AI-powered robot play badminton against humans

多客科技

作者：微信文章


Scientists taught an AI-powered robot how to play badminton against humans

From building cars to flipping burgers, robots today do a lot. And now researchers working in Switzerland have trained a robot called “ANYmal” to play badminton, too. Badminton is a fast-moving sport a bit like tennis: players hit an object called a shuttlecock back and forth with rackets over a net on a court.

The robot is only as good as a seven-year-old playing badminton. Even so, it’s an exciting step forward. The researchers published their study in May in a scientific journal.

About The Robot

ANYmal is a dog-like robot with four legs. It weighs 110 pounds and is about 1.5 feet tall. It was developed at the Robotic Systems Lab at ETH Zürich, a university in Switzerland. Having four legs helps ANYmal climb rocks and jump over things. The robot uses AI, or artificial intelligence. AI is a type of computer system. It can perform some things that humans do, like spot patterns and make decisions.

Researchers have added arms to these dog-like robots. They’ve trained the robots to fetch things and open doors. But making a robot move and see things correctly in a fast-changing environment like a sports game is still hard. "Sports is a good application for this kind of research,” said Yuntao Ma, who worked on the study. That’s “because you can gradually increase the competitiveness or difficulty."

Teaching A New Dog New Tricks

In this study, Ma and his team attached a moving arm with a badminton racket at a 45-degree angle to a standard ANYmal robot. (An angle is a figure formed by two rays meeting at a common point. Angles are measured in units called degrees.)

With the arm, the robot looked more like giraffe than a dog. (That is, if a giraffe could hold a badminton racket in its mouth.) The researchers designed a built-in system that controlled how the robot’s arm and legs moved. ANYmal also had two cameras that acted as eyes: they helped the robot figure out where the shuttlecocks were going.

Computer software then taught the robot to become a badminton player. In ANYmal’s case, it used trial and error to learn to spot and track the shuttlecock, move toward it, and swing the racket.

To do this, the researchers first created a computer simulation of a badminton court, with the robot's virtual version standing in the center. (A simulation is a model or imitation of a real-life object or situation.) Virtual shuttlecocks were served from near the center of the opponent's side of the court. The robot had to track the shuttlecock’s position and guess its trajectory, or flight path.

When playing badminton, human players have to figure out the best angles to make good shots. (Imagine a pair of scissors. Now imagine each blade is a ray — if you open them, the space between the blades is an angle.) Understanding the trajectory of the shuttlecock is also important in badminton because it helps a player tell where it will land. The badminton shot called a forehand smash, for example, is hit with power and speed downward into the opponent’s court. The angle and steepness of the shuttlecock’s trajectory make it hard for the opponent to return the shot.
But before it could do any of that, ANYmal had to learn how to hit a shuttlecock first. The researchers made a strict training plan. There were thousands and thousands of trials of this simulation training. 文章出自《Newsela》重点短语：

teach sb. to do sth.（教某人做某事）train sb. to do sth.（训练某人做某事）develop a robot（研发机器人）publish a study（发表研究）take an exciting step forward（迈出激动人心的一步）increase the difficulty（增加难度）

play badminton（打羽毛球, 球类运动不用定冠词“the”）hit a shuttlecock（击打羽毛球）swing the racket（挥拍）

重点句型：描述 “能力与功能”

句型：主语 + can /could + 动词原形 + 其他例句：It can perform some things that humans do, like spot patterns and make decisions.（它能完成一些人类做的事，比如识别规律和做决定。

定语从句

句型：先行词 + that /which/who + 从句例句 1：Badminton is a fast-moving sport that is a bit like tennis.（羽毛球是一项快速的运动，有点像网球。）例句 2：Yuntao Ma is the person who worked on the study.（马运涛是参与这项研究的人。）

中文翻译参考：
科学家教会人工智能机器人与人类打羽毛球

从制造汽车到制作汉堡，如今的机器人能做很多事。现在，瑞士的研究人员还训练出了一款名为 “ANYmal” 的机器人，让它学会了打羽毛球。羽毛球是一项节奏快速的运动，有点类似网球：运动员用球拍将一个名为 “羽毛球” 的物体隔网在球场上来回击打。

这款机器人的羽毛球水平仅相当于七岁儿童。即便如此，这仍是一项激动人心的进步。研究人员已于 5 月在某科学期刊上发表了他们的研究成果。
关于机器人

ANYmal 是一款形似小狗的四足机器人，体重 110 磅（约 50 公斤），身高约 1.5 英尺（约 46 厘米）。它由瑞士苏黎世联邦理工学院（ETH Zürich）的机器人系统实验室研发。四条腿的设计能帮助 ANYmal 攀爬岩石、越过障碍物。该机器人搭载了人工智能（AI）技术，AI 是一种计算机系统，能完成一些人类的行为，比如识别规律、做出决策。

研究人员为这类狗形机器人加装了机械臂，并训练它们完成取物、开门等动作。但在体育比赛这种快速变化的环境中，要让机器人准确移动、识别物体，难度依然很大。参与此项研究的马运涛（音译）表示：“体育是这类研究的理想应用场景。” 因为 “你可以逐步提高比赛的竞争性或难度”。
教 “新狗” 学 “新招”

在这项研究中，马运涛及其团队为标准版 ANYmal 机器人加装了一条机械臂，机械臂末端的羽毛球拍呈 45 度角固定（角是由两条有公共端点的射线组成的图形，单位为 “度”）。

加装机械臂后，这款机器人看起来更像长颈鹿，而非小狗（试想一下，如果长颈鹿能用嘴叼着羽毛球拍，大概就是这个样子）。研究人员设计了一套内置系统，用于控制机器人手臂和腿部的运动方式。ANYmal 还装有两个充当 “眼睛” 的摄像头，帮助它判断羽毛球的飞行方向。

随后，研究人员通过计算机软件让机器人学会打羽毛球。具体来说，ANYmal 通过 “试错法” 学习识别、追踪羽毛球，向羽毛球移动，并挥拍击打。

为实现这一目标，研究人员首先搭建了一个羽毛球场地的计算机模拟场景，让机器人的虚拟版本站在场地中央（模拟是对现实物体或场景的建模或模仿）。虚拟羽毛球会从对手场地的中线附近发出，机器人需要追踪羽毛球的位置，并预判其飞行轨迹。

人类打羽毛球时，需要判断最佳角度以打出好球（想象一把剪刀，把两片刀刃看作两条射线 —— 打开剪刀时，刀刃之间的空间就是一个角）。理解羽毛球的飞行轨迹同样重要，因为这能帮助球员判断球的落点。例如，羽毛球中的 “正手扣杀” 动作，就是用力量和速度将球向下击向对手场地。羽毛球飞行轨迹的角度和陡峭程度，会让对手难以回球。

但在做到这些之前，ANYmal 首先要学会如何击打羽毛球。研究人员制定了严格的训练计划，仅模拟训练就进行了成千上万次。