If you've ever watched professional ping-pong players on TV, you may have wondered if -- due to their rapid-fire pace of play -- they were actually robots.
Well wonder, no more. A German research team created the first robot that can not only play ping-pong like a human, but can also improve its game by learning from its human opponents like the ones you've maybe seen whizzing around the table in a blur.
The robot -- the brainchild of PhD candidate and robotic researcher Katharina Muelling at the Technical University of Darmstadt -- consists of a robotic arm to which a ping-pong paddle is attached, as well as a camera that provides a view of the table and area of play.
Researcher Katharina Muelling poses with a ping pong playing robot she and her team at the Technical University of Darmstadt in Germany designed and built. The robot is comprised of an arm to which a paddle is attached as well as a camera that watches the table and area of play, responding to the opponent's moves. (Source: The Technical University of Darmstadt)
Muelling's work focuses on developing robots that can achieve motor control and perform complex motor-oriented tasks, as well as learn to adapt as they perform them -- a concept called "kinesthetic teach-in." She outlined her work on the robot in a paper available online.
As described by Muelling, she and her team used table tennis as a "benchmark task" to design the mixture of motor primitives (MoMP) algorithm controlling the robot, allowing it to understand one basic set of movements and then dynamically apply those movements as it goes along. (Watch a video of this process below.)
"The goal of this task is to learn autonomously from and with a human to return a table tennis ball to the opponent's court and to adapt its movements accordingly," she wrote in the paper, titled "Learning to Select and Generalize Striking Movements in Robot Table Tennis."
Once the robot learns movements from a human teacher -- the kinesthetic teach-in aspect of the basic task -- the machine's MoMP is programmed according to these movements. Following this, the robot's learning system identifies the movements it takes to hit the ball, generalizes them, and is able to apply them later throughout a game or prolonged play with an opponent, Muelling wrote. "The resulting system is able to return balls served by a ball launcher as well as to play in a match against a human."
From the looks of the video, Muelling's robotic ping-pong player isn't quite ready for the Olympics yet, but it could provide a steady opponent for practice -- the equivalent of a mechanical ball pitcher for batting practice or playing tennis against a wall.
I wouldn't think that a legal serve should confuse the robot. In order to operate at all it needs to know the ball's location in space as well the "field" (i.e., its side of the table). Not sure that it would be able to keep score, but I would think it would be relatively simple to discount any bounces on the far side of the net, considering everything else it is work off of.
This was pretty fun to watch as the robot learned and got better. I'm sure somebody will eventually figure out a useful application for this one-armed pongster even if it's only for ping pong training camps. Maybe it can be used to toss packs of peanuts into the stands during a ball game?
Good point about human verses bot reaction time - it reminded me of Data when he was tempted by the Borg Queen's offer to join her in First Contact - Captain Picard asked him how long he considered it and Data replied, "0.68 seconds sir. For an android, that is nearly an eternity."
Nice article, Elizabeth. I especially like the video. It seems were seeing more and more versions of humans against the machine. I love the fact that it learns. However, Chuck makes a good point about the backhand.
This is pretty amazing to see the robot learn how to play over time. At the Robot display at the Carnegie Science Center in Pittsburgh, they have a robot setup to play air hockey. The robot used vision to analyze the table then would only go on an offensive shot when it saw that there was a clear angle to the goal. At all other times, it stayed on defense. The robot did a pretty good job and won most of its matches.
Truchard will be presented the award at the 2014 Golden Mousetrap Awards ceremony during the co-located events Pacific Design & Manufacturing, MD&M West, WestPack, PLASTEC West, Electronics West, ATX West, and AeroCon.
Robots that walk have come a long way from simple barebones walking machines or pairs of legs without an upper body and head. Much of the research these days focuses on making more humanoid robots. But they are not all created equal.
The IEEE Computer Society has named the top 10 trends for 2014. You can expect the convergence of cloud computing and mobile devices, advances in health care data and devices, as well as privacy issues in social media to make the headlines. And 3D printing came out of nowhere to make a big splash.
For industrial control applications, or even a simple assembly line, that machine can go almost 24/7 without a break. But what happens when the task is a little more complex? That’s where the “smart” machine would come in. The smart machine is one that has some simple (or complex in some cases) processing capability to be able to adapt to changing conditions. Such machines are suited for a host of applications, including automotive, aerospace, defense, medical, computers and electronics, telecommunications, consumer goods, and so on. This discussion will examine what’s possible with smart machines, and what tradeoffs need to be made to implement such a solution.