Jon Krohn: 00:00
This is episode number 529 with Dave Niewinski, owner and host of Dave’s Armoury.
Jon Krohn: 00:14
Welcome to the SuperDataScience podcast. My name is Jon Krohn, a chief data scientist and best-selling author on deep learning. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. Thanks for being here today and now let’s make the complex simple.
Jon Krohn: 00:44
Welcome back to the SuperDataScience podcast. Today’s guest is the mad robotics genius, Dave Niewinski. Dave has a YouTube channel called Dave’s Armoury with over 10,000 subscribers. On his channel, he publishes videos of his robot creations, many of which feature deep learning AI algorithms in them to make them effective. And he publishes videos of his robot creations in order to teach people what’s possible with robotics today, including by providing the details of his hardware and software engineering approaches. Some of my favorite robots of his include one that exterminates weeds in his backyard using a flame thrower, one that brings cold beer from the fridge to wherever he is in his house, and one that keeps his kids out of his machine shop by finding them and hosing them with water. In addition to his enthralling Dave’s Armoury YouTube channel, Dave founded and owns a consulting business of the same name that allows him to apply his robotics expertise to automate and improve real-world industrial processes.
Jon Krohn: 01:46
In today’s episode, Dave details the specific robotics hardware and open-source software he incorporated into his wildest and most famous robots where machine vision algorithms, particularly deep learning models are critical for enabling robot functionality, his tips for folks who’d like to get it started in AI robotics themselves, and what excites him most about the societal impact AI robotics will have in our lifetimes. This episode will appeal to anyone who’s interested in discovering what’s possible with AI robotics today via a fun, lighthearted conversation. We do mention specific technical approaches here and there, but typically only briefly during higher level descriptions of robotics capabilities. And we do our best to break those technical terms down whenever we do use them. All right, you ready for this? Let’s do it.
Jon Krohn: 02:43
Dave, welcome to the SuperDataScience podcast. It’s awesome to have you here. Where in the world are you calling in from?
Dave Niewinski: 02:52
I’m in Canada, just outside of Waterloo, which is a small town just outside of Toronto.
Jon Krohn: 02:58
Well, as it happens, I know Waterloo very well because I went to high school and undergrad in Waterloo, Ontario. And I learned in talking to you just before we started filming that you actually live in a very small town outside of Waterloo, so small, I counted the streets on Google maps as we were talking, there were 12 of them and they’re all about a block or too long.
Dave Niewinski: 03:21
Yeah. Very, very small.
Jon Krohn: 03:23
Yeah. And yeah, so I wasn’t embarrassed after I saw how small it was that I hadn’t heard of this town, but yeah, love that area. People in Waterloo region are so nice. When I come home from New York I’m just… I’m at ease around everyone. Everyone seems to be happy. Everyone wants to make your day easier.
Dave Niewinski: 03:46
They’re very nice [crosstalk 00:03:48]. Terrible drivers in our area. It’s just the worst drivers.
Jon Krohn: 03:53
Oh man. Yeah. But well, yeah, the niceness counts for a lot. I’ll forgive them on the driving. So I was introduced to you by a mutual friend of ours, Graham McCormick, who is-
Dave Niewinski: 04:11
Hi, Graham.
Jon Krohn: 04:13
Who is the Corporate Engagement Officer at Wilfrid Laurier University, my undergraduate alma mater. And when he… So he introduced us over email. I went immediately to your YouTube channel, Dave’s Armory and I was blown away by your creativity, your execution. I instantly replied and was like, “Yes, we’ve got to get Dave on the SuperDataScience podcast right away. People are going to love to hear about what he’s working on.” So your YouTube channel has 10,000 subscribers and they come to see your incredible videos with a huge amount of technical detail from a hardware aspect, from software aspects on building AI powered robots. And so we’re going to talk about a bunch of these in detail, but it’s things like a Rubik’s cube solving robot, a weed exterminating robot that uses a flame thrower to get rid of the weeds, a robot that brings you cold beer on demand and one that races against your daughter to build a robot unicorn out of Lego. I mean, there’s just such a diversity of ideas. I can’t wait to get inside of your brain. Let’s start with one of these robots that I was just watching the video about and this is actually at time of recording your most popular video is a cornhole playing robot. So for our listeners outside of North America, cornhole is a game where you have these little like sandbags, I guess they’re supposed to be full with corn and you throw them a short distance, like 10 or 15 feet into a little hole in a wood board and yeah, that’s cornhole. So you have typically-
Dave Niewinski: 06:11
Bean bag toss.
Jon Krohn: 06:12
… so you have two… Bean bag toss, exactly. You’re just trying to get a little bean bag into a whole at a not arduous distance. It’s very popular at weddings. It’s a popular drinking game.
Dave Niewinski: 06:22
Yeah, you do not have to be sober to play.
Jon Krohn: 06:27
Probably, yeah, almost nobody sober does play. Though I learned a lot about it in your video about the corn playing robot. So that there’s a professional league. I wonder if they train-
Dave Niewinski: 06:37
Yeah, it’s a whole league for it. It’s crazy.
Jon Krohn: 06:40
The American Cornhole league, I guess. I absolutely love this video and it’s a good example of the level of detail that you go into in all of your videos I imagine. So you start off by explaining the cornhole game and then you start off by getting a robot to cut you the boards that you need to play cornhole. And it’s the same robot, Susan, that also later is your teammate in a cornhole game. So I thought that was cool in and of itself. So I mean, so tell us about it. So first you start off by, so you program it to cut the holes into the wood boards. How do you do that?
Dave Niewinski: 07:30
So a lot of arms are used in manufacturing. So cutting something like wood is pretty easily or pretty easy to do or something like aluminum. So I’d like to show the arms doing common things, like common arm things for the industrial audience, but at the same time, arms are super flexible. So there is a ton of room for arms to grow into, robots in general to grow into a ton of different areas that they’re not commonly used in. So once Susan’s done cutting the boards, I clean them up and turn them into cornhole boards, but then using a Jetson, which is basically if a video card and a raspberry pie had a baby. So using the Jetson to be able to use OpenCV to be able to find the hole in the cornhole board no matter where it is in front of it, and then always throw the bean bag perfectly every time to just drain shots.
Jon Krohn: 08:29
Awesome. So OpenCV, that is a software library. You can use it in Python to perform. So the CV in that stands for computer vision. So it’s an open computer vision library. And so you used it to… Did you have to do any training with a neural network or it kind of just worked?
Dave Niewinski: 08:49
So not for OpenCV. So OpenCV is visual processing. So you can pick out like specific colors or shapes or regions, you can do processing on the video or on the-
Jon Krohn: 09:02
Cool.
Dave Niewinski: 09:02
Frames themselves as they’re coming in from the video. So none of that was using AI. That’s more, I don’t want to say traditional kind of programming, but it’s more scripting language, less AI in training.
Jon Krohn: 09:17
Cool. But it still takes advantage of yeah, this computer vision library.
Dave Niewinski: 09:21
It still uses CUDA and GPU to do parallel processing really fast, which is why you need something like a Jetson or a video card to be able to do real time work with it.
Jon Krohn: 09:31
Got it. Got it. Got it. Yeah. So CUDA for listeners who don’t know it, a popular programming language for programming graphics processing units and graphics processing units or GPUs, which were originally primarily used to render graphics in computers. So if you’re playing a 3D video game or you are editing video footage on your computer, you need graphics cards to do that work efficiently. But that same kind of computation, these simple linear algebra operations that happen to render those 3D graphics are the same kinds of mathematical operations that we need to train neural networks including neural networks that we use for machine vision like convolutional neural networks. And so, yeah, so you can parallelize computation in real time like you’re saying on GPUs and like the Jetson that you just mentioned that I had never even heard of before.
Jon Krohn: 10:29
So it’s kind of like a lightweight, low cost relatively piece of hardware that you can just put right on the robot and in real time be doing computer vision. Cool. Wow. So you use OpenCV and the Jetson on Susan, this robot arm, and then you then modeled the physics of, okay, you detect where the hole is with OpenCV and then you say, okay, that tells me how far away the hole is. And then you programmed in the physics, the kind of trajectory that the bag would need to take and Susan, just executes that for you? It seems really hard to me.
Dave Niewinski: 11:16
Well, it’s like grade 12 physics, or probably grade 12 where it’s acceleration and velocity and all of those normal things where you get the trajectory of the bag and the arm itself because it’s an arm can do really repeatable, really accurate motions over and over again. So as long as you can tell the arm exactly the motion that it needs the bag to do, or it needs the end of the arm to do, then you should just be draining shots as long as long as you can figure out where it needs to go.
Jon Krohn: 11:52
Cool.
Jon Krohn: 11:55
This episode is brought to you by SuperDataScience, our online membership platform for learning data science at any level. Yes, the platform is called SuperDataScience. It’s the namesake of this very podcast. In the platform, you’ll discover all of our 50 plus courses, which together provide over 300 hours of content with new courses being added on average once per month. All of that and more you get as part of your membership at SuperDataScience. So don’t hold off, sign up today at www.www.superdatascience.com. Secure your membership and take your data science skills to the next level.
Jon Krohn: 12:32
Nice. So then once you had the robot ready to go, you hosted your own little tournament and it looks… So Susan was your teammate in this cornhole game. Tell us about it. How did it go?
Dave Niewinski: 12:47
Well, we won in some of the testing. So you need to load the bag into the end of Susan. Yeah. So once we had the robot all set up, it was Susan and I against James and Ben. And so we were throwing back and forth and Susan drained every single shot. And then right near the end, James, on the other team, of course, I had him sticking the bean bags in the gripper on Susan. And he deliberately started putting them in the wrong way so that Susan wasn’t throwing them consistently anymore. So she started to miss some shots by the end, but we still won.
Jon Krohn: 13:22
Oh, nice. And James himself is pretty well known on YouTube.
Dave Niewinski: 13:29
Yeah. So he’s the hacksmith from Hacksmith industries. I ended up meeting him through a whole other series of events, but he has a large facility kind of where I live that has 3 phase 600 volt power, which you need for an arm of that size or bigger. And I don’t have that at home because Canada. So I was able to put Susan at his shop and be able to do a bunch of work there, which was great. So he does a lot of videos making like props and real world examples of things that you see in movies like Marvel, Star Wars, all that kind of stuff.
Jon Krohn: 14:12
Cool. Yeah. So I mean kind of tangential, but that also sounds like a pretty cool channel for our listeners to check out. All right. So we’ve got the cornhole playing robot. Clearly you nailed that one. Susan was able to crush your opponents with you. Another really cool video was your shop defense robot. So tell us about that one.
Dave Niewinski: 14:33
So I have two kids. They’re two and four now. I like to try to include them in either videos or just work that I’m doing whenever I can just to try to teach them stuff. So I’ve made… But I don’t want them in the shop when I’m not there because saws and hammers and all that stuff. So instead of just locking the door like a normal person, I got an arm and a stereo camera from Zed 2 from Stereolabs and another Jetson to be able to find the kids in front of the robot using AI and then take a hose and spray them like cats just to keep them away.
Jon Krohn: 15:14
Oh, that’s great. So in that kind of case for the machine vision aspect in particular, how did you program that with this one?
Dave Niewinski: 15:25
So that’s two different things kind of happening at once. The first one is using an AI that can detect people. So like a normal, something like a COCO data set you can use to train that. I didn’t actually train that AI, that came with the Zed SDK. So the stereo camera comes with some example AIs that you can use. So I use that instead training my own. But then once the video comes in and it finds the person, it then using the stereo camera can get a 3D map of the world in front of it, kind of like how your eyes do so it can take where it found the person, find where they are in that 3D map. And then it knows exactly where that person is in front of the robot, not just they’re there or they’re not, it knows they’re a meter ahead and slightly to the side. Then once their arm knows where they are, it can point at them with the hose and go to town.
Jon Krohn: 16:27
Cool. Well, that’s another really cool application. So we get your shop defense robot. Another one that I loved and it really shows us kind of how your mind works is you built a weed exterminating robot. So step one, okay, we’ve got to have some kind of computer vision system so that the robot can detect where weeds are on the grass.
Dave Niewinski: 16:52
So that has an arm that has a stereo camera built into the end of it. So kind of like the Zed, but it’s all together. So I put that on a mobile robot, which is basically a tank and then a big propane tank and a flame thrower so for that in each video, I try to kind of highlight a specific piece of technology to try to cover it all in detail in a single video, no one would ever watch them. But so in that one, I looked a lot more at how to actually train an AI how to record and data and how to use that data to actually do the training. So in that one I recorded a quick video of my lawn, I segmented out the individual weeds, then trained a network on Google Colaboratory, which is it’s a Jupyter notebook, but in Google and they have access, free access to GPUs. So you can be doing a lot of this AI training without needing to maintain or buy any of your own hardware, which was awesome.
Jon Krohn: 18:06
I love Colab on that. I use Colab for all of my online teaching because it’s so easy for me to just say, here’s the URL, let’s get going. You don’t have to install anything.
Dave Niewinski: 18:17
Well exactly. So I drive the robot around, it would point the camera at the ground. And if it saw a weed that the AI detected a weed, then it would turn the flame thrower on and burn right out of there. It was very therapeutic for me. A lot of the weeds actually came back through the dead spots. So there would just be a dead spot in the lawn with a weed in the middle, which felt pretty bad. But just watching it happen felt good for that brief moment.
Jon Krohn: 18:51
Yeah. That’s funny. And yeah. And then if the shop defense robot, if spraying water at your kids doesn’t work well enough, you’ve always got the flame thrower of it.
Dave Niewinski: 19:01
Absolutely.
Jon Krohn: 19:02
All right. So those are good ones. Another one I loved is the pumpkin carving robot. So at the time of filming, we’ve just finished Halloween. So this pumpkin carving robot, yeah, I guess that could also kind of work as a shop defense robot. You put a serrated knife on the end of the robot arm.
Dave Niewinski: 19:22
Yeah. There’s a couple different ways that you can obviously cut things. So in that video I used a knife, which is not a good way to be cutting things with a robot and Dremel which you could do like fine details with the Dremel, but it’s not as stabby horrory as a robot with a blade.
Jon Krohn: 19:41
So what’s a typical way that you would use, it would be some kind of saw?
Dave Niewinski: 19:45
Like to cut a pumpkin?
Jon Krohn: 19:47
Well, I mean, yeah. So for example, when you cut the cornhole boards, what do you use there on the robot arms? So like you said that a knife and the stabbing action is like not a great use of a robot arm. So what would you typically use?
Dave Niewinski: 20:02
Well, normally you use some kind of a spindle, right? A router or a spindle is basically like a drill sort of. It just spins it and then the robot moves it around and the spinning bit actually does the cutting. There are lots of cases though where you would use something like a knife. Ultrasonic cutting is a big one in like the plastics industry, like just trimming thermoform plastics or anything backing formed like car mats. So it’s a little blade that stabs in and out like 10,000 times a second and you can just rip around because the arm can position it at any angle so you can get lots of fine details around objects.
Jon Krohn: 20:45
Wow. That’s cool. All right, but that wasn’t our scenario here. You’ve just got a pumpkin. But you were like-
Dave Niewinski: 20:52
Yeah. And a knife I found at Value Village that I sharpened so much that it was too sharp.
Jon Krohn: 21:00
Nice. And for our listeners outside Canada, Value Village is a chain of thrift stores. So you can bring anything to Value Village, well, not anything, but you bring used clothing, used knives, household stuff, and then other people can go and buy it typically very cheaply. So yes, you got your Value Village knife, sharpen it up, and then-
Dave Niewinski: 21:23
Is Value Village like a Canadian only thing?
Jon Krohn: 21:26
I don’t know. I must admit since I’ve been in New York, I haven’t done much thrift shopping. We do have Goodwills. I see Goodwills around. But I haven’t seen a Value Village. I think it might be Canadian.
Dave Niewinski: 21:37
Okay. Yeah. Goodwill. Same thing.
Jon Krohn: 21:39
Yeah. Yeah. So, yeah, all right. So then what was tricky about that I guess is that similar to the kind of scenario, do you have to do any machine vision to like detect where the pumpkin is or do you kind of, yeah…
Dave Niewinski: 21:55
So I didn’t actually use any sort of AI or machine vision for that. That was more offline programming. So programming it in your computer without the robot, which makes the robot usually more accurate and more productive because you’re not taking the machine down to program it. Or for that actually I did a bunch of manual program with the UR I was using like the universal robot arm that I was using. You can grab it and move it around so it has hand guiding built in, so instead of just using the controller, you could physically grab the robot and position it where you want it. So that’s a good way for anyone who’s getting into robotics or newer users, it’s a lot easier for you just to say, robot go here, and you could grab the robot and move it.
Jon Krohn: 22:48
Cool. I am learning so much. I suspect our listeners are as well. All right. So just a few more. I could probably go through all of your videos and this would be super interesting, but we’re just trying to pick some of the most different, most interesting videos. So another one was a Rubik’s cube solving robot. Whoa. That’s something… There’s been in the last couple of years, there have been big papers in the industry from like big research outfits, I think like OpenAI had a big paper about a Rubik’s cube solving robot. And so you did it too.
Dave Niewinski: 23:26
So the algorithm for solving it, I just used a Python library that already existed. I didn’t write that myself. But being able to use the arm to with just a single arm and no gripper actually to just be able to pick up and reorient the arm or reorient the cube and solve it by hand. I raced it. I won, but that was also some really inefficient programming on my part. So it’s kind of cheating.
Jon Krohn: 23:55
That’s good. Like I’m definitely going to win this race so better put, so yeah. So that’s different from the situation where you’re playing cornhole where you knew that Susan was going to be on your own team.
Dave Niewinski: 24:07
Yeah. So I wanted her to do really well. Whereas it was me against Herman, Herman was the UR that I was using for that. Yeah, I didn’t really worry about programming it super efficiently. It’s more, yeah, so again, in each video I try to show kind of show a different benefit to arms or a different way you can do it. And being able to have the arm literally pick up the cube, look at it, figure out a solution and then do it, that’s more beneficial or that’s more the benefit that I was trying to show that you could put functionally almost anything in front of the robot and it can figure out a new path, a new solution to that problem versus something like hard automation, which only does one thing.
Jon Krohn: 24:57
Yeah. Yeah. Super cool. Yeah. And I feel like I’m starting to get familiar with your industry terminology. Terms like UR, when you throw them out there, I know exactly what you mean, a universal robot arm tool. All right. So another one, you got two robots, two different videos that could work with Lego. So one of them created Lego art. It created a, not a painting because it was Lego art, but something that looks like a painting of Darth Vader.
Dave Niewinski: 25:31
So yeah. Lego released an art series. They’ve got ones of The Beatles. They’ve got, I think a Marvel one, they’ve got some Star Wars ones. It’s really repetitive. So it’s literally pick up a little pip and stick it on I think a few thousand times, which is perfect for arms, just doing a simple task over and over and over again. But to program that manually would be a nightmare. I just programmed it with a simple Python script, which is on GitHub if anyone’s curious.
Jon Krohn: 26:04
Oh cool. Yeah. Great. We’ll be sure to include that in the show notes. And all right, so then you said that it was simple to have this highly repetitive task and okay, that makes sense to me. So placing these pips the right color in the right place, but something that sounds more complicated then is doing something like building a robot unicorn, but you did have a robot arm do that and race against your daughter. So how’d that go?
Dave Niewinski: 26:32
Herman won. My kid lost publicly on the internet. It’s a real strength building kind of thing. It’s real character moment for her. That one was using a gripper in a force torque sensor. So actually being able to grab the piece, know when it grabbed the piece or if it missed, and then when it’s putting in, push with a certain force in certain directions, because just having the pieces snapped together is surprisingly hard with Lego. It’s easy to do if you’re doing it by hand. But if you’re trying to do it blind and stick pieces together, you need to be able to feel the feedback when the piece is actually [inaudible 00:27:12] to one another.
Jon Krohn: 27:13
Wow. Yeah. So it sounds like that might have been the focus of that video explaining that part. Yeah. Cool. All right. And then just one last robot to round things off. This is one really after my heart because I don’t drink very much these days, but I love beer, Dave. It is the only… People often talk about wine in a lot of detail. People seem to know a lot about wine. I don’t know anything about wine. But I know a lot about beer. And so you have a robot that can bring you cold beer on demand. Wow.
Dave Niewinski: 27:52
Yeah. So that’s using a mobile robot from Clear Path Robotics. So it can drive around on its own and it has a LIDAR on it, which is a laser that spins around and it can see the world around it. Using that, it can make a map of my house, drive around my house on its own. I paired that with Google assistant or Google home. So I could say, Hey Google bring me a beer. It would wake up the robot, drive the robot to the fridge. Then the robot would use an arm with a gripper, open the fridge, pull out a beer, close the fridge and drive it to wherever I told it to go in the house.
Jon Krohn: 28:32
Wow. That’s cool. All right. So there’s eight examples for your listeners of some of the incredibly diverse and creative different kind of robots that Dave has built. You can get all of those on the Dave’s Armoury YouTube channel. So Dave, how do you come up with these ideas? Do you have a process or do you just… You’re just like, I’m annoyed by the weeds on my lawn. I’m going to get rid of them, I’m going to find a way to do it with a robot.
Dave Niewinski: 29:08
I don’t really have a process for coming up with videos, I sort of just see something and think that an arm or something could do that pretty well and then try to figure out what sort of technology I can highlight while doing it again. Trying to sort of teach people, but at the same time, I do have a sort of just a grid for thinking of what videos I should actually do. Because coming up with ideas is not so much the problem, it’s weeding out all of the bad ones. So trying to find videos that are relatable because I want people to not be afraid of robots. I want them to understand that they have a place, they have benefits and challenges, they’re not great for everything. They’re good for some other stuff.
Jon Krohn: 29:58
Don’t be afraid of my stabbing flame throwing robots. Nothing to be afraid of.
Dave Niewinski: 30:02
Yes. Don’t. They’re friendly. You want them to trust you. So I try to find videos that are somewhat relatable because I want people to be engaged and interested and comfortable with robots. But I also want the video to have some sort of challenge to it and be using a robot in a way that’s not a common use. Right? You can go on YouTube and you can see robots building cars or robots welding or robots assembling or any of those are very common. And there’s lots of examples of that. I want to show robots doing things in very uncommon spaces. For example, yeah, beer delivery or weed murdering.
Jon Krohn: 30:51
Nice. And it opens my mind up. I kind of, and I’m later in the episode I’m going to kind of ask you about what kind of education someone needs to be able to build some of these kinds of robots. And it’s going to be interesting to learn about that, but I kind of… Meeting you and seeing your YouTube channel, it makes me believe that like I could be creative with robots. There’s things I could do. Whereas previously, like I kind of thought, as you say, when I see a welding robot or a car building robot that seems like something unapproachable to me for some reason, but because these are tangible, kind of everyday applications of kinds of robots that would be cool in my life, yeah, I mean your YouTube channel has certainly had that effect on me, so I’m sure it has on other people as well. So I think you’re…
Dave Niewinski: 31:42
That’s great to hear. I’m glad.
Jon Krohn: 31:43
Yeah. All right, Dave, so dare I ask, what are some of the robots that we might see on your YouTube channel in the future?
Dave Niewinski: 31:55
So from a hardware standpoint, I’m still doing a bunch of work with KUKA. So I have another KUKA actually coming soon.
Jon Krohn: 32:05
What was KUKA used for?
Dave Niewinski: 32:06
So KUKA was Susan. So that was a KR20, which had a big controller that I couldn’t run at home. So I think the one that is coming next is either going to be a KR6 or a K10. So like a six or 10 kilogram payload. It’s more of a desk size robot that you can run on just normal 110. I’m actually working on, and depending when this video or this podcast comes out, I may have already released this video or I might not if I didn’t get to it. I’m working on a video about how to make an AI for a small arm that will find my cat and then use a laser pointer to play with my cat. So that’s like AI and a stereo camera and ROS and arms and [inaudible 00:32:59] kind of all that stuff. But again, everyone plays with their cat, right? So that’s coming. I want to make an autonomous lawnmower here using not a lawnmower of course, and it because Canada, I also want to try to make a robotic snowblower so I can get out of my driveway, but we’ll see if I get to that.
Jon Krohn: 33:22
Wow. Those are a lot of cool ideas. They all sound like classic Dave’s Armoury. Brilliant. So Dave’s Armoury the YouTube channel is not the only application of Dave’s Armoury. You also have a consulting business of the same name. And so tell us a bit about that, about like what you do with the consulting business, how you help people out.
Dave Niewinski: 33:47
Sure. So I started posting these videos on YouTube evenings of weekends, whenever I would get free time as limited as that is. But a bunch of people have seen those videos and reached out to me through Instagram or Twitter or wherever, or been local businesses in the area and said, oh well, I’ve been thinking about how to improve our manufacturing process or I’ve been thinking about how would I make a robot that drives around and does X. And so I end up doing a lot of consulting on the side for helping companies develop new technologies or develop new robots that fit with their current business or wherever their business is planning on going or helping companies try see where they could be using automation in their manufacturing, whether it’s packaging or something like that. Or just explaining what I would consider to be very simple things about robots that I’m seeing is not commonly known. For example, integrators, right? A robot integrator is a company that will look at your facility, look what you need to do, then pick which robots and hardware and safety and all of that you need. They put together a custom system for you and deliver it. And then that works in your facility. To me, I thought that was everyone knew about robot integrators. Apparently that’s not true. Surprisingly, unless you’ve worked with a robot integrator, surprisingly few people know that these companies exist.
Jon Krohn: 35:35
Super cool. All right. So is there a particular project or two that you’d be able to share that could be interesting for us?
Dave Niewinski: 35:45
I’m working on a project right now to make an autonomous lawnmower, a self-driving lawnmower. So in one of my past lives, I did mobile robotics using ROS, which is a framework for building robot software. It has tons of libraries, tons of tutorials. I would absolutely recommend it for anyone who’s looking to get into robotics or at least poke around or try stuff. IT’s ROS robot operating system, not an operating system. But so using some software kind of like that to be able to take an electric mower and have it drive itself around and mow a lawn for you.
Jon Krohn: 36:36
Cool. Yeah, that sounds very much like, yeah, I mean, that sounds like it could be one of your YouTube videos. So that’s cool that exactly the kind of stuff that you share with the world to educate us and maybe spark our own creativity about what we could be doing with robots in the real world, you’re also being able to directly leverage that to be making a real impact in the industry. Cool. All right. So Dave, we’ve talked about some of the kinds of machine learning software libraries that you’ve used. You’ve talked about Python, you’ve talked about OpenCV. You’ve talked about GitHub. You’ve talked about Robot OS. Are there any other particular tools, I guess particularly maybe software or data science related tools that might interest our listeners?
Dave Niewinski: 37:23
Well, we talked about Colaboratory, which is awesome. There’s a website called Roboflow that I actually found, I don’t know, a couple months ago. I used it in the weed killer video. That is actually a great website for doing like, you feed it in images, you can mark them out where they are, you can set up your different classes and it has built in augmentation too. So you can just say generate four times as many versions of pictures and it will just take them increase and decrease the contrast, do all that stuff. And you can do that a lot of different ways, but it’s just super easy on that. Jetson would probably be actually my biggest recommendation for something like this. You can get a Jetson Nano for like a hundred bucks, I think. And it’s a full on little computer. It has IO, it has CUDA cores. You can use it for machine learning. It’s low power. It’s cheap. So that’s an awesome way to start with any sort of this stuff. I had also said ROS, which is, again, there’s a simulation. So anyone who wants to do robotics, but doesn’t have a robot, which is I would assume most people, you can run simulations in it so you can functionally have a robot without needing the hardware or really needing one. That’s a great way to start.
Jon Krohn: 38:49
Awesome. So tell us a bit about how you got started in all this. So we know about kind of where you are now, you have this flourishing YouTube channel, you have this consulting business, how did you end up here?
Dave Niewinski: 39:04
So I went to university for engineering. Between semesters in university, I worked at a robot integrator. So just doing wiring and painting and not so much software, but a lot more of kind of the hardware install stuff. I went from there to doing offline programming software for robots. So that’s having a basic simulation in your computer that’s creating programs for the robot directly from the CAD model of whatever of part you want to create. And then taking that simulation, that data, and having a real world robot run it. It’s faster to program. It makes your robot far more efficient. You can start running one off parts on your robot instead of needing 5,000 parts before it’s worthwhile. That got me very close with a lot of the robot manufacturers around, so I would go on site because I had to troubleshoot and work through a lot of the hardware, understand a lot of the hardware and the software inside of the robot before I could make any of that work. And that was kind of the bedrock for a lot of where I started from there. Then I went into mobile robotics, which is ROS and AI. It’s instead of making a program then telling the robot to run it, it’s a much tighter connection between the two where you’re receiving sensor data live, you’re processing it, you’re coming up with solutions, answers, and then driving the hardware directly. So that’s usually a closed loop live control of the robot. And that got me into a lot of different AI stuff. I did a bunch of wiring and software and that kind of stuff in school. So sort of a melding of all of those different things sort of got me here, which is also sort of a melding of a whole bunch of different things that turns into fire robots and kid defense.
Jon Krohn: 41:15
Cool. That was a really great summary. Following on from that, what would you recommend to somebody who wants to get started out building their own AI robots? I mean, you already talked about something. So you talked about ROS, you talked about Roboflow, Jetson Nano, obviously those kinds of things, but I mean, in terms of an education, you have an engineering background, does somebody need an engineering background? I guess probably a lot of our listeners they already have some familiarity with programming. So if you already have a little bit of programming experience, what should you do next?
Dave Niewinski: 41:55
So in my videos, I use different cameras, different power supplies, different programming languages, different networking stuff, different arms, different AI, there’s a ton of different pieces that can all come together in a robot system. So to work, you don’t have to know robot arms to have a significant input in like a shop defense robot. You wouldn’t need to necessarily know arms to do a lot of that as long as you can do AI or you know cameras really well or you understand ROS. So, I mean, normally in most cases, as like an employee in the company, you wouldn’t be doing all 10 pieces of this. You’d be doing maybe the AI and the vision or maybe the arm control. So I would say don’t worry about trying to do all of it. Pick a piece that you really like. If you really like AI, that is a huge stepping stone to making a robot smarter. Or if you really like wiring, learn how to do wiring and some basic a control stuff so you can make grippers or different attachments to arms or different control systems that communicate between devices. And you can again use that to make robots smarter or to make your own robots down the road with motors and controllers and wiring and [inaudible 00:43:38] and stuff. I know that was a really long-winded answer of just kind of do what you like and it’s a step in the right direction.
Jon Krohn: 43:48
Dave, it was a beautiful answer. I was actually like in awe the entire time you were saying that. It was perfect.
Dave Niewinski: 43:53
I’d like to thank my mom, my dad, the academy.
Jon Krohn: 43:55
Cool. All right. So Dave, do you have some master plan with Dave’s Armoury? We’ve got snow blowing robots coming out maybe soon. I mean, what’s the like big vision if any?
Dave Niewinski: 44:17
I mean my long term goal with all of this I think would be for it to become self-sustaining enough where I can wake up in the morning, go into my shop, dick around with robots all day and then go home. Just the freedom to be able to any project I want to make, regardless of whether I think it’s going to be good on YouTube or not, or regardless of whether I have a specific project, just that freedom to be able to make whatever I want with whatever hardware I have. I like that flexibility.
Jon Krohn: 44:58
Cool. That’s an awesome dream. And I believe you can achieve it.
Dave Niewinski: 45:02
And beer.
Jon Krohn: 45:02
You are, yeah, I need a whole army of beer retrieving robots and step one, army of beer retrieving robots. Step two, question mark. Step three, enormous profits.
Dave Niewinski: 45:17
Yeah. World domination.
Jon Krohn: 45:19
Exactly. Cool. So, all right. So here’s kind of a big one, kind of a big question. It’s one of my favorite questions to ask on the show, but I don’t actually ask it all that often. It’s kind of like something when I save for guests that I think you’re going to have a really interesting take on this. So you are working in a field that changes rapidly, probably every five years, it’s kind of a completely different landscape in terms of what’s available for hardware, what’s available for software, what you can do, what applications are possible. So there are these various trends, these exponential trends, each one is exponential and they combine together to mean that the pace of technology is moving extremely fast every year. So we’ve got data storage is extremely cheap and always getting cheaper. Compute is cheap. You’re talking about these Jetson Nano’s with GPUs on them for a hundred dollars that don’t take very much power. I mean, amazing. There’s more and more sensors everywhere, and those are getting cheaper. You’ve got these stereo sensors like we’ve got tons of interconnectivity, internet bandwidth, people are sharing their innovations. People are committing to GitHub repositories and uploading YouTube videos like you are about how to be doing various aspects from a software aspect, training a neural network all the way through to the wiring the robotics. We can learn about so much of this stuff online. So with all of these different exponential trends all happening, what excites you about what might happen in our lifetime with say AI robotics or in your kids’ lifetime?
Dave Niewinski: 47:13
Oddly enough, the one thing I think I’m most excited about is self-driving cars. They are robots in kind of the purest sense. I mean, they’re hidden inside a car, but they are still autonomous vehicles. And I can’t wait for traffic jams to be less and for me not to have to drive places.
Jon Krohn: 47:38
That ties back to the very beginning talking about Waterloo and how people are bad drivers.
Dave Niewinski: 47:42
Yeah. I am very excited to have that problem to sort of take care of itself.
Jon Krohn: 47:49
It’s the thing we started at the very beginning of the episode. We need to get the drivers out of Waterloo.
Dave Niewinski: 47:54
Yeah. It’s funny, I used to travel a lot for a previous job and always coming back to KW. It’s like just terrible drivers here no matter where I went. Always the worst here.
Jon Krohn: 48:07
KW for our listeners is Kitchener Waterloo, which is two separate cities that have both grown very large and now just overlap a lot.
Dave Niewinski: 48:15
They just meld it into one. Yeah, I’m very excited for something like self-driving cars, but I don’t know where any of this is going. Like you said, we keep getting more computing, more storage and more everything. But I mean, what’s the next big thing? I have no idea. I hope to be part of it whatever it is.
Jon Krohn: 48:46
Wow. Yeah. I think you are. Whatever it is, you are already part of it. All right. That’s awesome, Dave. Do you have a book recommendation for us by chance?
Dave Niewinski: 48:57
The last books, the last series I read was the Bobiverse. You read that?
Jon Krohn: 49:02
The Bobiverse?
Dave Niewinski: 49:03
Bobiverse. Yeah. It’s about a guy who dies and is reborn again in the body of a robot. And so he has like it’s his consciousness but with control of robotic hardware and he becomes a spaceship. It’s a very good book series. There’s three of them. Quite enjoyed it.
Jon Krohn: 49:26
Cool. Well, that is an awesome recommendation given the topic of this podcast. Nice work. Sometimes people go completely, you never know what book people are going to talk about, but your book was very much on topic. So that’s cool. So in terms of following you, obviously the Dave’s Armoury YouTube channel is a go-to spot. I cannot implore the listener more, if you are vaguely interested in how robots work, how you can be putting machine learning algorithms into robots and be doing meaningful work with it, Dave’s videos are amazing for explaining in a surprising amount of granular detail how you can be doing it yourself. So I highly recommend Dave’s Armoury YouTube channel. How else should people follow you or get in touch?
Dave Niewinski: 50:21
Either that or Instagram. I have an account on there. It’s Dave’s Armoury with a U because Canada.
Jon Krohn: 50:31
Yeah, the U in armoury. Yep. Cool. All right, we’ll be sure to include those in the show notes. Dave, thank you so much for taking the time to be with us.
Dave Niewinski: 50:42
Thank you for having me.
Jon Krohn: 50:42
I learned so much and I’m sure are the listeners did too. Hopefully I’ll have you on again soon sometime so you can tell us all about your new armies of robots that have developed in the meantime.
Dave Niewinski: 51:04
Thanks for having me. It was great to be here. I’ll be back.
Jon Krohn: 51:04
What a fun and utterly mad scientist, or should I say engineer Dave is. I love his creativity and I can’t wait to see what robots he conjures up next. In today’s episode, Dave greatly expanded my understanding of robotics. I hope he did for yours too. He detailed his mission to get people into robotics by publishing entertaining, but nevertheless highly technically informative content on his Dave’s Armoury YouTube channel. He talked about hardware such as universal robot arms, clear path mobile robots, Zed 2 stereo cameras and Jetson Nano GPUs. He also talked about software such as the OpenCV computer vision library, the ROS robot operating system, Roboflow for operationalizing machine vision algorithms in robots and Google Colab for training deep learning models freely and easily in the cloud. He also talked about robotic integrators that provide custom industrial robotics solutions, and that if you’d like to get into the AI robotics world yourself, you shouldn’t try to become an expert in everything at once, but perhaps focus on the AI modeling aspect in particular, especially if that’s already your strong suit. As always, you can get all the show notes, including the transcript for this episode, the video recording, any materials mentioned on the show, the URLs for Dave’s YouTube channel and Instagram profile, as well as my own social media profiles at www.www.superdatascience.com/529. That’s triple www.www.superdatascience.com/529.
Jon Krohn: 52:33
If you enjoyed this episode, I’d greatly appreciate it if you left a review on your favorite podcasting app or on the SuperDataScience YouTube channel. I also encourage you to let me know your thoughts on this episode directly by adding me on LinkedIn or Twitter and then tagging me in a post about it. Your feedback is invaluable for helping us shape future episodes of the show. All right, thanks to Ivana, Mario, Jaime, JP and Kirill on the SuperDataScience team for managing and producing another fun and informative episode for us today. Keep on rocking it out there folks and I’m looking forward to enjoying another round of the SuperDataScience podcast with you very soon.
