3888
04:09
A very simple form such as this one actually contains
80
233744
2655
04:12
a lot of information that may not be visible to the human eye.
81
236399
3592
04:15
So, for instance, we can plot the length of the edges.
82
239991
2481
04:18
White surfaces have long edges, black ones have short ones.
83
242472
3336
04:21
We can plot the planarity of the surfaces, their curvature,
84
245808
3577
04:25
how radial they are -- all information that may not be
85
249385
4087
04:29
instantly visible to you,
86
253472
1947
04:31
but that we can bring out, that we can articulate,
87
255419
2425
04:33
and that we can use to control the folding.
88
257844
3376
04:37
So now I'm not specifying a single
89
261220
1959
04:39
ratio anymore to fold it,
90
263179
2176
04:41
but instead I'm establishing a rule,
91
265355
2552
04:43
I'm establishing a link between a property of a surface
92
267907
2436
04:46
and how that surface is folded.
93
270343
2973
04:49
And because I've designed the process and not the form,
94
273316
2951
04:52
I can run the process again and again and again
95
276267
2322
04:54
to produce a whole family of forms.
96
278589
2535
05:09
These forms look elaborate, but the process is a very minimal one.
97
293385
4443
05:13
There is a simple input,
98
297828
1129
05:14
it's always a cube that I start with,
99
298957
1952
05:16
and it's a very simple operation -- it's making a fold,
100
300909
3576
05:20
and doing this over and over again.
101
304485
3952
05:24
So let's bring this process to architecture.
102
308437
2280
05:26
How? And at what scale?
103
310717
1600
05:28
I chose to design a column.
104
312317
1728
05:29
Columns are architectural archetypes.
105
314045
3376
05:33
They've been used throughout history to express ideals
106
317421
3067
05:36
about beauty, about technology.
107
320488
5075
05:41
A challenge to me was how we could express
108
325563
1922
05:43
this new algorithmic order in a column.
109
327485
3296
05:46
I started using four cylinders.
110
330781
2984
05:49
Through a lot of experimentation, these cylinders
111
333765
3776
05:53
eventually evolved into this.
112
337541
3135
05:56
And these columns, they have information at very many scales.
113
340676
4616
06:01
We can begin to zoom into them.
114
345292
2518
06:03
The closer one gets, the more new features one discovers.
115
347810
3555
06:07
Some formations are almost at the threshold of human visibility.
116
351365
3672
06:10
And unlike traditional architecture,
117
355037
1952
06:12
it's a single process that creates both the overall form
118
356989
2784
06:15
and the microscopic surface detail.
119
359773
5345
06:21
These forms are undrawable.
120
365118
2691
06:23
An architect who's drawing them with a pen and a paper
121
367809
3376
06:27
would probably take months,
122
371185
1792
06:28
or it would take even a year to draw all the sections,
123
372977
2384
06:31
all of the elevations, you can only create something like this
124
375361
2304
06:33
through an algorithm.
125
377665
2136
06:35
The more interesting question, perhaps, is,
126
379801
2032
06:37
are these forms imaginable?
127
381833
2629
06:40
Usually, an architect can somehow envision the end state
128
384462
2635
06:42
of what he is designing.
129
387097
1864
06:44
In this case, the process is deterministic.
130
388961
2825
06:47
There's no randomness involved at all,
131
391786
2359
06:50
but it's not entirely predictable.
132
394145
2007
06:52
There's too many surfaces,
133
396152
1576
06:53
there's too much detail, one can't see the end state.
134
397728
3610
06:57
So this leads to a new role for the architect.
135
401338
3214
07:00
One needs a new method to explore all of the possibilities
136
404552
3611
07:04
that are out there.
137
408163
1776
07:05
For one thing, one can design many variants of a form,
138
409939
3232
07:09
in parallel, and one can cultivate them.
139
413171
2214
07:11
And to go back to the analogy with nature,
140
415385
2248
07:13
one can begin to think in terms of populations,
141
417633
2496
07:16
one can talk about permutations, about generations,
142
420129
3528
07:19
about crossing and breeding to come up with a design.
143
423657
4863
07:24
And the architect is really, he moves into the position
144
428520
2385
07:26
of being an orchestrator of all of these processes.
145
430905
3432
07:30
But enough of the theory.
146
434337
2344
07:32
At one point I simply wanted to jump inside
147
436681
2527
07:35
this image, so to say, I bought these red and blue
148
439208
3829
07:38
3D glasses, got up very close to the screen,
149
443037
2936
07:41
but still that wasn't the same as being able to
150
445973
2072
07:43
walk around and touch things.
151
448045
2352
07:46
So there was only one possibility --
152
450397
1903
07:48
to bring the column out of the computer.
153
452300
3097
07:51
There's been a lot of talk now about 3D printing.
154
455397
2920
07:54
For me, or for my purpose at this moment,
155
458317
2959
07:57
there's still too much of an unfavorable tradeoff
156
461276
3082
08:00
between scale, on the one hand, and resolution and speed, on the other.
157
464358
6783
08:07
So instead, we decided to take the column,
158
471141
2257
08:09
and we decided to build it as a layered model,
159
473398
2439
08:11
made out of very many slices, thinly stacked over each other.
160
475837
4128
08:15
What you're looking at here is an X-ray
161
479965
1897
08:17
of the column that you just saw, viewed from the top.
162
481862
2775
08:20
Unbeknownst to me at the time,
163
484637
2000
08:22
because we had only seen the outside,
164
486637
2376
08:24
the surfaces were continuing to fold themselves,
165
489013
2160
08:27
to grow on the inside of the column,
166
491173
2264
08:29
which was quite a surprising discovery.
167
493437
2546
08:31
From this shape, we calculated a cutting line,
168
495983
3614
08:35
and then we gave this cutting line to a laser cutter
169
499597
3024
08:38
to produce -- and you're seeing a segment of it here --
170
502621
3814
08:42
very many thin slices, individually cut, on top of each other.
171
506435
4891
08:49
And this is a photo now, it's not a rendering,
172
513480
2813
08:52
and the column that we ended up with
173
516293
1800
08:53
after a lot of work, ended up looking remarkably like the one
174
518093
3047
08:57
that we had designed in the computer.
175
521140
3710
09:00
Almost all of the details, almost all of the
176
524850
2121
09:02
surface intricacies were preserved.
177
526971
3040
09:08
But it was very labor intensive.
178
532626
2346
09:10
There's a huge disconnect at the moment still
179
534972
2401
09:13
between the virtual and the physical.
180
537373
2760
09:16
It took me several months to design the column,
181
540133
2126
09:18
but ultimately it takes the computer about 30 seconds
182
542259
2718
09:20
to calculate all of the 16 million faces.
183
544977
2848
09:23
The physical model, on the other hand,
184
547825
1961
09:25
is 2,700 layers, one millimeter thick,
185
549786
4208
09:29
it weighs 700 kilos, it's made of sheet that can cover
186
553994
4147
09:34
this entire auditorium.
187
558141
2138
09:36
And the cutting path that the laser followed
188
560279
2088
09:38
goes from here to the airport and back again.
189
562367
5117
09:43
But it is increasingly possible.
190
567484
1749
09:45
Machines are getting faster, it's getting less expensive,
191
569233
2611
09:47
and there's some promising technological developments
192
571844
2760
09:50
just on the horizon.
193
574604
1783
09:52
These are images from the Gwangju Biennale.
194
576387
3072
09:55
And in this case, I used ABS plastic to produce the columns,
195
579459
3524
09:58
we used the bigger, faster machine,
196
582983
1854
10:00
and they have a steel core inside, so they're structural,
197
584837
3057
10:03
they can bear loads for once.
198
587894
2976
10:06
Each column is effectively a hybrid of two columns.
199
590870
2015
10:08
You can see a different column in the mirror,
200
592885
3409
10:12
if there's a mirror behind the column
201
596294
2050
10:14
that creates a sort of an optical illusion.
202
598344
3072
10:17
So where does this leave us?
203
601431
1831
10:19
I think this project gives us a glimpse of the unseen objects that await us
204
603262
4468
10:23
if we as architects begin to think about designing not the object,
205
607730
3883
10:27
but a process to generate objects.
206
611613
3393
10:30
I've shown one simple process that was inspired by nature;
207
615006
3273
10:34
there's countless other ones.
208
618279
2848
10:37
In short, we have no constraints.
209
621127
3447
10:40
Instead, we have processes in our hands right now
210
624574
3570
10:44
that allow us to create structures at all scales
211
628144
4581
10:48
that we couldn't even have dreamt up.
212
632725
3032
10:51
And, if I may add, at one point we will build them.
213
635757
4819
10:56
Thank you. (Applause)
214
640576
6623
ABOUT THE SPEAKER
Michael Hansmeyer - Computational architectMichael Hansmeyer is an architect and programmer who explores the use of algorithms and computation to generate architectural form.
Why you should listen
Classical architecture is defined by "orders" -- ways to connect a column to a building, to articulate the joining of materials and structural forces. Colloquially, these orders are based on elemental forms: the tree trunk, the plank, the scroll, the leaf. Michael Hansmeyer is adding a new elemental form: the subdivision algorithm. He turns his math and programming skills to making ornate, organic, hyperdetailed columns generated from lines of code and then comped up in cross-sections of cardboard, almost as if they're being 3D printed.
His recent work with cupolas and domes is even more mesmerizing, like looking deep inside an organic form of near-unbearable complexity. See images at digital-grotesque.com >>
Michael Hansmeyer | Speaker | TED.com