機器人完成了人類“最艱難的任務”:組裝宜家傢俱

Robots have taken our jobs, learned our chores and beaten us at our own games.

機器人搶走了我們的工作，學會了做家務，還在比賽中打敗了我們。

Now researchers in Singapore say they have trained one to perform another task known to confound humans: figuring out how to assemble furniture from Ikea.

現在，新加坡的研究人員們表示，他們已經成功地訓練出機器人執行另一個備感驚訝的人類所熟知的任務：想辦法組裝宜家傢俱。

A team from Nanyang Technological University programmed a robot to create and execute a plan to piece together most of Ikea’s $25 solid-pine Stefan chair on its own, calling on a medley of human skills to do so. The researchers explained their work in a study published on Wednesday in the journal Science Robotics.

南洋理工大學(Nanyang Technological University)的一個團隊給一個機器人編了程序，讓其創造並執行計劃，自主組裝宜家售價25美元的松木斯第芬(Stefan)椅的大部分，要做到這一點需要各種人類的技能。研究人員週三發表在《科學機器人》(Science Robotics)雜誌上的研究報告對他們的研究進行了說明。

“If you think about it, it requires perception, it requires you to plan a motion, it requires control between the robot and the environment, it requires transporting an object with two arms simultaneously,” said Dr. Quang-Cuong Pham, an assistant professor of engineering at the university and one of the paper’s authors. “Because this task requires so many interesting skills for robots, we felt that it could be a good project to push our capabilities to the limit.”

“想一想，這需要洞察力，需要你對行動做出規劃，需要在機器人和環境間的控制，需要同時用兩個機械臂運送一件物品，”範光強（Quang-Cuong Pham，音）博士說，他是該校的工程學助理教授，也是該論文的作者之一。“因爲這項任務對機器人來說需要大量有意思的技能，我們覺得這會是一個將我們的能力拓展到極致的好項目。”

He and his Nanyang colleagues who worked on the study, Francisco Suárez-Ruiz and Xian Zhou, aren’t alone.

他和南洋理工大學一道進行這項研究的同事弗朗西斯科·蘇亞雷斯-魯伊斯(Francisco Suárez-Ruiz)及周憲（音）並非獨行者。

In recent years, a handful of others have set out to teach robots to assemble Ikea furniture, a task that can mimic the manipulations robots can or may someday perform on factory floors and that involves a brand many know all too well.

近年來，其他一些人也開始教機器人組裝宜家傢俱，這個任務可以模擬機器人有朝一日能夠或者可能在工廠裏進行的操作，而且它涉及一個許多人熟知的品牌。

“It’s something that almost everybody is familiar with and almost everybody hates doing,” said Ross A. Knepper, an assistant professor of computer science at Cornell University, whose research focuses on human-robot interaction.

“這是幾乎每個人都熟悉，也是每個人都討厭做的事情，”康奈爾大學(Cornell University)計算機科學助理教授羅斯·A·克內珀(Ross A. Knepper)說，他的研究主要聚焦在人機交互方面。

In 2013, Mr. Knepper was part of a team at the Massachusetts Institute of Technology that presented a paper on its work in the area, describing the “IkeaBot” the team created, which could assemble the company’s Lack table on its own.

2013年的時候，克內珀曾是麻省理工大學(Massachusetts Institute of Technology)一個研究團隊的成員，該團隊發表了一篇相關領域的論文，對他們創造的“宜家機器人”(IkeaBot)進行了描述，這個機器人可以自主組裝宜家的拉科(Lack)桌。

But chairs, with backs, stretchers and other parts, pose a more complex challenge; hence the interest of the Nanyang researchers.

但組裝帶有靠背、橫檔及其他部件的椅子難度更大，因此這激起了南洋理工大學研究人員的興趣。

Their robot was made of custom software, a three-dimensional camera, two robotic arms, grippers and force detectors. The team chose only off-the-shelf tools, in order to mirror human biology.

他們的機器人由特別設計的軟件、一個立體照相機、兩隻機械臂、抓握器及力感應器組成。爲了模仿人類生物學，該團隊選擇的都是已有的現成工具。

“Humans have the same hardware to do many different things,” Dr. Pham said. “So this is kind of the genericity that we wanted to mimic.”

“人類有同樣的硬件來做許多不同的事情，”範光強說。“所以這就是我們想模仿的泛型。”

Also like humans, the robot had a little help to start: It was fed a kind of manual, a set of ordered instructions on how the pieces fit together. After that, though, it was on its own.

同樣地，像人類一樣，這款機器人在開始組裝傢俱時也沒得到什麼幫助：提供給它的是某種指南，一系列關於如何將各個部位組裝起來的指令。在那之後，它就只能靠自己了。

The robot proceeded in three broad phases, spread out over 20 minutes 19 seconds.

機器人分三個大階段開始進行，共耗時20分鐘19秒。

First, like humans, it took some time to stare at the pieces scattered before it.

首先，像人類一樣，它花了些時間盯着散落在自己面前的各個部位。

The robot spent a few seconds photographing the scene and matching each part to the one modeled in its “manual.”

機器人花了幾秒，對眼前的景象進行拍照，把每個部位與“指南”裏的進行匹配。

Then, over more than 11 minutes, the robot devised a plan that would allow it to quickly assemble the chair without its arms knocking into each other or into the various parts.

然後，在11分鐘多的時間裏，機器人設計了一個方案，能夠讓它在機械臂不相互碰撞或撞到不同零件的情況下，快速將椅子組裝起來。

Finally, it put the plan in motion over the course of nearly nine minutes. The robot used grippers to pick up the wooden pins from a tray and force sensors at its “wrists” to detect when the pins, searching in a spiral pattern, finally slid into their holes. Working in unison, the arms then pressed the sides of the chair frame together.

最終，它在近9分鐘時間裏執行了這個方案。機器人利用抓握器從一個托盤裏拿起木釘，位於其“腰部”的力感應器會以螺旋方式進行搜索，檢測木釘何時應當被置入孔中。兩條機械臂協同合作，然後將椅子的框架向內按壓，合在一起。

Of course, the robot didn’t succeed right away. There were several failed attempts along the way and researchers tweaked the system before the robot was finally able to assemble the chair on its own.

當然，機器人也不是立刻就取得了成功。期間也有不少失敗，在機器人最終能夠自主組裝椅子前，研究人員對系統進行了不少改動。

The accomplishment was the culmination of three years of work, but the team is eager to see what else it can automate, Dr. Pham said.

這一成就是三年努力的結果，但範光強說，該團隊還渴望看到這個機器人能自動操作其他什麼東西。

With the help of experts in artificial intelligence, the researchers may be able to create a robot that can build a chair by following spoken directions or by watching someone else do it first, he said. Or maybe, he said, they’ll eventually develop one that assembles furniture in a way that is truly human: by ignoring the manual altogether.

他說，在人工智能專家的幫助下，研究人員們可能會創造出一款機器人，它會聽從語音指令，或先觀看別人組裝，再自己來動手。又或者，他說，他們最終能開發出一款能真正以人類的方式組裝傢俱的機器人：完全忽視組裝指南。