TED2003
Torsten Reil: Animate characters by evolving them
托斯腾·雷尔将生物研究用于动画制作
Filmed:
Readability: 3.5
363,842 views
托斯腾·雷尔探讨了怎样用生物学做出更自然的动画人物 — 通过从内到外的人物设计，通过使用骨骼，肌肉和神经系统。这是他2003年在TED的演讲。在侠盗飞车4中能看到他现在的工作成果。
Torsten Reil - Animating neurobiologist
By coding computer simulations with biologically modeled nervous systems, Torsten Reil and his company NaturalMotion breathe life into the animated characters inhabiting the most eye-poppingly realistic games and movies around. Full bio
By coding computer simulations with biologically modeled nervous systems, Torsten Reil and his company NaturalMotion breathe life into the animated characters inhabiting the most eye-poppingly realistic games and movies around. Full bio
Double-click the English transcript below to play the video.
00:15
I'm going to talk about a technology that we're developing at Oxford now,
0
3000
4000
我想谈谈一个我们正在牛津开发的技术，
00:19
that we think is going to change the way that
1
7000
3000
我们认为这个技术可以改变
00:22
computer games and Hollywood movies are being made.
2
10000
3000
电脑游戏和好莱坞电影的制作方法
00:26
That technology is simulating humans.
3
14000
3000
这个技术就是模拟人类。
00:29
It's simulated humans with a simulated body
4
17000
3000
这个模拟人类有电脑模拟的身体，
00:32
and a simulated nervous system to control that body.
5
20000
3000
和电脑模拟的神经系统来控制身体。
00:36
Now, before I talk more about that technology,
6
24000
3000
现在，在我开始讲解这个技术之前，
00:39
let's have a quick look at what human characters look like
7
27000
3000
让我们先快速浏览一下在现今的电脑游戏中
00:42
at the moment in computer games.
8
30000
3000
游戏人物角色是什么样子的。
00:45
This is a clip from a game called "Grand Theft Auto 3."
9
33000
3000
这是从一个名为“侠盗飞车3”的游戏中摘出的片断。
00:48
We already saw that briefly yesterday.
10
36000
2000
我们昨天已经见过了。
00:50
And what you can see is -- it is actually a very good game.
11
38000
3000
可以看出，这是个非常精美的游戏。
00:53
It's one of the most successful games of all time.
12
41000
3000
这是有史以来最成功的电脑游戏之一。
00:56
But what you'll see is that all the animations in this game are very repetitive.
13
44000
4000
但是你也可以发现这个游戏中所有的模拟动作都有很大的重复性。
01:00
They pretty much look the same.
14
48000
2000
所有的动作看起来都一样。
01:02
I've made him run into a wall here, over and over again.
15
50000
3000
我让这个人物一次又一次地撞到墙上，
01:05
And you can see he looks always the same.
16
53000
2000
你可以发现每次它（的反应）看起来都一样。
01:07
The reason for that is that these characters
17
55000
3000
这是因为这些动画人物
01:10
are actually not real characters.
18
58000
2000
并不是真正的“人”，
01:12
They are a graphical visualization of a character.
19
60000
4000
它们是“人”的虚拟图像。
01:16
To produce these animations, an animator at a studio has to anticipate
20
64000
5000
设计师做出这些动画人物的时候
01:21
what's going to happen in the actual game,
21
69000
3000
得要猜想游戏中将会发生什么情境，
01:24
and then has to animate that particular sequence.
22
72000
3000
然后还得根据这些特定的情境画出（动画人物的反应）。
01:27
So, he or she sits down, animates it, and tries to anticipate what's going to happen,
23
75000
4000
所以设计师得坐下来，画出动画，还要猜想将会发生什么，
01:31
and then these particular animations are just played back
24
79000
3000
所以这些人物的反应都是重复播放的设定动作，
01:34
at appropriate times in the computer game.
25
82000
3000
只是根据电脑游戏中特定的情境播放罢了。
01:37
Now, the result of that is that you can't have real interactivity.
26
85000
5000
结果游戏中并没有真的互动。
01:42
All you have is animations that are played back
27
90000
3000
有的仅仅是重复播放的动画，
01:45
at more or less the appropriate times.
28
93000
2000
关键只是选择合适的时机罢了。
01:47
It also means that games aren't really going to be as surprising as they could be,
29
95000
5000
这也说明现在的电脑游戏并不具备它们应有的新鲜度，
01:52
because you only get out of it, at least in terms of the character,
30
100000
3000
因为所有你能得到的动画效果，至少在人物动作方面，
01:55
what you actually put into it.
31
103000
2000
仅仅是设计者储存在游戏中的那些。
01:57
There's no real emergence there.
32
105000
2000
并不会有真正的效果。
01:59
And thirdly, as I said, most of the animations are very repetitive because of that.
33
107000
4000
第三点，如我所说，绝大部分的动画重复性都很大，也是因为都是预存的。
02:03
Now, the only way to get around that
34
111000
2000
唯一的解决办法
02:05
is to actually simulate the human body
35
113000
2000
就是实实在在地模拟人体的反应，
02:07
and to simulate that bit of the nervous system of the brain that controls that body.
36
115000
5000
模拟人的神经系统是怎样控制身体的运动的。
02:12
And maybe, if I could have you for a quick demonstration
37
120000
3000
这样吧，如果我能借你做个演示
02:15
to show what the difference is --
38
123000
2000
来展示（真人的反应）的不同 —
02:17
because, I mean, it's very, very trivial.
39
125000
4000
因为，要知道，有的差异是很微妙的。
02:21
If I push Chris a bit, like this, for example, he'll react to it.
40
129000
3000
好比我轻轻推克里斯一下，就像这样，他就会给出反应。
02:24
If I push him from a different angle, he'll react to it differently,
41
132000
3000
如果我从不同的角度推他，他的身体反应会不同。
02:27
and that's because he has a physical body,
42
135000
2000
那是因为他有一个真正的身体，
02:29
and because he has the motor skills to control that body.
43
137000
3000
他有那些运动技能，能够控制他的身体。
02:32
It's a very trivial thing.
44
140000
2000
这是非常微妙的。
02:34
It's not something you get in computer games at the moment, at all.
45
142000
2000
这不是你能从现今的电脑游戏中得到的反应。
02:36
Thank you very much. Chris Anderson: That's it?
46
144000
2000
谢了克里斯。（克里斯·安德森 问：“这样就行了？”）
02:38
Torsten Reil: That's it, yes.
47
146000
2000
是的，就这样。
02:40
So, that's what we're trying to simulate --
48
148000
1000
这就是我们试图模拟的 —
02:41
not Chris specifically, I should say, but humans in general.
49
149000
4000
不是单单模拟 Chris 这个人，而是，所有的真人身体的反应。
02:46
Now, we started working on this a while ago at Oxford University,
50
154000
5000
在牛津大学我们已经开始一段时间的研究了，
02:51
and we tried to start very simply.
51
159000
2000
我们一开始做得很简单，
02:53
What we tried to do was teach a stick figure how to walk.
52
161000
3000
仅仅是训练这个火柴人走路。
02:56
That stick figure is physically stimulated. You can see it here on the screen.
53
164000
3000
这个火柴人的身体是能接受信号刺激的。你在屏幕上能看见。
02:59
So, it's subject to gravity, has joints, etc.
54
167000
3000
所以它能对重力有反应，它也有关节，等等。
03:02
If you just run the simulation, it will just collapse, like this.
55
170000
3000
如果你开始模拟程序，它会跌倒，就像这样。
03:05
The tricky bit is now to put an AI controller in it
56
173000
4000
现在难就难在怎样把人工智能控制器放进去，
03:09
that actually makes it work.
57
177000
2000
让它顺利工作。
03:11
And for that, we use the neural network, which we based on
58
179000
3000
我们用的是模拟的神经系统，
03:14
that part of the nervous system that we have in our spine
59
182000
2000
它是根据真的人体的脊髓中
03:16
that controls walking in humans.
60
184000
2000
控制走路的那部分神经系统设计的。
03:18
It's called the central pattern generator.
61
186000
2000
名字叫中枢模式发生器。
03:20
So, we simulated that as well, and then the really tricky bit
62
188000
3000
我们模拟了这个神经系统之后，真正的难点
03:23
is to teach that network how to walk.
63
191000
2000
就是教会这个神经系统如何控制身体来走路。
03:25
For that we used artificial evolution -- genetic algorithms.
64
193000
4000
为了解决这个问题我们使用的是人工进化系统 — 基因模拟算法。
03:29
We heard about those already yesterday,
65
197000
2000
我们昨天也听说过了。
03:31
and I suppose that most of you are familiar with that already.
66
199000
3000
这里我就当你们大部分人都知道那是怎么回事。
03:34
But, just briefly, the concept is that
67
202000
2000
但是，简短地说，这个概念是
03:36
you create a large number of different individuals --
68
204000
3000
你首先制造出一大批个体来，各个不同，
03:39
neural networks, in this case --
69
207000
2000
这里我们做的是一大批神经系统。
03:41
all of which are random at the beginning.
70
209000
2000
开始时，每个的工作模式都是随机的。
03:43
You hook these up -- in this case, to the virtual muscles
71
211000
2000
你把这些接到这个火柴人上 —这里我们把这些神经系统
03:45
of that two-legged creature here --
72
213000
3000
接到这些两腿人的模拟肌肉系统上 —
03:48
and hope that it does something interesting.
73
216000
3000
然后就等着它们随机工作。
03:51
At the beginning, they're all going to be very boring.
74
219000
2000
一开始火柴人的反应都挺没劲的。
03:53
Most of them won't move at all,
75
221000
2000
绝大部分都不动，
03:55
but some of them might make a tiny step.
76
223000
2000
有些动起来了，也仅仅是迈一小步。
03:57
Those are then selected by the algorithm,
77
225000
2000
这些让火柴人动起来的神经系统会被挑出来，
03:59
reproduced with mutation and recombinations to introduce sex as well.
78
227000
4000
用来生产下一代，其中加入了基因变异和基因重组，也加入了性别。
04:03
And you repeat that process over and over again,
79
231000
2000
然后你不断重复这些步骤，
04:05
until you have something that walks --
80
233000
2000
直到你得到真的能够让火柴人走路的神经系统 —
04:07
in this case, in a straight line, like this.
81
235000
2000
就像这里这个能走一条直线的。
04:09
So that was the idea behind this.
82
237000
2000
这就是这个实验想做的。
04:11
When we started this, I set up the simulation one evening.
83
239000
3000
当我们开始做时，我在一天晚上启动了一个实验，
04:14
It took about three to four hours to run the simulation.
84
242000
3000
整个模拟试验一般要三四个小时。
04:17
I got up the next morning, went to the computer and looked at the results,
85
245000
4000
第二天早上起来我跑去看结果，
04:21
and was hoping for something that walked in a straight line,
86
249000
3000
满怀希望能得到一个走直线的动画人，
04:24
like I've just demonstrated,
87
252000
2000
就像我给你们看的那个，
04:26
and this is what I got instead.
88
254000
2000
结果我看到的是这个。
04:28
(Laughter)
89
256000
10000
（笑声）
04:38
So, it was back to the drawing board for us.
90
266000
3000
我们又得从头来过。
04:42
We did get it to work eventually,
91
270000
3000
最后我们当然做成了，
04:45
after tweaking a bit here and there.
92
273000
2000
改改这里改改那里。
04:47
And this is an example of a successful evolutionary run.
93
275000
3000
这里是个进化成功的例子。
04:50
So, what you'll see in a moment is a very simple biped
94
278000
3000
你马上将要看到的是一个非常简单的两足动物
04:53
that's learning how to walk using artificial evolution.
95
281000
3000
用人工进化的方法学习如何走路。
04:56
At the beginning, it can't walk at all,
96
284000
2000
一开始它一点也走不了，
04:58
but it will get better and better over time.
97
286000
2000
但是走的越来越好。
05:02
So, this is the one that can't walk at all.
98
290000
3000
这里是那个一点也不会走路的。
05:05
(Laughter)
99
293000
6000
（笑声）
05:11
Now, after five generations of applying evolutionary process,
100
299000
3000
现在，当运用了人工进化，五代之后
05:14
the genetic algorithm is getting a tiny bit better.
101
302000
3000
基因算法变得好些了。
05:17
(Laughter)
102
305000
8000
（笑声）
05:25
Generation 10 and it'll take a few steps more --
103
313000
2000
第十代，这个两足动物能多走几步了。
05:31
still not quite there.
104
319000
2000
但是还是不太行。
05:34
But now, after generation 20, it actually walks in a straight line without falling over.
105
322000
5000
现在，二十代后，它能真的走出一条直线来，也不会跌到。
05:40
That was the real breakthrough for us.
106
328000
3000
这对我们来说是个真正的突破。
05:43
It was, academically, quite a challenging project,
107
331000
3000
这是个学术上非常有挑战性的项目，
05:46
and once we had reached that stage, we were quite confident
108
334000
3000
而且当我们能做到这一步的时候，我们就有信心
05:49
that we could try and do other things as well with this approach --
109
337000
3000
能够挑战其他的目标，比如用这个方法 —
05:52
actually simulating the body
110
340000
2000
真的来模拟人体，
05:54
and simulating that part of the nervous system that controls it.
111
342000
3000
模拟用来控制身体运动的那部分神经。
05:57
Now, at this stage, it also became clear that this could be very exciting
112
345000
3000
并且那个阶段我们确信这个技术将会
06:00
for things like computer games or online worlds.
113
348000
3000
对电脑游戏或者网络世界有很大意义。
06:03
What you see here is the character standing there,
114
351000
2000
现在你看到的是一个模拟人物，站在那里，
06:05
and there's an obstacle that we put in its way.
115
353000
2000
前面我们放了一个障碍物。
06:07
And what you see is, it's going to fall over the obstacle.
116
355000
5000
你将会看到，它会被障碍物绊倒。
06:12
Now, the interesting bit is, if I move the obstacle a tiny bit to the right,
117
360000
3000
现在，有趣的是，如果我把障碍物向右边挪一点，
06:15
which is what I'm doing now, here,
118
363000
2000
就像这样，
06:17
it will fall over it in a completely different way.
119
365000
4000
它会以一种非常不同的方式跌倒。
06:24
And again, if you move the obstacle a tiny bit, it'll again fall differently.
120
372000
5000
再来一次，如果你把障碍物再挪一点，它跌到的方式又会不同。
06:29
(Laughter)
121
377000
2000
（笑声）
06:31
Now, what you see, by the way, at the top there,
122
379000
2000
现在，你看到的是，喔，顺便一提，在屏幕的上面，
06:33
are some of the neural activations being fed into the virtual muscles.
123
381000
3000
是我们传入肌肉的一些神经的活动，
06:36
Okay. That's the video. Thanks.
124
384000
2000
好的，这里就是这个片段。谢谢。
06:38
Now, this might look kind of trivial, but it's actually very important
125
386000
3000
好的，这看起来微不足道但是其是非常重要，
06:41
because this is not something you get at the moment
126
389000
2000
因为这是现今你在任何的
06:43
in any interactive or any virtual worlds.
127
391000
2000
虚拟世界或者互动游戏中能够看到的。
06:48
Now, at this stage, we decided to start a company and move this further,
128
396000
3000
现在，从这个阶段，我们决定成立一个公司，开始进一步的研究，
06:51
because obviously this was just a very simple, blocky biped.
129
399000
3000
因为很明显这只是个非常简单的块状两足动物，
06:54
What we really wanted was a full human body.
130
402000
2000
我们真的想做的是一个完整的模拟人体。
06:56
So we started the company.
131
404000
1000
这样我们成立了公司。
06:57
We hired a team of physicists, software engineers and biologists
132
405000
5000
我们吸收了一些物理学家，软件工程师和生物学家
07:02
to work on this, and the first thing we had to work on
133
410000
3000
一起工作，我们做的第一件事是
07:05
was to create the human body, basically.
134
413000
4000
做出一个模拟人体。
07:09
It's got to be relatively fast, so you can run it on a normal machine,
135
417000
3000
这个模拟人需要很轻巧，这样你才能在一般的电脑上运作。
07:12
but it's got to be accurate enough, so it looks good enough, basically.
136
420000
3000
但是也需要非常地准确，才会好看。
07:15
So we put quite a bit of biomechanical knowledge into this thing,
137
423000
3000
所以我们夹了不少生化科技进去，
07:18
and tried to make it as realistic as possible.
138
426000
4000
使它看起来像是真的。
07:22
What you see here on the screen right now
139
430000
2000
这里你在屏幕上看到的，
07:24
is a very simple visualization of that body.
140
432000
2000
是这个模拟人体的一个简单版，
07:26
I should add that it's very simple to add things like hair, clothes, etc.,
141
434000
4000
值得一提的是给它加上头发，衣服之类的是非常简单的。
07:30
but what we've done here is use a very simple visualization,
142
438000
3000
但是我们决定做得很简洁
07:33
so you can concentrate on the movement.
143
441000
2000
这样大家的注意力才会集中在它的动作上。
07:35
Now, what I'm going to do right now, in a moment,
144
443000
3000
现在，我想做的是，
07:38
is just push this character a tiny bit and we'll see what happens.
145
446000
3000
推这个模拟人一下下，看看会发生什么。
07:46
Nothing really interesting, basically.
146
454000
2000
没什么有趣的。
07:48
It falls over, but it falls over like a rag doll, basically.
147
456000
3000
它跌到了，像个假娃娃。
07:51
The reason for that is that there's no intelligence in it.
148
459000
3000
这是因为我们没有放人工智能进去。
07:54
It becomes interesting when you put artificial intelligence into it.
149
462000
4000
当我们放人工智能进去就有趣多了。
07:58
So, this character now has motor skills in the upper body --
150
466000
4000
现在这个模拟人的上半身有运动技能。
08:02
nothing in the legs yet, in this particular one.
151
470000
2000
下半身没有，在这个人体里。
08:04
But what it will do -- I'm going to push it again.
152
472000
3000
但是马上你会看到 — 我现在在推它一下，
08:07
It will realize autonomously that it's being pushed.
153
475000
2000
它有自主意识被推了，
08:09
It's going to stick out its hands.
154
477000
2000
它会张开双臂，
08:11
It's going to turn around into the fall, and try and catch the fall.
155
479000
3000
向跌倒的方向转身，试图撑住。
08:20
So that's what you see here.
156
488000
2000
这就是你在这里看到的。
08:22
Now, it gets really interesting
157
490000
2000
如果下半身也加上人工智能，
08:24
if you then add the AI for the lower part of the body as well.
158
492000
4000
就更有趣了。
08:28
So here, we've got the same character.
159
496000
2000
这里看到的是同一个模拟人，
08:30
I'm going to push it a bit harder now,
160
498000
2000
我现在推得更狠一些，
08:32
harder than I just pushed Chris.
161
500000
2000
比刚才我推克里斯更用力些。
08:34
But what you'll see is -- it's going to receive a push now from the left.
162
502000
4000
你将看到我从左边推它。
08:41
What you see is it takes steps backwards,
163
509000
2000
你将会看到它向后退 —
08:43
it tries to counter-balance,
164
511000
2000
试图平衡自己，
08:45
it tries to look at the place where it thinks it's going to land.
165
513000
4000
试图向下看。
08:49
I'll show you this again.
166
517000
2000
我马上播放给你看。
08:51
And then, finally hits the floor.
167
519000
3000
最后它倒到地上。
08:55
Now, this becomes really exciting
168
523000
3000
现在当你从不同角度推它的时候，
08:58
when you push that character in different directions, again, just as I've done.
169
526000
5000
它的反应就有趣多了，就像我刚做的。
09:03
That's something that you cannot do right now.
170
531000
4000
这是你现在得不到的。
09:07
At the moment, you only have empty computer graphics in games.
171
535000
3000
现在你从游戏里只能看到空洞的动画图像。
09:10
What this is now is a real simulation. That's what I want to show you now.
172
538000
3000
这里我给你看的是真的模拟人生。
09:13
So, here's the same character with the same behavior I've just shown you,
173
541000
3000
这里是同一个模拟人，能作相同的反应。
09:16
but now I'm just going to push it from different directions.
174
544000
2000
现在我从不同角度推它，
09:18
First, starting with a push from the right.
175
546000
2000
先从右边退，
09:23
This is all slow motion, by the way, so we can see what's going on.
176
551000
3000
这是慢动作这样我们能看得更清楚。
09:26
Now, the angle will have changed a tiny bit,
177
554000
3000
现在我换个角度，
09:29
so you can see that the reaction is different.
178
557000
4000
你看一看到它的反应是不同的。
09:33
Again, a push, now this time from the front.
179
561000
3000
再来一次，推一下，从前面
09:37
And you see it falls differently.
180
565000
2000
你看它跌到的方式又变了。
09:39
And now from the left --
181
567000
2000
现在从左边。
09:43
and it falls differently.
182
571000
2000
又不一样。
09:45
That was really exciting for us to see that.
183
573000
2000
我们看到这个结果很激动，
09:47
That was the first time we've seen that.
184
575000
2000
这是第一次我们看到成果。
09:49
This is the first time the public sees this as well,
185
577000
2000
今天是第一次向公众发表。
09:51
because we have been in stealth mode.
186
579000
2000
因为我们还在保密阶段的左，
09:53
I haven't shown this to anybody yet.
187
581000
2000
我还没给人看过。
09:55
Now, just a fun thing:
188
583000
2000
现在，只是好玩，
09:57
what happens if you put that character --
189
585000
2000
看看如果我们把它放在不同情境会发生什么 —
09:59
this is now a wooden version of it, but it's got the same AI in it --
190
587000
2000
这是个木头人的版本，放了人工智能进去 —
10:01
but if you put that character on a slippery surface, like ice.
191
589000
2000
我们把它放在光滑的表面上，比如冰面。
10:03
We just did that for a laugh, just to see what happens.
192
591000
3000
我们这么做只是为了看看会发生什么滑稽的事。
10:06
(Laughter)
193
594000
1000
（笑声）
10:07
And this is what happens.
194
595000
2000
结果是这样。
10:09
(Laughter)
195
597000
3000
（笑声）
10:12
(Applause)
196
600000
3000
（鼓掌）
10:15
It's nothing we had to do about this.
197
603000
2000
我们没有人为的加东西进去。
10:17
We just took this character that I just talked about,
198
605000
2000
我们只是用了类似的模拟人，
10:19
put it on a slippery surface, and this is what you get out of it.
199
607000
3000
把它放在光滑的表面上，结果就是这样。
10:22
And that's a really fascinating thing about this approach.
200
610000
3000
这正是这个研究的过人之处，
10:26
Now, when we went to film studios and games developers
201
614000
3000
当我们把这个结果给电影工作者和游戏的设计者看的时候，
10:29
and showed them that technology, we got a very good response.
202
617000
3000
向他们展示这项技术，我们得到了很好的反响。
10:32
And what they said was, the first thing they need immediately is virtual stuntmen.
203
620000
4000
他们最需要的是虚拟替身。
10:36
Because stunts are obviously very dangerous, they're very expensive,
204
624000
4000
因为替身工作非常危险，雇替身的花费也很大。
10:40
and there are a lot of stunt scenes that you cannot do, obviously,
205
628000
2000
很多使用替身的片断还不能用真人，
10:42
because you can't really allow the stuntman to be seriously hurt.
206
630000
3000
因为这些特技太危险了，真人会受伤。
10:45
So, they wanted to have a digital version of a stuntman
207
633000
3000
所以他们很需要一个数码版的替身，
10:48
and that's what we've been working on for the past few months.
208
636000
2000
这就是我们过去几个月里做的项目。
10:50
And that's our first product that we're going to release in a couple of weeks.
209
638000
5000
这是我们的第一个产品，几周后会上市。
10:55
So, here are just a few very simple scenes of the guy just being kicked.
210
643000
5000
这里是一些非常简单的情境，比如这个人刚被踢了一脚。
11:00
That's what people want. That's what we're giving them.
211
648000
2000
这是电影公司需要的，我们只是按他们的需要做。
11:02
(Laughter)
212
650000
7000
（笑声）
11:09
You can see, it's always reacting.
213
657000
2000
你可以看到，模拟人总会反应，
11:11
This is not a dead body. This is a body who basically, in this particular case,
214
659000
4000
这不像是没有生命的个体。这是一个能够感觉到施力的身体，
11:15
feels the force and tries to protect its head.
215
663000
2000
在这个特定情境下，能够自我保护。
11:17
Only, I think it's quite a big blow again.
216
665000
2000
我认为这个看起来很真实，
11:19
You feel kind of sorry for that thing,
217
667000
2000
所以人们开始觉得不忍。
11:21
and we've seen it so many times now that
218
669000
2000
我们这些人看得次数太多了，
11:23
we don't really care any more.
219
671000
2000
已经漠不关心了。
11:25
(Laughter)
220
673000
1000
（笑声）
11:26
There are much worse videos than this, by the way, which I have taken out, but ...
221
674000
4000
很多片断更糟糕，我都不能给你们看。
11:31
Now, here's another one.
222
679000
2000
这里是另一个片断，
11:33
What people wanted as a behavior was to have an explosion,
223
681000
4000
他们要求的是一个对爆炸情境的反应，
11:37
a strong force applied to the character,
224
685000
2000
好像有一个强大的力量施加到这个人物身上，
11:39
and have the character react to it in midair.
225
687000
2000
这个人物需要在空中作反应。
11:41
So that you don't have a character that looks limp,
226
689000
2000
你不会想要一个软绵绵无生气的模拟人，
11:43
but actually a character that you can use in an action film straight away,
227
691000
3000
你要的是一个能够在动作电影中用的，
11:46
that looks kind of alive in midair as well.
228
694000
2000
在空中的反应看起来是活的模拟人。
11:48
So this character is going to be hit by a force,
229
696000
2000
现在这个模拟人会被重重一击，
11:50
it's going to realize it's in the air,
230
698000
2000
它会意识到自己飞到空中，
11:52
and it's going to try and, well,
231
700000
3000
它会试图，
11:55
stick out its arm in the direction where it's landing.
232
703000
2000
向它跌落的方向伸出手臂。
11:59
That's one angle; here's another angle.
233
707000
3000
这里是从一个角度看，这里是从另一个角度看。
12:02
We now think that the realism we're achieving with this
234
710000
2000
我们现在认为我们可以成功地
12:04
is good enough to be used in films.
235
712000
2000
使用这些电脑替身。
12:06
And let's just have a look at a slightly different visualization.
236
714000
3000
然我们看看一些不同的情境，
12:09
This is something I just got last night
237
717000
2000
这里是我昨晚上拿到的结果，
12:11
from an animation studio in London, who are using our software
238
719000
3000
伦敦的一个动画公司用我们的软件做的，
12:14
and experimenting with it right now.
239
722000
2000
他们正在实验把这个做到游戏中去。
12:16
So this is exactly the same behavior that you saw,
240
724000
3000
这是同一个模拟人，
12:19
but in a slightly better rendered version.
241
727000
4000
但是稍微精美一些的版本。
12:23
So if you look at the character carefully,
242
731000
3000
如果你仔细地看，
12:26
you see there are lots of body movements going on,
243
734000
2000
你可以看到很多的肢体动作，
12:28
none of which you have to animate like in the old days.
244
736000
2000
这些动作不需要你画出来，
12:30
Animators had to actually animate them.
245
738000
2000
用老方法动画人需要真的把它们画出来，
12:32
This is all happening automatically in the simulation.
246
740000
2000
这里都是自然而然在模拟人身上发生的。
12:34
This is a slightly different angle,
247
742000
2000
这是另一个角度，
12:39
and again a slow motion version of this.
248
747000
2000
还是慢动作。
12:41
This is incredibly quick. This is happening in real time.
249
749000
4000
这一切都发生得很快，实时发生的。
12:45
You can run this simulation in real time, in front of your eyes,
250
753000
2000
你可以实时跑这个程序，这些动作就会发生在你眼前。
12:47
change it, if you want to, and you get the animation straight out of it.
251
755000
3000
你可以改变这个程序，然后得到不同的动作。
12:50
At the moment, doing something like this by hand
252
758000
2000
目前，手工绘制出这些动作，
12:52
would take you probably a couple of days.
253
760000
2000
恐怕要花好几天。
12:55
This is another behavior they requested.
254
763000
3000
这是另一个他们需要的动作。
12:58
I'm not quite sure why, but we've done it anyway.
255
766000
2000
我不知道为什么，但是我们按需作活。
13:00
It's a very simple behavior that shows you the power of this approach.
256
768000
2000
这是个非常简单的特技但是你可以看到这个技术的潜力。
13:02
In this case, the character's hands
257
770000
2000
这里模拟人的手
13:04
are fixed to a particular point in space,
258
772000
2000
被固定在空中一个特定的点上，
13:06
and all we've told the character to do is to struggle.
259
774000
3000
然后我们让这个模拟人挣扎。
13:09
And it looks organic. It looks realistic.
260
777000
3000
它的动作看起来很自然，很真实。
13:12
You feel kind of sorry for the guy.
261
780000
2000
你甚至会为“他”难过。
13:14
It's even worse -- and that is another video I just got last night --
262
782000
3000
这个更残忍 — 这是另一个昨晚我才拿到的片断 —
13:17
if you render that a bit more realistically.
263
785000
2000
如果你把人物作得更真实一点。
13:23
Now, I'm showing this to you just to show you
264
791000
2000
现在我给你们看这个片断，
13:25
how organic it actually can feel, how realistic it can look.
265
793000
2000
你可以感受到这些动作有多自然，看起来多真实。
13:27
And this is all a physical simulation of the body,
266
795000
3000
而这全是因为我们有这个模拟的身体，
13:30
using AI to drive virtual muscles in that body.
267
798000
3000
我们用了人工智能来驾驭这些肌肉。
13:35
Now, one thing which we did for a laugh was
268
803000
3000
现在，我想给你们看一个只是做了好玩的片断
13:38
to create a slightly more complex stunt scene,
269
806000
2000
我们只是想做一个更复杂的特技，
13:40
and one of the most famous stunts is the one where James Bond
270
808000
3000
而一个非常著名的特技动作是当零零七
13:43
jumps off a dam in Switzerland and then is caught by a bungee.
271
811000
4000
在瑞士跳入一个大水库，然后被蹦极绳子救了。
13:48
Got a very short clip here.
272
816000
2000
这里是这个真的电影片断。
13:54
Yes, you can just about see it here.
273
822000
2000
你可以看到。
13:56
In this case, they were using a real stunt man. It was a very dangerous stunt.
274
824000
3000
这里他们请了一个真的特技演员，但是这是个非常危险的特技。
13:59
It was just voted, I think in the Sunday Times, as one of the most impressive stunts.
275
827000
3000
我想在“周日报导”上这个特技刚被评为最惊人的特技之一。
14:02
Now, we've just tried and -- looked at our character and asked ourselves,
276
830000
3000
这里，我们试着用我们的模拟人，然后问自己
14:05
"Can we do that ourselves as well?"
277
833000
2000
“我们的模拟特技人能做到么？”
14:07
Can we use the physical simulation of the character,
278
835000
2000
我们能不能用这个模拟的人物，
14:09
use artificial intelligence,
279
837000
2000
用我们的人工智能方法，
14:11
put that artificial intelligence into the character,
280
839000
2000
把人工智能放在模拟人身体里，
14:13
drive virtual muscles, simulate the way he jumps off the dam,
281
841000
4000
让控制肌肉，模拟真人跳入水库，
14:17
and then skydive afterwards,
282
845000
2000
跳进去，
14:19
and have him caught by a bungee afterwards?
283
847000
2000
然后被蹦极绳拉回来？
14:21
We did that. It took about altogether just two hours,
284
849000
3000
我们真的做到了。仅仅两个小时，
14:24
pretty much, to create the simulation.
285
852000
2000
就作出了这个模拟情境。
14:26
And that's what it looks like, here.
286
854000
2000
这里是它看起来的样子。
14:37
Now, this could do with a bit more work. It's still very early stages,
287
865000
3000
我们可以做得更好一点，这个研究还在初级阶段，
14:40
and we pretty much just did this for a laugh,
288
868000
2000
我们仅仅是作了好玩的。
14:42
just to see what we'd get out of it.
289
870000
2000
只是看看我们能做出什么来。
14:44
But what we found over the past few months
290
872000
2000
我们过去几个月的工作证明了
14:46
is that this approach -- that we're pretty much standard upon --
291
874000
3000
我们的技术是非常非常
14:49
is incredibly powerful.
292
877000
2000
有潜力的。
14:51
We are ourselves surprised what you actually get out of the simulations.
293
879000
4000
对于我们能够从这些模拟人身上得到的结果，我们自己也惊讶了。
14:55
There's very often very surprising behavior that you didn't predict before.
294
883000
4000
我们常常得到非常出乎意料的结果。
14:59
There's so many things we can do with this right now.
295
887000
2000
我们的技术有很多应用。
15:01
The first thing, as I said, is going to be virtual stuntmen.
296
889000
3000
第一个，像我说的，是数码特技替身，
15:04
Several studios are using this software now to produce virtual stuntmen,
297
892000
4000
一些电影工作室已经在用这个软件做数码替身了。
15:08
and they're going to hit the screen quite soon, actually,
298
896000
2000
这些特技镜头会在很短的时间内上映，
15:10
for some major productions.
299
898000
2000
很多大的制作都用它们。
15:12
The second thing is video games.
300
900000
3000
第二个应用就是电脑游戏。
15:15
With this technology, video games will look different and they will feel very different.
301
903000
4000
用这个技术电脑游戏看起来会很不同，感觉上也会很不同。
15:19
For the first time, you'll have actors that really feel very interactive,
302
907000
3000
历史上第一次你会看到演员的非常真实的互动，
15:22
that have real bodies that really react.
303
910000
2000
真的身体，真的反应。
15:24
I think that's going to be incredibly exciting.
304
912000
3000
我认为这非常令人激动。
15:27
Probably starting with sports games,
305
915000
2000
最先我们会用在运动性的游戏上，
15:29
which are going to become much more interactive.
306
917000
2000
比起平常的游戏它们更需要互动性。
15:31
But I particularly am really excited
307
919000
1000
但是我本人最感到激动的应用
15:32
about using this technology in online worlds,
308
920000
3000
是将这个技术用在网络世界里
15:35
like there, for example, that Tom Melcher has shown us.
309
923000
3000
比如像汤姆·梅尔彻给我们展示的那样。
15:38
The degree of interactivity you're going to get
310
926000
2000
我们能够得到的互动性
15:40
is totally different, I think, from what you're getting right now.
311
928000
3000
是非常不同的，我认为，和你能从老方法中得到的相比。
15:44
A third thing we are looking at and very interested in is simulation.
312
932000
4000
这是第三个我们非常感兴趣的应用。
15:49
We've been approached by several simulation companies,
313
937000
2000
已经有一些公司联系我们了，
15:51
but one project we're particularly excited about, which we're starting next month,
314
939000
3000
但是有一个特别的项目我们感到特别的激动，我们下个月会开始做，
15:54
is to use our technology -- and in particular, the walking technology --
315
942000
4000
是将我们的技术，特别是模拟走路这方面，
15:58
to help aid surgeons who work on children with cerebral palsy,
316
946000
4000
用在帮助那些治疗儿童脑瘫的医生
16:02
to predict the outcome of operations on these children.
317
950000
3000
来预计手术的愈后效果。
16:05
As you probably know,
318
953000
2000
如你们所知，
16:07
it's very difficult to predict what the outcome of an operation is
319
955000
3000
通常脑瘫病人手术后的行走能力的结果
16:10
if you try and correct the gait.
320
958000
2000
是非常难以预料的，
16:12
The classic quote is, I think, it's unpredictable at best,
321
960000
3000
经典说法是，我认为，人们目前认为
16:15
is what people think right now, is the outcome.
322
963000
3000
最难预料的，就是愈后结果。
16:18
Now, what we want to do with our software is allow our surgeons to have a tool.
323
966000
4000
现在，我们想做的是用我们的软件，帮助医生预测。
16:22
We're going to simulate the gait of a particular child
324
970000
3000
我们想模拟每个孩子的步态，
16:25
and the surgeon can then work on that simulation
325
973000
3000
医生能够用这个模拟人物，
16:28
and try out different ways to improve that gait,
326
976000
2000
试不同的手术方案，来帮助调整步态，
16:30
before he actually commits to an actual surgery.
327
978000
3000
在他们真正做手术之前。
16:33
That's one project we're particularly excited about,
328
981000
2000
这个项目我们感到特别激动，
16:35
and that's going to start next month.
329
983000
2000
下个月会启动。
16:39
Just finally, this is only just the beginning.
330
987000
3000
最后，我想说这只是个起步，
16:42
We can only do several behaviors right now.
331
990000
2000
我们现在只能做一些特定的动作，
16:44
The AI isn't good enough to simulate a full human body.
332
992000
3000
我们用的人工智能还不能模拟整个人体的神经系统。
16:47
The body yes, but not all the motor skills that we have.
333
995000
3000
我们能模拟整个身体，但是不能模拟所有的动作功能。
16:50
And, I think, we're only there if we can have something like ballet dancing.
334
998000
3000
我希望能做出像芭蕾舞这类的动作来，
16:53
Right now, we don't have that
335
1001000
2000
现在还做不到。
16:55
but I'm very sure that we will be able to do that at some stage.
336
1003000
2000
但是我确信我们将来可以。
16:57
We do have one unintentional dancer actually,
337
1005000
3000
我们碰巧做出了一个舞者，
17:00
the last thing I'm going to show you.
338
1008000
2000
这是我想给你们看的最后一个人物。
17:02
This was an AI contour that was produced and evolved --
339
1010000
3000
这个人工智能我们做出来 —
17:05
half-evolved, I should say -- to produce balance, basically.
340
1013000
3000
部分上是 — 保持平衡的。
17:08
So, you kick the guy and the guy's supposed to counter-balance.
341
1016000
3000
你踢这个人一脚，他应该会平衡自己。
17:11
That's what we thought was going to come out of this.
342
1019000
3000
这也是我们原本预计的结果。
17:14
But this is what emerged out of it, in the end.
343
1022000
2000
结果这是我们最后看到的东西。
17:17
(Music)
344
1025000
10000
（音乐起）
17:27
Bizarrely, this thing doesn't have a head. I'm not quite sure why.
345
1035000
3000
看起来怪怪的，这个舞者没有头，我不知道为什么，
17:31
So, this was not something we actually put in there.
346
1039000
2000
但是我们通常不用头。
17:33
He just started to create that dance himself.
347
1041000
4000
他自顾自地开始跳舞了。
17:37
He's actually a better dancer than I am, I have to say.
348
1045000
3000
我得说，他比我还强些。
17:41
And what you see after a while --
349
1049000
2000
你将会看到的 —
17:43
I think he even goes into a climax right at the end.
350
1051000
2000
这个舞最后还来了一个高潮呢。
17:49
And I think -- there you go.
351
1057000
3000
我想，就是这里。
17:52
(Laughter)
352
1060000
2000
（笑声）
17:54
So, that all happened automatically. We didn't put that in there.
353
1062000
2000
这都是自动产生的，我们没有加这个舞进去。
17:56
That's just the simulation creating this itself, basically.
354
1064000
3000
这都是这个模拟人自己产生的。
17:59
So it's just --
355
1067000
2000
这样 —
18:01
(Applause)
356
1069000
1000
（掌声）
18:02
Thanks.
357
1070000
2000
谢谢。
18:05
Not quite John Travolta yet, but we're working on that as well,
358
1073000
3000
他不像约翰·特拉沃尔塔那样帅，但我们在向这个方向努力。
18:08
so thanks very much for your time.
359
1076000
2000
谢谢你们给我时间。
18:10
Thanks.
360
1078000
1000
谢谢
18:11
(Applause)
361
1079000
1000
（掌声）
18:12
CA: Incredible. That was really incredible.
362
1080000
2000
克里斯·安德森 ：“令人惊叹，这真是太令人惊叹了。”
18:14
TR: Thanks.
363
1082000
1000
托斯腾·雷尔：“谢谢。”
ABOUT THE SPEAKER
Torsten Reil - Animating neurobiologistBy coding computer simulations with biologically modeled nervous systems, Torsten Reil and his company NaturalMotion breathe life into the animated characters inhabiting the most eye-poppingly realistic games and movies around.
Why you should listen
From modeling the mayhem of equine combat in Lord of the Rings: Return of the King to animating Liberty City gun battles in Grand Theft Auto IV, Torsten Reil's achievements are all over the map these days. Software that he helped create (with NaturalMotion, the imaging company he co-founded) has revolutionized computer animation of human and animal avatars, giving rise to some of the most breathtakingly real sequences in the virtual world of video games and movies- and along the way given valuable insight into the way human beings move their bodies.
Reil was a neural researcher working on his Masters at Oxford, developing computer simulations of nervous systems based on genetic algorithms- programs that actually used natural selection to evolve their own means of locomotion. It didn't take long until he realized the commercial potential of these lifelike characters. In 2001 he capitalized on this lucrative adjunct to his research, and cofounded NaturalMotion. Since then the company has produced motion simulation programs like Euphoria and Morpheme, state of the art packages designed to drastically cut the time and expense of game development, and create animated worlds as real as the one outside your front door. Animation and special effects created with Endorphin (NaturalMotion's first animation toolkit) have lent explosive action to films such as Troy and Poseidon, and NaturalMotion's software is also being used by LucasArts in video games such as the hotly anticipated Indiana Jones.
But there are serious applications aside from the big screen and the XBox console: NaturalMotion has also worked under a grant from the British government to study the motion of a cerebral palsy patient, in hopes of finding therapies and surgeries that dovetail with the way her nervous system is functioning.
More profile about the speakerReil was a neural researcher working on his Masters at Oxford, developing computer simulations of nervous systems based on genetic algorithms- programs that actually used natural selection to evolve their own means of locomotion. It didn't take long until he realized the commercial potential of these lifelike characters. In 2001 he capitalized on this lucrative adjunct to his research, and cofounded NaturalMotion. Since then the company has produced motion simulation programs like Euphoria and Morpheme, state of the art packages designed to drastically cut the time and expense of game development, and create animated worlds as real as the one outside your front door. Animation and special effects created with Endorphin (NaturalMotion's first animation toolkit) have lent explosive action to films such as Troy and Poseidon, and NaturalMotion's software is also being used by LucasArts in video games such as the hotly anticipated Indiana Jones.
But there are serious applications aside from the big screen and the XBox console: NaturalMotion has also worked under a grant from the British government to study the motion of a cerebral palsy patient, in hopes of finding therapies and surgeries that dovetail with the way her nervous system is functioning.
Torsten Reil | Speaker | TED.com