Why Don’t We Have Better Robots Yet? | Ken Goldberg | TED

翻译人员: Yip Yan Yeung 校对人员: suya f.

I have a feeling most people in this room would like to have a robot at home.

我猜在座的大多数人 家里都有机器人。

It'd be nice to be able to do the chores and take care of things.

会做家务、 会处理事情真是太好了。

Where are these robots?

这些机器人在哪里？

What's taking so long?

为什么花了这么长的时间？

I mean, we have our tricorders,

我们有三轴飞行器， 还有卫星。

and we have satellites.

We have laser beams.

我们有激光束。

But where are the robots?

但是机器人在哪里？

(Laughter)

（笑声）

I mean, OK, wait, we do have some robots in our home,

好吧，等等，

我们家里确实有一些机器人，

but, not really doing anything that exciting, OK?

但是没在做那些酷炫的事，对吧？

(Laughter)

（笑声）

Now I've been doing research at UC Berkeley for 30 years

我已经在加州大学伯克利分校

和我的学生一起研究了 30 年的机器人，

with my students on robots,

and in the next 10 minutes,

在接下来的 10 分钟里，

I'm going to try to explain the gap between fiction and reality.

我将尝试解释 虚构与现实之间的差距。

Now we’ve seen images like this, right?

我们已经看到了 这样的影像，对吧？

These are real robots.

这些是真实的机器人。

They're pretty amazing.

它们太神奇了。

But those of us who work in the field,

但那些在该领域工作的人，

well, the reality is more like this.

好吧，现实更像这样。

(Laughter)

（笑声）

That's 99 out of 100 times, that's what happens.

100 次中的 99 次是这样， 事实就是这样。

And in the field, there's something that explains this

在这个领域，

that we call Moravec's paradox.

“莫拉维克悖论” 可以解释这个现象。

And that is, what's easy for robots,

对机器人来说很容易的任务，

like being able to pick up a large object,

比如捡起一个大型物体，

large, heavy object,

大而重的物体，

is hard for humans.

对人类来说是很难的。

But what's easy for humans,

但是对人类很容易的任务，

like being able to pick up some blocks and stack them,

比如捡起一些方块， 把它们叠起来，

well, it turns out that is very hard for robots.

对机器人来说非常困难。

And this is a persistent problem.

这是一个持续存在的问题。

So the ability to grasp arbitrary objects is a grand challenge for my field.

抓住任意物体的能力 在我这个领域是一项艰巨的挑战。

Now by the way, I was a very klutzy kid.

顺便说一句， 我是一个笨手笨脚的孩子。

(Laughter)

（笑声）

I would drop things.

我会拿不住东西。

Any time someone would throw me a ball, I would drop it.

只要有人丢给我一个球， 我都接不住。

I was the last kid to get picked on a basketball team.

我是最后一个 被选入篮球队的孩子。

I'm still pretty klutzy, actually,

我其实还是很笨拙，

but I have spent my entire career studying how to make robots less clumsy.

但我整个职业生涯都在研究 如何让机器人不那么笨拙。

Now let's start with the hardware.

让我们从硬件开始。

So the hands.

双手。

Now this is a robot hand, a particular type of hand.

这是一只机器人的手， 一种特殊的手。

It's a lot like our hand.

这很像我们的手。

And it has a lot of motors, a lot of tendons

你可以看到它有很多马达、 很多肌腱和电缆。

and cables as you can see.

So it's unfortunately not very reliable.

不幸的是，它不是很可靠。

It's also very heavy and very expensive.

它还很重，而且非常昂贵。

So I'm in favor of very simple hands, like this.

我更喜欢更简单的手，就像这个。

So this has just two fingers.

它只有两根手指。

It's known as a parallel jaw gripper.

它被称为“平行夹爪”。

So it's very simple.

它很简单。

It's lightweight and reliable and it's very inexpensive.

它轻巧可靠，而且非常便宜。

And if you're doubting that simple hands can be effective,

如果你在怀疑 这简单的手有多有效，

look at this video where you can see that two very simple grippers,

看看这个视频，你可以看到 两个非常简单的夹爪，

these are being operated, by the way,

顺便提一句，它们是由人 像控制木偶那样操控的。

by humans who are controlling the grippers like a puppet.

But very simple grippers are capable of doing very complex things.

但是非常简单的夹爪 能够做非常复杂的事情。

Now actually in industry,

其实在行业中，

there’s even a simpler robot gripper, and that’s the suction cup.

还有一种更简单的机械夹爪， 那就是吸盘。

And that only makes a single point of contact.

只需要一个接触点。

So again, simplicity is very helpful in our field.

同样，“简单” 在我们的领域中非常有用。

Now let's talk about the software.

我们来聊聊软件。

This is where it gets really, really difficult

这就是非常困难的地方，

because of a fundamental issue, which is uncertainty.

源于一个根本性的问题， 那就是“不确定性”。

There's uncertainty in the control.

控制中存在不确定性。

There’s uncertainty in the perception.

感知中存在不确定性。

And there’s uncertainty in the physics.

物理上存在不确定性。

Now what do I mean by the control?

我说的“控制”是什么意思？

Well if you look at a robot’s gripper trying to do something,

如果你看看机器人的抓手尝试做某事，

there's a lot of uncertainty in the cables and the mechanisms

就会发现电缆和机械装置 存在着很大的不确定性，

that cause very small errors.

而这些不确定性会导致非常小的错误。

And these can accumulate and make it very difficult to manipulate things.

而且它们会累积起来， 使操纵物体变得非常困难。

Now in terms of the sensors, yes,

就传感器而言，没错，

robots have very high-resolution cameras just like we do,

机器人像我们一样， 拥有非常高分辨率的摄像头，

and that allows them to take images of scenes in traffic

使它们能够拍摄交通、

or in a retirement center,

退休中心、

or in a warehouse or in an operating room.

仓库或手术室中的场景图像。

But these don't give you the three-dimensional structure

但是这些并不能给你 现况的三维结构。

of what's going on.

So recently, there was a new development called LIDAR,

最近，有一项名为 “激光雷达”（LIDAR）的新技术，

and this is a new class of cameras that use light beams to build up

是一种新型摄像机，使用光束

a three-dimensional model of the environment.

建立环境的三维模型。

And these are fairly effective.

它们相当有效。

They really were a breakthrough in our field, but they're not perfect.

它们确实是我们领域的一项突破， 但它们并不完美。

So if the objects have anything that's shiny or transparent,

如果物体有任何闪亮或透明的东西，

well, then the light acts in unpredictable ways,

那么，光线就会 出现无法预测的情况，

and it ends up with noise and holes in the images.

最终图像中会产生噪点和孔洞。

So these aren't really the silver bullet.

因此，它们并不是真正的万灵药。

And there’s one other form of sensor out there now called a “tactile sensor.”

现在还有另一种形式的传感器， 叫做“触觉传感器”。

And these are very interesting.

它们非常有趣。

They use cameras to actually image the surfaces

它们用摄像头对表面摄影，

as a robot would make contact,

就像机器人接触表面那样，

but these are still in their infancy.

但它们仍处于起步阶段。

Now the last issue is the physics.

最后一个问题是物理。

And let me illustrate for you by showing you,

我来展示一下，

we take a bottle on a table

我们把一个瓶子放在桌上，

and we just push it,

推它，

and the robot's pushing it in exactly the same way each time.

机器人每次的推动方式完全一样。

But you can see that the bottle ends up in a very different place each time.

但是你可以看到，瓶子每次都会 出现在一个截然不同的地方。

And why is that?

为什么？

Well it’s because it depends on the microscopic surface topography

这是因为这取决于在瓶子滑动时，

它下方出现的微观表面地形。

underneath the bottle as it slid.

For example, if you put a grain of sand under there,

比如，如果你在下面放了一粒沙子，

it would react very differently than if there weren't a grain of sand.

相较没有那粒沙子， 反应会截然不同。

And we can't see if there's a grain of sand because it's under the bottle.

我们看不出是否有沙子， 因为它在瓶子下面。

It turns out that we can predict the motion of an asteroid

事实证明，我们可以预测 一百万英里以外的小行星的运动，

a million miles away,

far better than we can predict the motion of an object

效果远比我们预测一个物体

as it's being grasped by a robot.

被机器人抓住时的运动要好得多。

Now let me give you an example.

我来举个例子。

Put yourself here into the position of being a robot.

把自己置于机器人的角色。

You're trying to clear the table

你想清理桌子，

and your sensors are noisy and imprecise.

但你的传感器噪声很大， 而且不精确。

Your actuators, your cables and motors are uncertain,

你的驱动器、电缆和电机不稳定，

so you can't fully control your own gripper.

所以你无法完全控制自己的夹爪。

And there's uncertainty in the physics,

物理上也存在不确定性，

so you really don't know what's going to happen.

所以你真的不知道会发生什么。

So it's not surprising that robots are still very clumsy.

因此，机器人仍然非常笨拙 也就不足为奇了。

Now there's one sweet spot for robots, and that has to do with e-commerce.

机器人有一个最佳用处， 与电子商务有关。

And this has been growing, it's a huge trend.

这种趋势一直在增长， 是一个巨大的趋势。

And during the pandemic, it really jumped up.

在疫情期间，它突飞猛进。

I think most of us can relate to that.

我认为我们很多人都有所体会。

We started ordering things like never before,

我们比以往都更常订购商品，

and this trend is continuing.

而且这种趋势仍在继续。

And the challenge is to meet the demand,

挑战在于满足需求，

we have to be able to get all these packages delivered in a timely manner.

我们必须及时交付所有包裹。

And the challenge is that every package is different,

挑战在于每个包裹都不一样，

every order is different.

每个订单都不一样。

So you might order some some nail polish and an electric screwdriver.

你可以订购一些指甲油 和一把电动螺丝刀。

And those two objects are going to be

而这两个商品将位于

somewhere inside one of these giant warehouses.

一个巨型仓库内的某个位置。

And what needs to be done is someone has to go in,

得有人进去，

find the nail polish and then go and find the screwdriver,

找到指甲油、找到螺丝刀，

bring them together, put them into a box and deliver them to you.

把它们放在一起、

放到一个盒子里、 交到你手上。

So this is extremely difficult, and it requires grasping.

这是极其困难的，需要抓取的动作。

So today, this is almost entirely done with humans.

眼下几乎完全由人类完成。

And the humans don't like doing this work,

而且人类不喜欢做这项工作，

there's a huge amount of turnover.

流失率很大。

So it's a challenge.

因此，这是一个挑战。

And people have tried to put robots

人们试图将机器人 放入仓库来完成这项工作。

into warehouses to do this work.

(Laughter)

（笑声）

It hasn't turned out all that well.

结果没那么好。

But my students and I, about five years ago,

但是，大约五年前，我和我的学生

we came up with a method, using advances in AI and deep learning,

想出了一种方法， 利用 AI 和深度学习的进步，

to have a robot essentially train itself to be able to grasp objects.

让机器人进行自我训练， 使其能够抓住物体。

And the idea was that the robot would do this in simulation.

我们想的是机器人 通过模拟进行这一过程。

It was almost as if the robot were dreaming about how to grasp things

就像是机器人想象如何抓住物体，

and learning how to grasp them reliably.

并学习如何稳定地抓住它们。

And here's the result.

以下是结果。

This is a system called Dex-net

这是一个名为 DEX-net 的系统，

that is able to reliably pick up objects

它能够稳定地拾取

that we put into these bins in front of the robot.

我们放在机器人前面的桶里的物体，

These are objects it's never been trained on,

这些是它从未接受过训练的物体，

and it's able to pick these objects up

它能够一遍又一遍地 捡起这些物体

and reliably clear these bins over and over again.

并稳定地清空这几个桶。

So we were very excited about this result.

我们对这个结果感到非常兴奋。

And the students and I went out to form a company,

我和学生们 出去创立了一家公司，

and we now have a company called Ambi Robotics.

现在我们有一家 名为 Ambi Robotics 的公司。

And what we do is make machines that use the algorithms,

我们让机器通过算法，

the software we developed at Berkeley,

即我们在伯克利开发的软件，

to pick up packages.

拾取包裹。

And this is for e-commerce.

这是为电子商务开发的。

The packages arrive in large bins, all different shapes and sizes,

这些包裹装在形状和大小 各不相同的大箱子里，

and they have to be picked up,

必须根据邮编将它们拾起、 扫描，然后放入较小的箱子里。

scanned and then put into smaller bins depending on their zip code.

We now have 80 of these machines operating across the United States,

我们在美国各地 有 80 台此类机器在运行，

sorting over a million packages a week.

每周分拣超过一百万个包裹。

Now that’s some progress,

已经取得了一些进展，

but it's not exactly the home robot that we've all been waiting for.

但这并不完全是 我们一直在期待的家用机器人。

So I want to give you a little bit of an idea

因此，我想向你们介绍一下 我们正在进行的一些新研究，

of some the new research that we're doing

to try to be able to have robots more capable in homes.

目的是让机器人 在家中发挥更大的能力。

And one particular challenge is being able to manipulate deformable objects,

一个特别的挑战是 操纵可变形的物体，

like strings in one dimension,

例如一维的绳索、

two-dimensional sheets and three dimensions,

二维的纸张和三维的物体，

like fruits and vegetables.

例如水果和蔬菜。

So we've been working on a project to untangle knots.

我们在做一个解开绳结的项目。

And what we do is we take a cable and we put that in front of the robot.

我们要做的是把一根电缆 放在机器人前面。

It has to use a camera to look down, analyze the cable,

它得用摄像头向下看、 分析电缆、

figure out where to grasp it

想出抓住什么位置、

and how to pull it apart to be able to untangle it.

如何把它抻开才能解开。

And this is a very hard problem,

这是一个非常棘手的问题，

because the cable is much longer than the reach of the robot.

因为电缆比机器人的 触及范围要长得多。

So it has to go through and manipulate, manage the slack as it's working.

它必须仔细检查和操纵， 在操作过程中留意松弛的部分。

And I would say this is doing pretty well.

我觉得它做得还挺不错。

It's gotten up to about 80 percent success

可以达到约 80% 的成功率，

when we give it a tangled cable at being able to untangle it.

让它解开一条打结的电缆。

The other one is something I think we also all are waiting for:

我觉得我们还在等待另一个机器人：

robot to fold the laundry.

叠衣服的机器人。

Now roboticists have actually been looking at this for a long time,

机器人专家其实已经研究了很长时间，

and there was some research that was done on this.

并且已经对此进行了一些研究。

But the problem is that it's very, very slow.

但问题在于它非常非常慢。

So this was about three to six folds per hour.

一个小时叠 3 到 6 次。

(Laughter)

（笑声）

So we decided to to revisit this problem

我们决定重新审视这个问题，

and try to have a robot work very fast.

试着让机器人快速工作。

So one of the things we did was try to think

我们做的一件事是试着

about a two-armed robot that could fling the fabric

让一台双臂机器人 像我们叠衣服的时候抖一抖衣物，

the way we do when we're folding,

and then we also used friction in this case to drag the fabric

我们还利用摩擦力拖动衣物，

to smooth out some wrinkles.

扯平褶子。

And then we borrowed a trick which is known as the two-second fold.

我们还参考了 “两秒叠衣服”的技巧。

You might have heard of this.

你可能听说过这个。

It's amazing because the robot is doing exactly the same thing

太神奇了， 因为机器人就在做这件事情，

and it's a little bit longer, but that's real time,

虽然时间长了点， 但这个是原倍速的。

it's not sped up.

没有开倍速。

So we're making some progress there.

我们已经取得了一些进展。

And the last example is bagging.

最后一个例子是装袋。

So you all encounter this all the time.

所有人都总是会遇到这种情况。

You go to a corner store, and you have to put something in a bag.

你去街角的商店， 你要在袋子里放点东西。

Now it's easy, again, for humans,

对于人类来说，这很容易，

but it's actually very, very tricky for robots

但其实对机器人来说 非常非常棘手，

because for humans, you know how to take the bag

因为对于人类来说， 你知道如何拿起袋子、

and how to manipulate it.

如何操纵它。

But robots, the bag can arrive in many different configurations.

但是对机器人来说， 袋子会以不同的形态出现。

It’s very hard to tell what’s going on

很难分辨发生了什么，

and for the robot to figure out how to open up that bag.

机器人也很难弄清楚 如何打开那个袋子。

So what we did was we had the robot train itself.

于是我们让机器人自我训练。

We painted one of these bags with fluorescent paint,

我们在袋子上涂了荧光涂料，

and we had fluorescent lights that would turn on and off,

配上可以开关的荧光灯，

and the robot would essentially teach itself how to manipulate these bags.

机器人自学如何操纵这些袋子。

And so we’ve gotten it now up to the point

我们现在已经能让它

where we're able to solve this problem about half the time.

解决这个问题的速度提速一倍。

So it works,

有效，但我得说我们还没完全达成。

but I'm saying, we're still not quite there yet.

So I want to come back to Moravec's paradox.

我想回过头来谈谈莫拉维克悖论。

What's easy for robots is hard for humans.

对机器人来说容易的事情 对人类来说却很难。

And what's easy for us is still hard for robots.

对我们来说容易的事情 对机器人来说很难。

We have incredible capabilities.

我们拥有卓越的能力。

We're very good at manipulation.

我们非常擅长操纵。

(Laughter)

（笑声）

But robots still are not.

但是机器人仍然不是。

I want to say, I understand.

我想说，我明白。

It’s been 60 years,

已经过去了 60 年，

and we're still waiting for the robots that the Jetsons had.

我们仍在等待 杰森一家拥有的机器人。

Why is this difficult?

为什么这很难？

We need robots because we want them to be able to do tasks that we can't do

我们之所以需要机器人，是因为 我们希望它们完成我们无法完成

or we don't really want to do.

或不太想做的任务。

But I want you to keep in mind that these robots, they're coming.

但我希望你知道， 这些机器人，它们即将到来。

Just be patient.

耐心点。

Because we want the robots,

因为我们想要机器人，

but robots also need us

但机器人也需要我们

to do the many things that robots still can't do.

去做许多机器人仍然做不到的事情。

Thank you.

谢谢。

(Applause)

（掌声）