Domino Robot: Creative Indoor Positioning Demo | Marvelmind

0:00 Hello, colleagues. Many of our customers are very special, and they do absolutely amazing and sometimes even crazy things. This is just another very special case, and let me show it to you a bit. He's a friendly little robot that's super good at only one thing: setting up a ton of dominoes, really, really well. So this is the robot as you see it—it's a pretty crazy robot that is doing the domino stuff. And of course, as any robot, it needs positioning. I just came across this on YouTube by accident, and of course I couldn't not point out that there is a Marvelmind beacon for positioning. The guys actually approached us in April saying that—

0:59 You are not helping us well, and we didn't know who they are, et cetera. While I went to Twitter and accidentally found that the guys are pretty prominent on Twitter and even more prominent on YouTube. Yeah, so you can watch the original video. Just a few comments from our side: the robot is, first of all, great. This is the real robot, and I would really recommend everyone to look at how it is built because it is built with many good and essential things in it. Now, first of all, as just mentioned, it is designed for a particularly single task, which is probably the key element for modern robots. If you try to build a robot that would be doing everything for everyone, then most probably it will not do anything for anyone. But one particular thing, tuned for this—

1:57 And it can beat a human. So in this particular case, it easily beat the human in placing these, because it was using a completely different approach in building the dominoes. The robot needs localization. There are several ways you can do this: optical, lidar, and ultrasound. So I'm happy to see that eventually the guys chose us for positioning, and they did it perfectly as I see, because they did it for location and direction. As you see, there's quite a large base between the stationary beacons. If each mobile beacon can give you plus-minus two centimeter accuracy, then with this base you can have, I guess, fewer than one degree accuracy of the robot's position. And since you have a pretty—

2:56 Solid frame, then you will have end-to-end precise location and precise direction at the same time, absolutely, according to the blueprint and how it must be done. Okay, we jumped a bit further, and as you can see, this small dot is not just a dot—it's a Marvelmind stationary beacon. So in order to position the mobile robot, you need a modem, a central controller of the system, one per system, stationary beacons. This one, I guess there must be another one, another one, another one. So for 2D tracking, for example like this, it would be sufficient to have only two mobile beacons, for example here and here, okay. The guys placed them like this, which is also fine. And then beacons on the robot in order to get location and direction. Also, I really like—

3:54 Points the original idea and how this idea over the years was morphed to the end robot. Just watch it exactly: do you reload it and how it did? No, you see? Reloading, moving, placing, localizing, et cetera. All these are separate tasks and doesn't necessarily need to solve them in the same way. So they did really brilliantly on all these ways, and I really like that they didn't try to build in a human way. Because humans are doing one, one by one, by one. Okay, humans have better fingers and better capabilities in many ways, but they cannot do 300 pieces at once. But the robot can do that. So they did it perfectly from this direction as well. Exactly where to drop each domino and—

4:54 What color it should be and how do you make it so reliable that it doesn't screw up once in a hundred thousand drops of a domino? A system to reliably tackle issues at scales like this is just going to be inherently super complex. Here we were. Now, before I show you how it all works, I first wanted to put them in a head-to-head competition to see how good. Here you see they do employ ready-to-use stuff, and that's another great element. Like many of our other customers, you know, if you need positioning, of course you can reinvent the wheel, and sometimes that's good, and it's very good. But sometimes when you build something more complex like the whole robot, it's easier to take navigation from us, for example, than the camera from, let's say, another supplier. Then they had the lovely motors, controllers, and then they built the system. Yeah, it may not look pretty from a layman's point of view, but from my—

5:53 Point of view, it's absolutely excellent and brilliant robot. Oh my god, wow! And look, so this is the mobile deck, and this is the mobile beacon. And once again, spot the stationary beacons on the walls. Dominoes all at once was really the key that helped us crack the scale and reliability issues, and it's probably my favorite part. I love it. I really love it. The guys did it like it really must be done. So this is the thing which does the job, and it's done not in the human way—it's done in the robotics way, and separately, by the way. So there is a separate machine which is putting, loading, you know, the massive 300 pieces at once. Great! So they're using wheels again—

6:51 Special wheels which allow them to turn at one spot and move in different directions. You know, for the robot, you don't need a nice chassis. You need an easy chassis because you need to do many, many, many adjustments. You know, don't limit yourself. Put huge batteries and don't worry about them at all. Of course, you know, use something ready like this one, like this one, like motor controllers, cameras, and then you know, assemble the robot and that's it. And then you can tune. Then they can polish. They can do something. But don't limit yourself right from the beginning. Do, or use as proven systems as possible in the beginning at least, and then you can polish something a bit later. You'll have to see the beacons, okay? So you can see. So this is a mobile beacon, and this is a stationary beacon, and I'm talking about beacons on the robot. You have seen already a few times, but—

7:50 This is much faster because he doesn't have to wait. He just comes in here to the docking station, and the lower platform slides over. So the bottom layer of 300 dominoes gets loaded up all at once. And we also had a bot, which is great. This is absolutely great. You see? They're loading and driving was a separate thing, just meant that while the robot, the mobile robot, is placing the dominoes and driving around, and their, their—the other robot—can do the parallel. That's great because humans cannot do more than one. A human can do this, but if it's a single human, a human can do either placing dominoes or packing or moving, et cetera. The robots can do this in parallel, which, you know, speeds up significantly. Let's switch further back. Uploading system using a tray. Just in case, at any point the robot arm wasn't working. Besides the hot—

8:50 Wheels, tracks. There's a ton of 3D-printed parts throughout the build that we either printed ourselves, or if we were in a pinch, no. As I mentioned, that's absolutely a great approach. This is what we do as well. You know, currently as I speak, behind me there's a printer, and a couple more in front of me. Easy, flexible. Use them as we do. Friends at Matterhackers helped us out. So that's the loader. Now, how about the dominator himself? How does Dom know exactly where to go in the room to drop a domino? So we, and that is the key point about Marvelmind. It was a short one, but pretty essential, frankly speaking. Because in one way or another you do need to know the location. Again, I would say it could be us, it could be lidar, or it could be optical plus odometry. But since they're using these—

9:49 special wheels. I guess the odometry was not as precise, and optical plus automatic may be not precise enough. But many robots—they do, you know, Kiva robots which became Amazon robots—they're using, and majority of other robots are using this optical plus odometry way of positioning, which is great sometimes, but sometimes not, because not always you can place QR codes or some other optical elements on the floor in order to eliminate the accumulated errors by their odometer. And often these optical labels, they are wearing out, so you need to do something with them. You cannot do this, or let's say you cannot place the QR codes above, because then you don't need like

10:48 on the ceiling. Then you don't need enough accuracy. So this is why there are not too many solutions that give centimeter-level accuracy. And Marvelmind, of course, is great, particularly for precise tracking. Pre-programmed the route for all 102,000 dominoes, so the robot knows exactly where to go right from the start. Then as we're driving around, we use these indoor GPS sensors to track the position of the robot so it knows exactly where it is. And then as we get closer to the place the dominoes need to drop, we use these IR cameras, which is absolutely great. So there's no point for, let's say, trying to get something which you can achieve closer, like optical, for example. So sensor fusion—sensor fusion, sensor fusion—is the key. And that's a great sense of fusion between odometry. Maybe they're using, I don't know, the

11:46 details, us and optical. Because optical, the closer you come, there, higher the accuracy you can get. And you can adjust against, not absolute coordinates of the room. You don't need. You need to adjust against their local things, like objects, like domino pieces. So this is why the combination between several sensors, or several inputs, is there. Is the best approach that are tracking markers on the ground to make sure the robot lines up perfectly every single time. So the vision for reaches again one more time: a great approach. Use a sensor combination. Combine different types of sensory inputs, because none of them are good enough for all the cases. Which is the best for your practical case? It depends on the case. For people thinking, it's one story. For their own, another story. For virtual reality, third. For robots of this type, is fourth.

12:46 Forklifts, boats—you know, humans, they have very, very many different ways, and the approach is slightly different. But their approach in terms of sensor fusion is common. It's only a combination of what sensor inputs you can use. Is that we can set up? We could turn off the lights and leave and come back the next morning, and you've got like a full field of dominoes set up. And that is the most, you know, rewarding—that you can do something really autonomous. And this autonomous, you give the task and forget. I know. And then robot just drives. You: Yes. You do monitor? Yes, you do. You control it. But if everything is set up and the task is not interrupted by something, then you just come and the job is done. It is very satisfying to work all night in the dark, just fine. All right. Now, how about these super cool wheels? So these are called—yeah, those are absolutely perfect wheels. We are not using them

13:45 because usually they are a bit expensive and they work pretty well on these kinds of surfaces. But for their particular case, absolutely brilliant and exactly what must be used: omni wheels. And they're awesome because they let you translate in any direction you want. So this is way better than like your car, where if you need to move a little bit to the left or right, you have to make like a five-point turn. So with these guys, you can move any direction you want to adjust for small corrections in the placement of the dominoes. Yeah, that's important task because this is why we are using, you know, tank-like rotation. So our robots are usually can turn on the point exactly, because otherwise you would need to move many times back and forth, left and right, and often you don't have their space. So in that case, they can really move in each direction. In our case, you can turn and move a bit for

14:44 forward or backward. It's on the spot, so it's not like on the car. It's like a tank. But with omni wheels, it's even easy, because you can move from any direction using this, you know, special. But limitation is also obvious. So it is relatively expensive and it cannot drive on, let's say, all conditions like industrial conditions. But otherwise, the robot is great. Let me slow down and finish. Absolutely great. I'm happy to present that our customers are building absolutely amazing stuff, and I hope you will build something like this or even more fancy. Thank you very much.