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

譯者: 麗玲 辛

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

我猜想，在座大多數人 都希望家裡有機器人。

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

會做家事、處理事務，真是太好了

Where are these robots?

這些機器人在哪裡？

What's taking so long?

怎麼要等這麼久？

I mean, we have our tricorders,

我的意思是，我們有三度儀，

and we have satellites.

還有衛星，

We have laser beams.

雷射光束。

But where are the robots?

但機器人在哪裡？

(Laughter)

（笑聲）

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

我的意思是，好的，等等， 我們家裡確實有些機器人，

but, not really doing anything that exciting, OK?

但是，沒有做什麼酷炫的事，對吧？

(Laughter)

（笑聲）

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

我已經在加州大學柏克萊分校 和學生共同研究機器人 30 年，

with my students on robots,

and in the next 10 minutes,

接下來的 10 分鐘裏，

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

我會試著解釋科幻小說 與現實之間的差距。

Now we’ve seen images like this, right?

我們看過這樣的影像，對吧？

These are real robots.

這是真正的機器人，

They're pretty amazing.

很棒的機器人。

But those of us who work in the field,

但對我們在這領域工作的人，

well, the reality is more like this.

嗯，真實情況更像這樣。

(Laughter)

（笑聲）

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

100 次中的 99 次都是這樣。

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

在這個領域，有個理論 可以解釋這一點，

that we call Moravec's paradox.

我們稱之為莫拉維克悖論。

And that is, what's easy for robots,

對機器人來說很容易的任務，

like being able to pick up a large object,

例如，拿起一個大型物體，

large, heavy object,

又大又重的物體，

is hard for humans.

人類來做卻很難。

But what's easy for humans,

但對人類來說很容易的事情，

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

例如撿起一些積木，堆疊起來，

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

事實證明這對機器人來說非常困難。

And this is a persistent problem.

這個問題一直存在。

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

因此，抓取任意物體的能力 對我的領域來說是巨大的挑戰。

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

順便一提，我小時候笨手笨腳的。

(Laughter)

（笑聲）

I would drop things.

我會掉東西。

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

每當有人扔球給我，我都接不到。

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

打籃球時，我是最後被選的那個人。

I'm still pretty klutzy, actually,

事實上，我仍然很笨拙，

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

但我整個職業生涯都在研究 如何讓機器人不要那麼笨拙。

Now let's start with the hardware.

讓我們從硬體開始，

So the hands.

首先，手。

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

這是機器人手，某種特定類型的手。

It's a lot like our hand.

這很像我們的手。

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

如您所見，這有很多馬達，

and cables as you can see.

很多肌鍵和電纜。

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.

它很重，也非常昂貴。

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

所以我更喜歡像這樣簡單的手，

So this has just two fingers.

只有兩根手指，

It's known as a parallel jaw gripper.

被稱為平行夾爪。

So it's very simple.

它很簡單，

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

輕巧可靠，並且非常便宜。

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

如果您懷疑這樣簡單的手是否有效，

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

請看這個影片，你可以看到 兩個非常簡單的夾爪，

these are being operated, by the way,

順便一提，這些夾爪由人操控，

by humans who are controlling the grippers like a puppet.

就像控制木偶一樣。

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

但非常簡單的夾爪機器 能做非常複雜的事情。

Now actually in industry,

事實上，在業界，

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

還有一種更簡單的機械夾爪， 那就是吸盤。

And that only makes a single point of contact.

只需要一個接觸點。

So again, simplicity is very helpful in our field.

同樣，「簡單」在這個領域很有助益。

Now let's talk about the software.

現在讓我們談談軟體。

This is where it gets really, really difficult

這就是很難的地方，

because of a fundamental issue, which is uncertainty.

源於一個根本問題，不確定性。

There's uncertainty in the control.

控制方面有不確定性。

There’s uncertainty in the perception.

感測方面有不確定性。

And there’s uncertainty in the physics.

而且物理上也存在不確定性。

Now what do I mean by the control?

控制是什麼意思？

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

當你觀察機器人的夾爪試圖做某事，

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

你會發現電纜和機械裝置 存在許多不確定性，

that cause very small errors.

可能導致微小的錯誤。

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

這累積起來， 會讓物品操縱變得非常困難。

Now in terms of the sensors, yes,

就感測器而言，是的，

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

機器人像我們一樣，擁有 非常高解析度的攝像頭，

and that allows them to take images of scenes in traffic

使得它們能夠拍攝交通場景、

or in a retirement center,

退休中心、

or in a warehouse or in an operating room.

倉庫或手術室的影像。

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

但這並不能提供現況的三維結構。

of what's going on.

So recently, there was a new development called LIDAR,

最近，有一項新的技術， 稱為光學雷達 (LIDAR)，

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

這是一種新型攝影機， 使用光束建立環境的三維模型。

a three-dimensional model of the environment.

And these are fairly effective.

這相當有效。

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

這確實是我們領域的突破， 但這並不完美。

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

如果物體會反光或透明，

well, then the light acts in unpredictable ways,

那麼光線會出現不可預測的狀況，

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

最終會在影像中產生雜訊和孔洞。

So these aren't really the silver bullet.

所以這不是真正的解決方法。

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

現在還有另一種形式的傳感器， 稱為「觸覺感應器」。

And these are very interesting.

這很有趣。

They use cameras to actually image the surfaces

他們使用攝影機，對表面進行攝影，

as a robot would make contact,

如同機器人接觸表面，

but these are still in their infancy.

但這仍處於起步階段。

Now the last issue is the physics.

最後一個問題是物理。

And let me illustrate for you by showing you,

讓我向您展示，

we take a bottle on a table

我們在桌子上放一個瓶子，然後推它，

and we just push it,

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

機器人每次都以完全相同的方式推動。

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

但您可以看到，瓶子到的地方 每次都不一樣。

And why is that?

為什麼會這樣？

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

這是因為這取決於瓶子滑動時，

underneath the bottle as it slid.

它下方的微觀表面地形。

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

例如，如果你在那裡放一粒砂粒，

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

相較於沒有沙粒時，反應會非常不同。

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

而且我們無法看到是否有沙粒， 因為那是在瓶子下面。

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

事實證明，我們可以預測 一個百萬英里遠的小行星的運動，

a million miles away,

far better than we can predict the motion of an object

準確度比我們預測一個物體 被機器人抓取時的運動要好得多。

as it's being grasped by a robot.

Now let me give you an example.

容我給您一個例子。

Put yourself here into the position of being a robot.

把自己置於機器人的角色。

You're trying to clear the table

您正在清理桌子，

and your sensors are noisy and imprecise.

而您的感應器噪音很大且不精確。

Your actuators, your cables and motors are uncertain,

您的驅動器、纜線和馬達不穩定，

so you can't fully control your own gripper.

因此您無法完全控制自己的夾具。

And there's uncertainty in the physics,

物理上有不確定性，

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

所以您真的不知道會有什麼狀況。

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

因此，機器人仍十分笨拙， 也就不足為奇。

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

現在機器人有個絕佳用途， 那就是電子商務。

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

這個趨勢一直在增長，是巨大的趨勢。

And during the pandemic, it really jumped up.

在疫情期間，更是突飛猛進。

I think most of us can relate to that.

我想大多數人都有體會到這一點。

We started ordering things like never before,

我們比以往更常訂購商品，

and this trend is continuing.

而且這種趨勢仍在繼續。

And the challenge is to meet the demand,

挑戰在於滿足需求，

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

我們必須能及時寄送所有包裹。

And the challenge is that every package is different,

挑戰在於每個包裹都是不同的，

every order is different.

每個訂單都不同。

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

您可能訂購幾瓶指甲油 和一把電動螺絲起子。

And those two objects are going to be

這兩個物體會放在 一個巨型倉庫的某個地方。

somewhere inside one of these giant warehouses.

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

必須要有人進去，

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

找到指甲油，然後去找螺絲刀，

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

把它們放在一起， 放進盒子裡，然後寄給您。

So this is extremely difficult, and it requires grasping.

這非常困難，需要抓取動作。

So today, this is almost entirely done with humans.

現今，這幾乎完全是用人力完成。

And the humans don't like doing this work,

人們不喜歡做這項工作，

there's a huge amount of turnover.

人員流動量很大。

So it's a challenge.

所以這是一個挑戰。

And people have tried to put robots

人們試著將機器人放入倉庫 來做這項工作。

into warehouses to do this work.

(Laughter)

（笑聲）

It hasn't turned out all that well.

結果並不怎麼好。

But my students and I, about five years ago,

但大約五年前，我和我的學生 想出了一種方法，

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

利用人工智慧和深度學習的進展，

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

讓機器人進行自我訓練，抓取物體。

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

我們的想法是讓機器人 在模擬中做到這件事。

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

這幾乎就像機器人想像如何抓住東西，

and learning how to grasp them reliably.

並學習如何穩定地抓住它們。

And here's the result.

這就是結果。

This is a system called Dex-net

這是一個名為 Dex-Net 的系統，

that is able to reliably pick up objects

它能夠穩定地從它面前的桶中，

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

撿起我們放在裏面的物品。

These are objects it's never been trained on,

這些是它從未受過訓練的物體，

and it's able to pick these objects up

它能夠一次又一次地將這些物體

and reliably clear these bins over and over again.

拿起來，確實地清空桶子。

So we were very excited about this result.

因此，我們對這個結果非常興奮。

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

我和學生們出來成立一家公司，

and we now have a company called Ambi Robotics.

現在我們有一家 名為安比機器人的公司。

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

我們所做的就是製造機器，

使用我們在伯克萊開發的 演算法和軟體，

the software we developed at Berkeley,

to pick up packages.

來撿取包裹。

And this is for e-commerce.

這是用於電子商務。

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

包裹放在大型桶中送來， 有各種形狀和大小，

and they have to be picked up,

機器人必須拿取包裹、掃描、

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

再根據郵遞區號放入較小的桶中。

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

我們現在有 80 台這樣的機器， 在美國各地運行，

sorting over a million packages a week.

每週分類超過一百萬個包裹。

Now that’s some progress,

這只是一點進展，

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

還不是我們所有人 都在等待的家庭機器人。

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

因此，我想向您介紹

of some the new research that we're doing

我們正在進行的一項新研究，

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

目的是讓機器人在家庭中 發揮更大的作用。

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

一項特殊的挑戰是 能夠操縱可變形的物體，

like strings in one dimension,

例如一維的繩子、

two-dimensional sheets and three dimensions,

二維的布面和三維的物體，

like fruits and vegetables.

如水果和蔬菜。

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

我們一直在努力一個項目是解開繩結。

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

我們所做的就是拿一條電線， 放在機器人前面。

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

它必須使用攝影機 向下觀察，分析電線，

figure out where to grasp it

找出該從哪裡抓，

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

要怎樣拉開，才能解開。

And this is a very hard problem,

這是一個非常困難的問題，

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

因為電線比機器人的手長得多。

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

因此，它必須仔細檢查、操作、 在過程中注意鬆開的部份。

And I would say this is doing pretty well.

我覺得這做得相當好了。

It's gotten up to about 80 percent success

當我們給它一條纏結的電線，

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

它能夠解開，成功率高達 80%。

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

另一個是我想我們都在等待的東西：

robot to fold the laundry.

摺衣服的機器人。

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

機器人專家實際上已經關注 這個問題很長一段時間了，

and there was some research that was done on this.

並且對此進行了一些研究。

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

但問題是它非常非常慢。

So this was about three to six folds per hour.

每小時大約摺三到六件。

(Laughter)

（笑聲）

So we decided to to revisit this problem

因此，我們決定重新審視這個問題，

and try to have a robot work very fast.

並試著讓機器人動作變快。

So one of the things we did was try to think

我們嘗試設計一個雙臂機器人，

about a two-armed robot that could fling the fabric

它可以像我們摺衣服時那樣甩衣物，

the way we do when we're folding,

然後在這種情況下， 我們還使用摩擦力來拖動衣物，

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

to smooth out some wrinkles.

以撫平一些皺摺。

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

然後我們借用了一個技巧， 叫做「兩秒摺衣」。

You might have heard of this.

你可能有聽說過。

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

很神奇，機器人正在做完全相同的事情，

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

只是時間更長一點，

但這是即時影像，沒有快轉。

it's not sped up.

我們正在這方面取得一些進展。

So we're making some progress there.

And the last example is bagging.

最後一個例子是裝袋。

So you all encounter this all the time.

大家經常會遇到這樣的狀況。

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

您去街角商店， 必須把東西放進袋子裡。

Now it's easy, again, for humans,

對人類來說，這很容易，

but it's actually very, very tricky for robots

但對機器人來說，這非常非常棘手，

because for humans, you know how to take the bag

因為人類知道怎麼拿袋子，

and how to manipulate it.

如何使用它。

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

但對機器人來說， 袋子會以不同型態出現。

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

機器人很難辨定當時的狀況，

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

要怎麼打開袋子。

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

我們所做的是讓機器人自我訓練。

We painted one of these bags with fluorescent paint,

我們用螢光漆塗上一個袋子，

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

配上可以開關的螢光燈 ，

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

機器人自學如何操控這些袋子。

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

我們現在已經做到

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

用一半的時間就可以解決這個問題。

So it works,

這方法是有效的，

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

但我想說的是，我們還沒有完全做到。

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

我想回到莫拉維克悖論。

What's easy for robots is hard for humans.

對機器人來說容易的， 對人類來說很難。

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

而對我們來說容易的， 對機器人來說仍然很難。

We have incredible capabilities.

我們有令人難以置信的能力。

We're very good at manipulation.

我們很擅長操控。

(Laughter)

（笑聲）

But robots still are not.

但機器人仍然不擅長。

I want to say, I understand.

我想說，我明白。

It’s been 60 years,

60 年了，

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

我們還在等待 <<傑森一家>> 裏的機器人。

Why is this difficult?

為什麼這很難？

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

我們需要機器人，

因為我們希望他們能夠完成 我們做不到或不想做的任務。

or we don't really want to do.

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

但我希望你記住，機器人會到來的，

Just be patient.

請耐心等待。

Because we want the robots,

因為我們想要機器人，

but robots also need us

但機器人也需要我們

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

去做許多機器人仍然無法做到的事情。

Thank you.

謝謝

(Applause)

（掌聲）