Podcast: Teaching Robots to Play Badminton

In this episode, we cover how researchers trained a quadruped robot to play badminton, teaching it to track the shuttlecock, move into position, and swing a racket in real time, showing how AI can bring robots closer to human-like coordination.

This podcast is sponsored by Mouser Electronics. 

Episode Notes

(4:20) – Playing badminton against a robot

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn how cobots are driving Industry 5.0 with safer, smarter human-robot teamwork!

Become a founding reader of our newsletter: http://read.thenextbyte.com/

Transcript

Welcome back to the Next Byte podcast. In this one, we're talking about badminton and robots. And in your mind, you're thinking, what do these two have to do with each other? And at a glance, nothing, but at a much higher level, everything, from critical applications of remote surgeries to the next generation of manufacturing.

What's up friends, this is The Next Byte Podcast where one gentleman and one scholar explore the secret sauce behind cool tech and make it easy to understand.

Farbod: What's going on, folks? Welcome back to the Next Byte podcast on today's episode, Badminton and Robots. When was the last time you played badminton? I think for me it was like, oh God, fifth grade?

Daniel: I've played a couple summers ago. My aunt and uncle brought like a pop-up badminton kit to family vacation. Were all playing against one another. It's an interesting game. It's challenging. It's fun for all ages. It's one of those things where it's like raw speed and power is not the only thing that dictates how it will successfully or it's more about like being agile and having the right amount of touch.

Farbod: Absolutely. And as I've gotten older and slower, I'm wondering if I've still got it in me. So, an aside, but we should test out this theory, Daniel, sometime this weekend if you got the time.

Daniel: Yeah.

Farbod: But before we get into today's episode, let's talk about today's sponsor, Mouser Electronics. So, if you've been rocking with us, you know Daniel and I, love Mouser.   They're one of the world's biggest electronic distributors. simply by the nature of being that, they have access to a lot of cool things happening in both industry and academia. Now, sometimes they write articles about what's going on. Sometimes they do interviews. Well, we found this really interesting article that we're going to be linking into today's show notes about the human-robot collaboration, and specifically how it's empowering this industry 5.0 revolution. They talk about how you have this really high-tech set of factories that were already adopting collaborative robots to help expedite the different tasks that the operators on the factory floor are doing to ensure higher precision. It kind of reminds me of the automakers that have tons of robotic arms on their factory line. But then COVID hit and then even the folks who were a little bit hesitant or slow to adopt this technology quickly came on board because they were short staffed and they needed the extra support however they could get it. And now these cobots are being deployed at scale to get rid of the repetitive tasks that require a lot of precision and actually enabling the human operators to do things that require that human capital, that knowledge of everything that's going on, that intelligence which is great to see. I know robotics or artificial intelligence, anything tech, has been kind of a hot topic recently, especially as it pertains to reducing the workforce as people struggle to find jobs. But one of the things that this article highlights is that there's actually a limited number of people that want to do these types of jobs or are skilled enough to do these kind of jobs. And the addition of these collaborative robots are enabling them to operate at their best capacity. So that's another uh fun insight that I thought it would be worth sharing with you guys. But yeah.

Daniel: And we've been on a lot of factory visits recently. We've seen a lot of awesome applications of where robots and humans collaborate together. I feel like even a couple of years ago when we first started covering cobots here on the podcast, especially it was an academic term or theoretical term more so than a term that I've heard in industry or in practice. Now we've been in a lot of factory visits and I'm hearing operations managers and manufacturing engineers talk about cobots and bragging about the cobots that they've got in place. So, it's definitely something that's making its way from the academic realm, the theoretic realm into practice, into industry. And it's definitely cool to partner with Mouser along the way to get some education on these topics that are at the forefront and innovation and technology and making their way into the real world.

Farbod: Absolutely, on that note. Let's move over to today's article and we're going to be going all the way to ETH Zurich, again, playing badminton with robots, which seems like odd. Out of all these robotic applications, we've been seeing robots helping police departments to safely inspect potentially risky scenarios before humans are deployed or robots for search and rescue. Now you have badminton, which to me is the equal of that spot robot from Boston Dynamics dancing to Bruno Mars. Like super cool to look at, but I don't get it. Like what's the application here?  And turns out roboticists have been having robots do one form or another of badminton ping pong since the 1980s because it's the perfect task of perception, planning, and acting in a very short window of time.

Daniel: Yeah, in real time.   

Farbod: In real time, exactly. So, you're almost stress testing these capabilities that you would expect from a robot in these critical applications like search and rescue. And obviously a much safer setting.

Daniel: If you've seen a lot of robotics demos before the technology is completely mature, it almost looks like they're moving in slow motion. And that's because they are, right? Like they walk a little bit slower than you might see a normal person walking on the street. If you've seen any of these laundry folding robots demonstrations, it looks like they're folding robots really slow or a pick and place demonstration when they're first getting started, when the technology is not yet mature, they do move in slow time and they wouldn't be able to do something fast, like react in real time to hitting the shuttlecock in a badminton game. But playing badminton requires fast eyes to be able to recognize the shuttlecock and understand the trajectory of it moving quick feet to move the robot into the correct position, then also a good swing all at once. And it has to be coordinated and do it all at once. And that's why it's hard for humans to do it because it requires a lot of coordination. So, like you're saying, it's kind of this this pinnacle of understanding the vision systems, the control systems and several different control systems in concert, right? Being able to move the legs of this four-legged robot around as well as the arm that's holding the racket in the right position. It's something challenging and again, it may not have too much practical application. Like, do I want there to be robots everywhere playing badminton with everyone? Maybe, maybe not. But it is a way of kind of doing a strength test or flexing your muscles to some extent from a robotic perspective to say, hey, this is how far our technologies come. We can or cannot do this task. Or these are the limitations of our ability to do this challenging task, which requires, like you're saying, a perfect symphony and concert of not just can we create a robot and make it move around, can we understand vision, can we do complex path planning, can we perceive what's going on and can we also cause the robot to execute fast enough to be able to hit the shuttlecock before it hits the ground?

Farbod: Absolutely. And there's, you know, thinking back to when I played badminton, I think there's all these um nuances to hitting the ball specifically. For example, as it's coming your way, it's a very light feathery projectile so the breeze can carry it slightly right before it reaches you. Instinctively as a human, you know how to make those microadjustments over time to return the shuttlecock the right way. But for a robot, that's an incredibly difficult task. Because as the path of the shuttlecock makes its way towards the robotic arm, it needs to constantly think through the kinematics of where it's gonna land and how it needs to be returned to the other player. So again, all that to say, tough task that these guys have to do. I'm a big visual learner. This article that we're linking is gonna have a video of the robot playing the game, which I think is an absolute watch. One of the things that really stood out to me is just like a human, the robot, in an attempt to prioritize hitting the shuttlecock, slips every now and then, which goes to show how much it's starting to mimic and prioritize the task just as a human being would. Anything to make sure that you're winning the game or returning the shuttlecock. In terms of the value that this adds, of like the so what, again, we mentioned earlier, this seems trivial, this seems almost nonsensical to have a robot play game, but tying it back to what we were mentioning from Mouser in terms of manufacturing 5.0, Daniel and I visited Ed Mehr at Machina some time ago, I think almost a year ago at this point, where they have these two robots that are working on a piece of sheet metal together, right? So, pressing on either side to create, ideally, a very precise piece of the final product. And the only way that that can happen is if the robots are incredibly precise and in sync with each other. This kind of research development is what fuels applications like that to happen. Same goes for surgical robotics. Same goes, again, for search and rescue robotics. That's kind of the main value that I would want our listener to get out of this, is that, it might not seem that important, but these little baby steps is what enables the cool stuff that we actually see in the real world.

Daniel: And one thing that I think is really interesting is they said this is kind of a, it's not just a high-water mark or a massive new development in terms of playing a sport. Like, that's cool. They said this the first time a legged robot has played a fast sport on its own with no outside help. That feels like a couple of different qualifiers, but it's still really interesting to watch. And I just love watching the video of it playing. But in my mind, the way that they trained it is also very interesting, too. So, I've got this Anymal robot, which we've actually covered before on the podcast. It's a four-legged quadruped robot by Anybotics. And then they also have the Dyna Arm on top of it, which is the multi degree of freedom robotic arm. Those are two separate products and they're often trained separately and they're often operated separately, which is what makes these like combination super complex robotics super challenging is trying to build a model or control policy that moves all of this in perfect symphony. And in this case, they created and trained one control policy to move both the legs and the arms together. Which was also a unique approach. They didn't train just the arm how to do a swing and then train just the legs how to move around. They trained it all at once using a reinforcement learning simulation. Basically, the robot learned by trying over and over and over again, tens of thousands of times in simulation how to move its legs and its arms all together in symphony to hit the shuttlecock in the right time, which I thought was a very ambitious approach of theirs to like try and take all of this at once as opposed to piecing it out into separate movements and or just separate policies. But it also is very similar to the way that we learn how to play a sport in real life is like generally you're gonna go out and try all of the movement with your entire body all at once. And that's how you develop coordination. If someone told you like, oh, here's how you're gonna practice your badminton shot is like. I'm only going to train your legs. So can't move your upper body at all. You're only allowed to move your feet around. And I'm going to train you like that for a couple thousand hours. And then I'm going to train just your arms swinging for a couple thousand hours and then put you out into the real world where you've never practiced both together and say, bang, go do it, go play badminton. I guarantee it wouldn't work that well. But that's kind of how we've taken a lot of approach to different aspects of robotics. So, I appreciate the thought here around how to build coordination and how to play sports, which require a lot of good coordination, involves maintaining one control policy that moves all of the aspects of the robot together.

Farbod: Absolutely. Absolutely. So, to kind of tie things together, you know, you have this big challenge since the past, I don't know, 45-50 years of getting robots to do these tasks where you have to perceive something coming your way, doing the path planning to position yourself the right way and return in a specific format using a specific amount of force and actually execute it in time. And then you have all these researchers that have been trying to tackle it for a really, really long time. And now you have ETH Zurich, who's like, not only can we do it, but we're doing it with a moving robot, not something that's stationary. That's why it's kind of impressive. But in terms of the applications it can have for us, think critical applications like search and rescue, think like robotic surgery remotely, think manufacturing with six-degree axis robotic arms pushing metal together. That's where the value really comes into play for our day-to-day lives or things that the average engineer would care about. Yeah, I don't know anything else to add from your side?

Daniel: No, I mean they mentioned it's got more room to improve, or I think they said their record volley was 10 volleys back and forth. They want to add better cameras or movable head to it because right now the cameras are fixed mounted. So, imagine if you're trying to play badminton but you can't move your eyes or your head or your neck at all. So, it's like if you're completely stiff staring forward you'd have to move your torso around like crazy to be able to see what's going on. And they also mentioned that they're to try and train it soon to watch the human opponent the same way that we would watch the human opponent. So, you can kind of get some prediction on the way the shuttlecock's going to move by watching the trajectory of your opponent's as opposed to waiting to track the shuttlecock in the air. Again, it's more intuitive. That's why the net in badminton is see-through, not just like a brick wall. You're able to see what your opponents are doing and use that as context on how you're going to move. So, it's interesting. They're going to start adding more and more and more signals, more and more complexity to hopefully get to a point where this thing can play just about as well as a human can.

Farbod: Well, I'm excited for the robotic segment of the Olympics whenever this reaches that level of maturity. It's going to be a good time.

Daniel: Yeah, or autonomous battle bots. That'd be fun to watch too.

Farbod: That'd be very fun. And yeah, I think that's the pod, right?

Daniel: Yes, sir.

Farbod: Folks, thanks for listening. Catch you in the next one. Peace.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!

--

The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.