3. Growing Clearpath with Open Source Software, with Ryan Gariepy

2021-09-20 · 1:07:31

In this episode, Audrow Nash interviews Ryan Gariepy, Chief Technology Officer of both Clearpath Robotics and OTTO Motors. Ryan speaks about the origin of Clearpath, how Clearpath focuses on making reliable robots, and the future of robotics.




The transcript is for informational purposes and is not guaranteed to be correct.

(0:00:02) Audrow Nash

Hi, everyone, welcome to the Sense Think Act Podcast. I'm the host Audrow Nash. And in today's episode I interview Ryan Gariepy, Chief Technology Officer of Clearpath robotics. Clearpath robotics makes several mobile robotics platforms that are used by researchers and corporations. We talk about the origin of Clearpath robotics, how they have focused on making extremely robust robots, how Clearpath uses ROS 1 and ROS 2, and on the future of robotics, I thought it was very interesting to see how clear path as a company has matured with expectations and the robotics industry. And I really enjoyed hearing what Ryan thinks will be the future of robotics in the next two to five years. As always, big thank you to our founding sponsor, Open Robotics, and I hope you enjoy the episode. So to start, I would love to have you just introduce yourself and tell us a bit about ClearPath, too.

(0:01:08) Ryan Gariepy

Of course, the My name is Ryan Gariepy. I'm the CTO and co founder of Clearpath robotics. And I have been stuck in my office for about a year and a half and counting.

(0:01:22) Audrow Nash

All right, and so tell me a bit more about Clearpath. What's the kind of mission that you guys are going for? And how long has the company been around? Clearpath has been around for just over 12 years, probably 12 years, 12 years, one month and 21 days, I believe, as of today. Twenty one days! Something like that, the we started, we started as the company with an intent to be the company to go to when you had a robotics problem, and you didn't know where to go to solve it, or you had ideas on a problem that you thought you could solve with robots, but you weren't quite sure you weren't, you didn't know where to go. And it wasn't necessarily a consultancy, it was about building some of these products, these baseline products and hardware and software because it didn't exist at the time. And that's certainly been the case, we have seen the robotics market grow beyond our wildest dreams. And six years ago, we also started a sister company called auto motors, which is a more vertically integrated supplier of production grade autonomous materials, but autonomous mobile robots to the materials handling industry. And that's also been exciting. we've, we've actually driven, you know, millions and millions of hours unsupervised, fully autonomous on that side of the business, which I'm sure for all of the roboticists listening out there, they they appreciate the challenges inherent in that. Okay, so you started ClearPath wanting it to be for solving general robotics problems. How do you do approach like that? How did you approach making it so you could solve general robotics problems? Well, when we were started, or when we, when we got started, there was no standard way to build a robot or no standard way to build software, anyone doing research, whether it's theoretical, very academic research, or applications level Industrial Research, or government research, you were always building everything from scratch, the hardware, obviously, but also all of the software, right there, there was open source robotic software, but it wasn't really well used. And we want it to be that company, where you could go to and describe your problem, whether it's the research, the research problems, or the application you were trying to investigate. And that we would help you move forward without having to build your own expertise in robotics from the ground up, before you even started. You want, you know, researchers want to research or entrepreneurs want to want to test ideas. They don't want to be, you know, they don't want to be caught up in a year of learning how to build a robot before they even get started on what their idea is. It's demoralizing and frustrating. Hmm. And so Clearpath you guys have a bunch of general platforms effectively that people can build off of for their different applications, like on different surfaces and things. Yeah, we started we started in the OCR space. And that was because we'd realized and this this was back in 2008 2009. We realized that outdoors was where the really valuable use cases were going to start appearing. Were the new valuable use cases we're going to start appearing at least That's where the research was being done. And on top of that, on top of that, the on top of that, there was a lot of recent progress around things like sensors. For example, when I started my undergraduate degree it cost probably $3,000 to buy a an IMU, which was, which was half decent, which could tell you basically which way was up. And then through the end of my, under the end of my undergraduate, it was, you know, 30 bucks. Now wide, ours were still blindingly expensive at the time, but we were starting to see those prices drop we, for example, the the U r, g o four, which was the first really affordable navigation grade LIDAR, you know, small one about that big that that appeared through the end of my, if I recall, correctly through the, through the end of my undergraduate time. So we're starting to see this significant drop in sensors as well as a significant increase in the power capabilities of the robots. So now off road was all of a sudden possible, and it was this brand new space and research just for scale. How much do you think you could get a comparable IMU or Inertial Measurement Unit? For today? Like the same quality? Would it be like, my guess would be like 10 bucks or something? Or what do you think it would be? On the volume Far, far, far cheaper, for cheaper, the the most expensive part these days would be the engineering effort to stick it on the PCB. That's crazy, like between the IMU between the IMU themselves and all of the filtering software that now exists, and the processing software that now exists, like, your cell phone is probably better than those first time use we were using. Yeah, that's absolutely crazy. And then so outdoor, how did you identify that most? Or that there was a lot of potential interesting things for outdoor robotics applications? Like what were you looking at? Yeah, if we start from the research space, it was because there just weren't any research platforms out there. Or just there weren't many there were very very few research platforms out there, which could handle a reasonable outdoor terrain. There was a lot of indoor platforms like you know, there was a bunch of still a bunch of old like, our wi indoor platforms around there was is that is that a company? Or was Yeah, that was real real interface. Okay. And you had you know, the at the time it was mobile robots then became adept mobile robots and is now Homer on but you had there, the Pioneer three, three, dx was a very, very much a workhorse. You had people who are hacking roombas around and the first create was being used, but these were all indoor platforms. And the researchers there were obvious there obviously, there was Outdoor Research happening, right, like the CMU robotics Institute was, you know, had been doing outdoor robotics research for for decades. But it wasn't a very common thing. And we that's where we said it, that's where we're gonna get started. That's we're gonna make our mark by building these really solid robots that could unlock entirely brand new fields of field robotics work. That's really where we got started. It's almost ironic that auto was that when we started auto, it's almost entirely focused on indoors. But that's that's another story. Yeah, that is funny. Okay, so you were working on these platforms? And just Can you give me a sense of what they look like overall, like a lot of the platforms that are used for outdoor use? What do they look like? You know, generally most of the platforms we have their their black and yellow and they have four wheels. Yeah. And it's one of those and research is probably one of ours. Yeah. And they so they mostly they it's like a big flat spot on the top where you can attach different things that you might more cameras more, anything else. So yeah, that was generally what we wanted to build. I mean, we knew that the software space in robotics wasn't quite ready for the software platform yet, but we knew the world was ready for physical platforms and we took that somewhat literally. So building some big software platform, what do you mean, the worldwide software platform I would mean something you know, like the iPhone where you have, you can write software, right you don't need to change any hardware. We're now at the port Now further along that curve. But yeah, we we tried to make it really easy for us to integrate things we tried to make it easy for our customers to integrate things the sort of robots that if you decided that you're in the field and something needed to work and you could take a hammer to it, you could take a drill to it that it would work you can you know, pack it together and get it out do your research and get it back in the field we to the point where we would I don't know if we still sell these but we would sell replacement top plates only for when someone just wanted to, you know drill and tap the top plate for some research gear and then later they realize either they made a mistake. They really like it. And you can take the top plate with all your sensors, you could put it on the shelf and you could you could share the asset around right that's that's that was actually a very well, it's still a really common thing now that you have self driving vehicles where you would get a, you have a dozen researchers but only one physical robot, right? Because you could only afford one, one sensor setup. Now there's research labs where you can have things like, you'll have things ranging from, you know, the turtlebot, three, all the way up to like our, our war hog and moose platforms, for example, which are on a larger size. But it was it was very much like, Well, you know, at the time, you had the PRT robots in the field, and there you have an entire research lab, which was around those PR twos. But even for the labs, which couldn't afford that, they were still bought, they would buy like our Husky platform, which was our first and actually one of the most iconic platforms we make. And you'd have to share it, right you have this this little robot, it's got a bunch of sensors, you need to share it among all your students. That is nice. It's clever to just replace the top plate. That seems like a good solution. So let's see, are there any particularly interesting Apple, I'm sure that people are using your platforms for tons of things, but are there any that were like, particularly interesting or surprising uses? Well, if I go outside the if I if I ignore some of the confidential stuff, there's the one which I saw most recently, which, which surprised me and I didn't I didn't actually know about this one until I saw it in the news, but was a college which was using the boxer which is a industrial autonomous mobile robot that Clearpath sells, which is actually a variant of the auto Amr. So it's a boat, 100 kilos, 100 kilos built, like it can survive a factory because it can, and it was used, it was being used for robotics or applications research at St. Clair college, I believe. By recently, obviously, people have needed to do all sorts of, you know, working remotely schooling remotely and what have you. And they rigged it up as a, they may rigged it up as a moving clothes, horse or clothes, or like clothing model for lack of a better term, where they had a fashion, they have a fashion program at the school. And it was very important for them to see to see how the clothes would move. Like how like how they would drape when they would move and they get the and it's like you can't you're you're doing all this judging and exhibiting remotely and you can't you're not just gonna set them all up and mannequins and leave it like that and have the judges look at it so they would actually swap out these I can't exactly remember how they how they held the clothes on top the robotic It was a mannequin or mannequin or something else. But they would they would swap them out and and run the fashion show like that. I thought that was just you know, that was really really interesting. I mean, everyone hears about everyone. Here's what the disinfection robots that people built in the last year like I don't know how many have been built on our platform, but everyone hears about those but you never hear about these. These little like that. Yes, the clothes modeling robot. That's hilarious. What a great youth. Oh, yeah. Okay, and then one thing that's been very interesting to me from your previous talks and things, your robots they're designed to be super reliable, like go forever. Can you talk a bit about this from the from the beginning, we would, we would joke with our, with our prospects that that the reliability standard standard for our robots is is outlast the researchers. That was the that was the idea. Even at the clear path, even at the Clearpath robotics side of things where the standards are not as high as on the on the industrial front. But that was the goal it was you should never have to you should never have to cut short your field robotics research because of frankly, bad weather. And I think that comes from that comes from us all us all being Canadian are all the early the founders and the early employees all being Canadian, is that when you're out there testing your robots like one half of the effective school year the winter term, it's sort of slow and when the snow leaves it gets replaced by mud. So we really wanted to not get in the way and not have to worry about not have to worry about the snow or not have to worry about the rain. So that was that was our initial standard. And then we've kept pushing from there and this was a very interesting time for robotics because the this we were able to mature along with the market expectations. These days if you're starting robotics calm It's just got to work. Like, you know, you're compared against, you know, robots like, like ours are fetches as they are now, you're not compared against the robots that we had 10 years ago. But that's that's what the market does. Yeah, it matures. That is interesting. So it's really changed as the market has grown. And people expect a product now, what do you guys do differently, to make it so that your robots are even more reliable than they were in the past, which was still super reliable. The thing that's really helped is the fact that the rest of the world has now caught on to the potential power of robotics. So it means that it's easier for us to build partnerships with other companies who have a lot of some of the who have some of this fundamental fundamental expertise, like sensors, sensors, or actuators, or simulation systems. I think that's been that's been really helpful. So even a lot of the larger robots, for example, like Clearpath, robotics cells are built in partnership with vehicle suppliers. So where the robotics experts, but we actually work with other vehicle experts, or, or even more recently, more recently, it was very, you know, well covered in the robotics media that we did a, we're doing a we have rather a partnership with, with Boston Dynamics for the research space. And that just makes sense, right? Boston Dynamics is, is one of the leading if not the leading manufacturer of legged robots in the world. And we're the world's leading integrator and solver of these problems. So it was it just made sense that, that our clients in the world needed more legged robots. And now that that's possible. And why would we try to do that ourselves? Right, work with those who've already picked it up? Yeah. Interesting. So going back to working with the vehicle companies, so do take advantage of their existing supply chain and their existing processes? Or reps, how does that work? Ah, yeah, well, ya know, they go hand in some cases, the vehicle in some cases, the vehicle shows up, like at our door, where it's been a repurposed vehicle where they've remanufactured parts of it, and then we will take it and turn it from, you know, a, what might be described as a giant remote control car into this ROS enabled fully integrated research platform, you know, upgraded power systems, upgraded power systems, onboard compute all of the sensor options that we might have all of the customization options that we've got. So just like early days, you could call us up, you know, 10 years ago and say that my research, hypothetically, my research is in, you know, mobile manipulation and unknown environments. And we would turn around and say, Okay, well, maybe if your budget is x, then you can't do the 3d LIDAR. But you can get an arm like this, and you can get a camera. We've got that same, we've got that same. That same political experience for for everything else we do. It's really about providing like I said, this conference, one of the last conferences, I was actually able to speak out in person that are we do have very reliable robots. But in the end, what we, what we sell, is, is the ability to speed up the robotics development cycle. Hmm, definitely not true. And just out of curiosity, where are most of the manufacturers that actually build the robots you guys make? There's a lot we work with, in Well, in Ontario, Ontario has a lot of vehicle manufacturers, we've been we've been broadening that. So traditionally, we've been very heavy on still building our own vehicles. And where we've really got the broadest reach will be in the manipulation and setting spaces because those markets have been more mature for a while. In those in those you can really, you can imagine the countries in particular like you know, Germany and Japan are the two probably largest companies on those fronts, but that's probably not a surprise to anyone. America, Denmark, I mean, you are we were we had we worked with universal robots on a custom skew for ourselves and our clients. I don't know how you go, this was what sorry, stock keeping unit. It's a sort of custom configuration of their arm. So you're typically when you bought a universal robot arm, you needed to and you want it to do research with it. The Universal robot arm would have you'd have the arm and then you would have a control box which is a giant control box meant for industrial relatively decent size meant for industrial floors and it had three boards in it. It had an A power supply board, which was effectively a, you know, transformer and step down, or interactive rectifier and step down transformer. And so getting the power, yeah, power regulator, yeah, then you'd have a Safety Safety Board and a controller board and have that you have this giant enclosure, and a big power supply. But it turns out that most of our large vehicles, supply the power already supply the power at the spec needed, because they have a big DC, they have a big DC buses on them for power. So if you wanted to, if you wanted to integrate a mobile or AI, your robot onto a mobile robot, you would have to put in a an inverter like you would you would need an inverter to go up to AC, and then you would end up what would end up happening inside this black box, which was also kind of large was that then then it would end up losing more power, because you're just stepping it all the way down to DC. So we had a race to AC and then back to DC and then back to DC, right, which is generally a waste. Yep. So it's wasted amount of wasted space. And, and just extra cost. So we came to an agreement with them where we would, we would go by the arms and by the two boards, and then we would mount them ourselves. And you were you used to have this this ratchet, just huge package, that package was no longer there because we could mount the components inside of inside of the the chassis itself. And that was great. So that was the court sort of partnerships, which we were always forging, which was to open up these brand new markets for these partners, and serve our customers really cool in time because like we're already we've already got our hands full building the basis, we're not going to build the arms too. Yeah. So you said that you guys kind of your core competence would be like as an integrator, if I understood correctly of basically taking the platform and then adding pieces to it so that the robotics community or anyone who wanted to do something with the robot could use it. Right? Or more, more and more. So I mean, we're, we're very good at vehicle design. Yeah, but you know, we have been, there's just now that the world is picking up. These, you know, now that the world is really engaging with the robotics space in the automation space. It's, it's just, there's just more demand that we can almost deal with, like, it's one thing to maintain a vehicle demand, like a vehicle portfolio of a half dozen to a dozen vehicles, which is what we have, and we're proud of every single one of them. But there's so many different opportunities out there that we've had to, we've had to really partner in order to keep everyone keep growing to keep focused on what we're good at. Yeah. And then and so in this role, it's a lot of taking an existing platform and you retrofit it with components and software, or how would you think of that? Like, what do you do when you get a new platform, or it's probably quite case by case. But yeah, it's very case by case I mean, you we don't want to, we don't usually just, you know, buy some random, some like, I don't know, buy a car off the lot and customize it, we, we want to make sure that we want to make sure that we can provide this continual discontinuous solution for people, which means that we'll we'll usually have already spoken to the manufacturer, we've we've come to some kind of agreement with them, the you know, the degree of which is is always case by case, but we're not trying to guess at things right, we don't want to, we don't want to do something a little bit Renegade, where, you know, you you buy a, you know, you buy an ATV off the lot, and then you kind of hope it works. Because the customers, the customers expect it to be maintainable and for, you know, spare parts to be available. That's, that's the other reason why people will, will work with us is because when all of their grad students have left or when their research projects have gone on to, to production, those robots will still be there. And they want to, they want the next set of researchers in the next set of robotics developers to be able to, to just, you know, jump on board, for sure. Now, you mentioned how customers come to you with problems and then kind of you see if you can find something that fits the how has that changed while you've grown over the last 12 years? The problems are a lot more the problems are a lot more diverse. And what people are looking for mannequin the Yeah. Well, that one was good. We need to know about that one. They just they just did it. But on the plus side, I think they're is a growing appreciation for a growing appreciation for what robots can and can't do. Right? Like when we started the even the concept of a drone that you could just casually buy, it wasn't a thing. One of the, the, the company I, company I internship I interned at was called aireon Labs, and they've just been purchased by FLIR and they sold drones for first responders. And at the time that they had started, you couldn't buy an autonomous drone for autonomous quadcopter. There there was, there was no money that you could pay to the market, which would get you a reliable autonomous quadcopter, short of funding an entire robotics company yourself. So that was that was where we were right. And self driving cars. You know, the the urban challenge was, was just one at the time, I can't even remember what year the the urban challenge was. But that was just around that timeframe, like autonomous cars weren't a thing. So there was still a little bit of a sense of robots could either do nothing, or they could do everything. So early on, there was a lot more need to reset expectations. And on top of that, there was wood on top of that, or, yeah, so you're saying people would either think they could do everything or nothing, kind of like useless or they will just be the Jetsons Butler kind of thing effectively? Yeah, gotcha. And so bring people down to the level that robots actually can do. And kind of show them how they can be useful still, with what they have? Or is that we that's, that's where we are right now. Right? Like, you can. In when, when we had when we were starting, you know, there there would be there were probably a few 1000 people in the world, who knew who could reasonably describe the state of the self driving vehicle market, like where the technology readiness level of the self driving vehicle market. But now, I would charitably guess that there are millions of people on earth who can generally describe what a self driving vehicle is. And isn't, they know what it is. And they know it's not taking you to Starbucks anytime soon. Without a a well paid test driver in the front, are you still paying 10? Yeah. But that's that's not how it works. So you'd have, you'd either be, you know, selling to someone because you knew there's an opportunity there, and they didn't really think that it was possible, or they would come to you with an opportunity. Now, this is beyond the researchers, this is when the commercial demand started to take an uptake. But you'd have people coming in and saying, Yeah, like, Can I pay, you know, $5,000 to automate my old, clunky gas powered lawnmower, and have it run completely unsupervised, through the middle of a park in a crowded day. There was a lot more requests like that, huh, there are still there. Just to be clear, there are still requests like that. But they are. But there are far more requests, where the people who come in the door are already conditioned, and already educated about what robots can and can't do. And I think that's, that's a that's just wonderful for everybody, right? Like, that's saves everyone a whole lot of time. And so if you're getting requests like that, where they're coming in and asking if this is possible, and how much it would be, that means that you guys have to be extremely knowledgeable about the state of the art in probably a lot of different areas. Because I mean, there's the hardware, there's the software component, and all the specific things that are no sensors how to estimate things. Like you guys kind of have to be broadly competent in a lot of robotics. Yes. Yeah. It's a it's a challenge. Um, but the thing that gives us that's been interesting, well, and I'll speaking for me personally, and it's one of the reasons I love this work so much, is because every day there's a new, there's some new development out in the world, in the robotics space that's gonna make that's gonna make a new application possible. It's a new technology and like, every day, there's something new that I've that I can learn. Like, I just actually right before this, I was I was reading a paper or reading a paper put together that's, that's, um, I don't know if it's been accepted or if it's been final submissions for iros, which is a big robotics conference. And it's, it's authored by people who work for Clearpath that's, that's just great. And I reading through the paper and realizing that my math has really gone down the drain. But anyway, I digress. Um, the The other thing, which I mean, it sounds kind of intimidating, which is like our job is to stay up to date with everything in the field, which is robotics at all times. But it's as, as odd as it sounds. It's not that challenging, because we don't have to do any catch up. Like if you're starting a robotics company now. You have a lot of catching up to do. Right like you have decades. of progress and you have all sorts of you know labs to track you, you need to know where to pay attention who to pay attention to which conferences to go to like all of it, you need to catch up with the state of the art and I had that as you're doing that the state of the art always changing. But you know, we got to do that work before this the field before the the the robotics field really blew up. So now it's just it's it's day for day. So you wrote rode the wave, basically. And now you're still on it, which is fantastic. We're just we're just holding on now. But it's, it's a lot easier than then, than trying to catch it from nothing right now. How do you? Do you structure your team with like experts in specific areas? Or is it the team of generalists? Or how do you how would you think about it? So my team? Yeah, we're the team as a whole if you include the auto motors business unit, which is a large portion and, you know, sharing services and everything like that. We're about 200 to eight 280 to 300 people were in there. That range right now, are you all in Waterloo, Toronto area, or were in the bay, I believe, you know, we, our offices, our physical offices, we have four facilities in, in the Kitchener Waterloo Region. So if you know the area, there's one in Waterloo, there's two in Kitchener, and there's one on the near side of Cambridge, and everything else is everything else is remote work. And actually, a lot of that was like all of our sales team. And a lot of our field team were remote before even the pandemic hit that was had some, you know, engineers working remote here and there. But overall, you know, overall, Canada is probably you know, if I don't know these numbers offhand, but I would guess 90 to 95% of the people that we have our, you know, live live in Canada live in, you know, southwestern Ontario. And then though there's the next percentage, there'll be a rounding error of people who are full time around the world. So there's someone who's in someone in Mexico right now, there's someone in Spain, we have a full time member of our team in Japan, and then everyone else, everyone else is in the US. So there's a lot of a lot of salespeople in the US field service tax systems, engineers, people like that. Because there's really obviously a big export market for us is PS autonomous systems into the US. So gotcha. And then for like, Is it about the specialized versus generalist approach? Oh, yeah. Yeah. There's so there's, there's probably about there's under 20 people at the company who would have had robotics experience before they joined us, that would be Wow, my guess that's great. Okay. So whether that's whether that's mobile robotics experience, like, you know, AGV deployment in industry, or, you know, a Master's or PhD, it's, it's probably around that number. I know, the the autonomy group in the auto space, the on the, the auto business unit has 14, I believe, 14 people, it's all masters and PhDs, doing various sorts of product development and research, for example. But beyond that, we really try to bring in experts and experienced people in their fields, whether that's, you know, operations, manufacturing, finance, sales, wherever, and then even in engineering, like I have, I think it's seven people who report to me right now, not a single one, not a single one started in a robotic space, and only one of them had any experience in robotics before joining this. Wow. So it really is, and, but people learn really quickly, right? Like we as we've got this foundation is, you know, foundation of technology. The the analogy I often make is, if you go to, you know, you look at a company like Audi, for example, or you know, any established high tech company, how many people go into that high tech company with expertise in that the thing that makes that high tech company special, like, how many people work at SpaceX, and of those, how many are rocket scientists? Yeah, like how many people work at Audi and of those, how many people actually know how to build an engine? Right? Like, definitely, that seems like something like as a result of the robotics field maturing, I feel like like at the beginning, it was probably a lot more roboticists and now it's, people have other specializations that can contribute, because it's large enough that they can focus on something that they have skills for. Yeah, well, and that was actually a big thing for us when we decided to start getting into getting into the industrial space in general was hiring people who weren't roboticists, because you know this. This has changed over the years, but You know, back when we were starting, you know, maybe 10% of roboticist could write any kind of decent code that you would want running in a production application. And I'm probably being, I'm probably being generous there. And that's not an insult, because I don't think I'd count myself as one of those 10%. But when we started really going after the industrial market, it was important to hire people who had written good software. So that's, that's what we did. And these days, I'd say our ratio is for every person who writes, for every three people who write software, there's one roboticist and one tester, that's probably the rough ratio that we have. Gotcha. Very interesting. Let's see. So moving into software, you at Clearpath. Were some early adopters and supporter of the robot operating system. Ros, would you tell me a bit about that work? Yeah, I can still remember a lot of those discussions like they were yesterday, we started, we started the company formally in 2009. And at that point, we had the philosophy that there was always going to be more smart people outside of the company than inside a company. That's, that's almost like a statement of fact, for any company, no matter how big. And we said, we're going to start with, we're gonna start with open source. And that was number one for that, that general observation I just outlined. And the other one is that one of our largest competitors at the time, really, really charged a lot of researchers money for this expensive autonomy package. Like if you wanted to buy a laser, you also needed to buy this really expensive autonomy package. And we did market surveys of the early early researchers to those people who were who participated in that 12 year, 12 year old market survey, thank you. And almost every one of them said they didn't like paying for the software because they were going to write it themselves anyway. So I said, Okay, we will start with a small open source software package called player stage, because eba, which was a combination of three player, which is you can look at that as an equivalent to ROS stage, which is a 2d simulation environment meant for multi robot simulations and gazebo, which most people still know, because it's still around. simulation. And yeah, so we started with player was a message passing precursor to ROS. Right, right. This kind of thing. Okay, so you started with those. We started with those. And then in in 2010, we went to our first official trade show, which was ekra 2010. In the in Anchorage, Alaska. Oh, wow. The say we'll just say that those trade shows are a lot classier now. But we saw and met the crew from Willow Garage. And one of the things which was really interesting, and they had their PR two out, and they were doing all sorts of fancy stuff, and they're way, way better funded than we have. But the thing that really was impressive is that if you've ever installed player stage, it's a bit of a challenge. And, and the Willow Garage team had put so much effort into making ROS not only just a good piece of software, but also well documented and easy to install. And it just seemed like the way to go. And we knew that people from the player stage community had gone over to the, you know, being supported by were part of Willow Garage are being supported by Willow Garage like Brian querque, first and foremost. And then, so we said that, okay, this is going to be the way it's going to go. Right? This is the only place which is actually funding, not just some software, but the community around the software. So we basically turned at the drop of the hat and said, you know, probably that that would have been April, May and 2010. and I are going back to going back to Mike who's our first engineer, first engineer at the company saying like we need to, we need to start this supporting this ROS thing is starting to talk to our customers about it. So to the best of my knowledge, we were the first company. We're the first company in the world that probably made money off off of something tied to ROS. That's awesome. And that was the case for a very long time for years. A bunch of these other companies in the space, we're just having an HIE and I'm like we need to support ROS and that you could you could get these other robots to be supported but it was always some third party some third party like a library written by a grad student who happened to have that robot in the shop. You can like dross and these these days and then you know things continue to progress from there we have been talking no early days about the creation of at the time was called the ROS foundation. That was the the name for it. But now you know it is the open source robotics foundation. So that got founded I joined the board for that. The turtle bot turtle bot was an interesting story. So that that was obviously created by by Melanie wise, you know, at fetcher, I guess they were just acquired. So congratulations to them. And intelli foot is still at the store with Open Robotics. And they created this turtle bot thing, which is a really cool idea. And we had a smaller, a more extensible, but more expensive, but more far more expensive educational platform available at the time, and said no, like this is and the other thing that happened around the turtlebot was that the Microsoft Kinect came out, it was just the perfect sensor. Like that was the first commercially available RGB sensor. So then there's a camera and depth. Camera. Yes. And, and that meant that you never needed to worry about LIDAR, right? Like you didn't have to worry about the laser, at least for small scale stuff. So that was really cool. And it turns out that Melanie and Talia were tired of putting turtle bots together by hand. So we, we had a conversation and ended up that we we supplied Willow Garage with a ton of turtle quads, and then became the first manufacturer. First manufacturer turtle bots in the world. I can't I can't remember who else was there? There was Bill Morris from the company. He has his own business called I heard robots, or I heard engineering, I think it's changed between those names. And I think there was a third company at the time, I think they were they're off in Europe, but I can't remember who it was. And it was like that was like 10 years ago. But that was also a really interesting thing at the time that, you know, we decided we kill their own product, because we think of the turtlebot was going to gain momentum in it. It certainly did. And then yeah, other stuff around like ROSCon i was, that was what 2012 was the first ROSCon. So we helped, we helped found that. And that's still going strong. I know literally right before the right before the we got on this call here. I just finished my finished mice my share of the reviews for ROSCon 2021. So we Orleans in this year, New Orleans, that'll be a blast. But it's been exciting, right? Like the ROS community was huge, or it's it just completely mind blowing, the sort of things that we've seen here like we when, when you started when we started both pockets when we started there, ROS was out there. But even when we started supporting ROS, no manufacturer was out there supporting ROS and knows, like every sensor that you needed to get into ROS here it was written by, you know, Willow Garage, it was written by us, it was written by some grad student halfway around the world like, but now now you get a sensor or you get a robot from any reputable manufacturer, and there's probably a ROS driver for it. I remember getting a new prototype sensor from a, we'll just say it's a well known manufacturer. And the only instructions on how to use the sensor was was like install the package we gave you. And it's ROS launch this. Yeah, that was the instruction. And Previous to that, and previous to that, like every sensor manufacturer had their own their own, usually Java implemented sensor visualization package. Some still do, and I don't blame them. But it's been this, that's the sort of thing which is really interesting is, you know, first they would go from what's ROS from, you know, they just didn't know what it was they didn't care. And then later we would, you know, be trying to lean on them a little bit and say, Well, if you want us to sell this new sensor you have make it easier for us to build a ROS driver or like release your package online or do something. You know, at that point. It's like you I can I can almost guarantee you that your own engineers have a ROS driver if you go and ask them. So just put that on. Yeah. Yeah. And and now we're at the point where the sensors just you just come out of the box, you don't need to ask. And that's huge, right? Mm hmm. Let's see. So and you use ROS for everything connecting everything on your system, even early on. Is it true? Yeah. That's awesome. And so now how are you guys involved in the community with ROS? So you're obviously still a part of it? Yeah, definitely. Well, you know, still on the still on the board still running ROSCon. Our ability to stay up to date with all the recent versions ebbs and flows. So I apologize to anyone in the community who is or who, when we get when we caught a little get caught a little bit behind the ball on whatever the recent releases. All of the auto autonomy code is all based on ROS and uses the ROS middleware, it's on our fleet manager as well. Our inter fleet and robot comms on the, on the, for auto for our industrial Amr, or mobile robots. That's all process to now. So we've started to all the costs of their autos. Yeah, so all the autos all the autos talking to our fleet, our fleet manager, that's all based on ROS 2, since that's exciting three, four years back. And we were the, to the best of my knowledge, we were one of the first people to actually use production or ROS 2 in a production environment where, you know, if it failed, you would have angry phone calls. It was, you know, it wasn't through the entire stack, right? Like our autonomy system is still very ROS 1 based. Didn't that's just like a legacy thing. So we're talking about migration plans now. But, but we actually did an independent investigation when we needed to really scale up the amount of robots at a given factory. And we found that DDS was actually pretty good, that DDS was a good direction to go in and we said, okay, well for if DDS is what's coming out of this, then let's use ROS, too. So we did that. And it's Yeah, it's been great. See, I'm looking, I always forget what DDS stands for. data distribution service, data distribution service, and it's a really, it's a flexible middleware that allows you to kind of interpolate between TCP, which is lossless or not, like it doesn't lose data, and UDP which can lose data, but useful for like walking robots and this kind of thing, where you don't want to hold it up if you miss data. Things Yeah, so there's there's a bunch of stuff in there which is just which was already implemented, right? Like quality of service quality of service is a big thing. Some of the other stuff was just more of like how it was implemented. So for example, with with ROS 1, both the c++ language, the c++ language bindings, or the c++ library and the Python library, they both independently implemented the ROS 1 protocol. So it meant the with like so it meant that 20 was sometimes babies well twice as much maintain and then sometimes you had gotchas like especially early on where it's like oh that function and that function and subscribe well that only works if you use Python great well, or vice versa. And so that was like another big change that went with ROS 2 was you know, implementing everything at a you know, implement, implementing everything at a sink or the bindings on top of DDS at a single layer or a single language then and then building on top of that and then the other thing that's nice about that is that now it does make it easier to notionally migrate to some newer libraries like I know rust has been talked about a lot in the rust community these days but yeah, there's there's all sorts of other you know, little things like ROS 1 was really only built for a single very complicated robot like one robot that's doing very complicated things and specifically the PRT but then people started using it for fleets and robots really quickly or the other thing is that people want it people want to start running ROS on microcontrollers for obvious reasons and it's it's far easier to shrink bras to down and then I think yeah well they're they're definitely implementations of rosu that can run on a microcontroller I'm not sure how public they are it's always difficult to remember that so I won't mention any right now but yeah, you can you can fit like you can do a lot of ROS on a microcontroller which is now really interesting with the ROS 1 days you would have a package called ROS serial for example which if you wanted to talk to an Arduino you needed you put this like lightweight serial protocol on the ROS or on the Arduino or you know any AVR chip and you would basically tell it that he was allowed to use these know a few of these language or a few of these these message types but it wasn't really ROS so now we're able to do a lot more but there's a lot more consideration as well for you know real time real time nature of it which opens up the door to which opens up the door to you know safety critical ROS safety critical ROS implementation is you know, hard real time guarantees there's all sorts of interesting stuff there.

(0:49:03) Audrow Nash

I would say I mean and I read it, would you talk a bit about the safety critical ROS just kind of thing I don't know too much about it. Right says is it using standards around cross where you see you can guarantee some sort of performance and if it's real time you can deterministically say this is going to happen at this time or we know that this task will complete and this kind of thing. The the thing which is so obviously there's various levels of you know, safety of safety guarantees and up until now and this is actually including including our products for example, a lot of the safety happens at a lower layer under ROS right on the motor controllers are their safety relays or their safety PLCs or safety controllers programmed and you know, some some safe art for your jobs or whatever and That's not that's usually not in ROS now obviously your benefits in so many reasons there are so many ways to not having to do that and the nice thing is so Russ to on its own right now you know it's not certified it's not certified it's not you know if you use everything there in ROS you're going to be able to do things like sort of like like under a like a you know under the ages need some sort or yeah MISRA C language standards or c++ or like ISO 26262 or IEC 61 508 like basically if you love the standard libraries then you you probably know what I'm talking about do dash 178 is the aerospace one so I need ROS ROS to on its own because it supports by definition so many different languages like or so many functions in full c++ then it's not going to be certified on its own right because you can do things with c++ which are considered not safe if you really really try to be specific like and you're trying to write something that's you know safety readable very high levels like you know can run a nuclear power plant levels and and what have you that you need to avoid things like any kind of dynamic memory allocation ever. And so obviously, if you if you if the ROS community built a toolset that didn't ever let you use any dynamic memory allocation then that would we'll just say not be adopted because there's a lot of places where that doesn't matter. We actually built safety rated systems and we have we've made certain choices that optimize towards time to market at the expense of a little bit of extra cost on the safety system right but that means that we can use the full gamut of c++ or Python functions now but there are companies and there are initiatives out there where they have been able to trim it down but the benefit of ROS 2 is that it was is that you can at least trim it down with ROS 1 that was just it was just on tenable like you couldn't even consider it but now you can't like it's it's it's much more technically possible thing to build you know flight capable control software that uses ROS 2 as a as a messaging pass as a message passion framework

(0:52:23) Audrow Nash

for example. And then so you guys are migrating your Clearpath robot, you primarily run your ClearPath robots with ROS 2 correct Yeah all the all the all the custom software we do is ROS 1 right now

(0:52:39) Audrow Nash

Is ROS 1. Any so and you mentioned the auto where you guys are using ROS 2

(0:52:44) Ryan Gariepy

Sorry, sorry. autoware as in the Sofer autoware as in the community initiative, we don't use that what I was referring to was the autonomous code that are auto Ott Oh, autonomous mobile robots use gotcha so that's like everything there is remains you know ROS 1 for the time being I believe we have will be will start supporting ROS 2 will start supporting ROS 2 on some research platforms in the near future. But since I don't have the schedule up in front of me, I'm not going to I'm not going to actually say those dates because I like to support my engineering team the the I don't have a Twitter account where I just I just randomly named things on the on the the side of our of our autonomy software, we use ROS 2 for comms and we're started we are starting to consider the the migration process we just finished the Python three migration. So next up is the ROS 1 to ROS 2 migration. But as you can imagine, as you can imagine, when you have several dozen and software engineers writing code, and you know, in this code is also all going out to the factories, you we can't stop everything and just do a migration of, you know,

(0:54:06) Audrow Nash

five systems business continues of engineering. So I've been talking with a lot of that stuff if we and of course if we find any, you know, shortcuts along the way we'll we'll definitely be short sharing them with the community because I'm sure there's other people in our, in our situation, for sure. Are you getting comments from customers, or people that are coming to you? Or they're asking for ROS 2 or how's the community? Common? Yeah, it's becoming more and more common I can say from I can say from generally looking at from generally looking at you know, the interests at ROSCon there's a lot of people developing in ROS tooth right now. And they're developing you know, very real, very serious things in ROS 2. We're and we're definitely getting interest with ours. You know, for example, One of the very things, which is ROS 2, which is very clearly ROS 2 is that if you are going to interact the API to our indoor autonomy software, for example, which is what runs all of our industrial self driving vehicles or industrial mobile robots. That's all ROS. That's, that's the API. So if you want to go and talk to one of our talk to one of our autos, ROS, who is your only option?

(0:55:24) Audrow Nash

Gotcha. Now, let's see worse, I would love to talk about some other things. But seeing that we don't have that much time left, if we're trying to hit an hour, I would love to ask you some kind of larger questions and get your perspective on a few things. So the first, you are saying you caught the wave of robotics, really, and you guys have stayed on top of it now. And also, it seems to me from talking now, you guys also caught the open source wave, which is really exciting to be on both of those, like, just you invested in open source and the community and the amount of code out there. And the quality of code is just grown and grown. How would you? What advice would you give to a startup that is looking to do a hardware robotics startup, and they're just beginning now, and they don't have all this infrastructure, or expertise. So the first, the first bit of advice I'd give is to not is to know what you want, know what value you're bringing to the world. And partner as much as possible around that value. And what that means is, when we started our auto division, we had to build the robots from scratch, we had to build the autonomy from scratch, the fleet manager, the UI, all of our continuous integration systems, we did a bunch of like internal upgrades to like gazebo to make it fit in continuous integration systems. Like control software, a lot like all of this stuff, we did all that from scratch. But now you can basically partner or buy a lot of that stuff off the shelf. And I would really encourage any new startup in robotics to start with that and not try to build everything from scratch, have a very good idea of what they're bringing to the table. And if I look at a Yeah, if I look at the way, you know, a typical startup, you're either bringing a new technology to an existing market, or you're bringing a new market, or you're you're, you're taking an existing technology and applying it to a new market. And so the Airbnb and the Uber, for example, like they were bringing, they were, they had a market hypothesis, and they were bringing some some pretty standard technology there. Like at the time everyone had smartphones. And obviously, we all still have smartphones. But us we had we had a newer technology, which was being brought to an existing market, which was materials handling, right? And I'm using the auto example here. And the but you want to pick one, right? So are you. And these days, there's just so much technology out there. And really, I think a lot of the opportunity is all these new markets where you can apply robotics. Yeah. And that means not inventing the wheel along the way. And, you know, part of this I we spent a ton of time earlier on in this conversation talking about the ways that that ClearPath engages with people. So obviously we can support those, those sorts of customers, we've supported startups along the way. If and, you know, what, if if they're working with us, and then the startup turns around and says, You know what, I've got an order for 1000. And we're going to go and outsource that stuff to let's say, jaybo, for example, and because they're going to give us the best deal. Hey, that's cool. I don't I don't have a problem with that. Thank you. I'm glad you're succeeding, right. But there are companies like us, and then there's all sorts of off the shelf technology out there as well, that can help you along the way. So that would be the real thing. I'd say to those new robotics startups, like don't try to do all the hardware and all the autonomy and all the fee management and all the information that shows what you can use what you can't like I have I have spent those budgets. I know how much it costs. And I know Yeah, I know the stresses that it brings you. And it's it's not worth it. There's other ways to do it. Now at least. What do you think about this idea as like platform as a service or something like this for a business model? And what I mean by that would be, if I am a robotics company, I'm going to buy someone else's robot but my own code on it, and then have it do something useful. Do you think that will be a reasonable business model in the future for robotics startups? Like basically, so which are which side is the startup honor? And are they on this side of writing the app? Yes. Basically an app. Yeah, right, I think. Yeah, absolutely. I think that's going to be I think that's going to be the case, I think, early on. It's going to be combination of the app and some tooling, not perhaps not as complex as a robot, but that you're going to get the combination of the app and the tooling. But I definitely think that's going to be the case. And we've seen this actually, through the through the recent, you know, year and a half, we've seen a lot of companies with no robotics expertise whatsoever been able to build disinfection applications, right. And that's what they've done right? Is they built a new application. They built a new application. They built some hardware, they've stuck it on a robot, and they're good. So yeah, it's it's huge. I think, I think, right? I think for now, you're going to see it as you're not going to necessarily see app, like just someone who writes apps and makes money off the app. Or was there even the app plus tooling, I think there's still going to be like, app plus tooling, plus, you know, a sales effort on its own right? Like you're, if you're selling, if you make a disinfection robot, can you've done the app, you've built the hardware, and then you're going and you're going to sell it? Right? So I don't think we're yet as disser meat disintermediate disintermediated as Apple, Apple is, for example, where they also bring you the customers. But I definitely think that the time for getting a robot off the shelf and just putting like, and just talking to its API and throwing some tools on it. That time is now like, I think that is now two years ago, it wasn't then but but in the last year or so I think we're over that tipping point, which is really exciting. I mean, I wouldn't I there's, you're still there's still going to be lacking in that the caveat there is not every space, right? Like there's no autonomy out there, which can take like, you know, what, a one or two killer robot and run it around children safely. I think that's not an off the shelf tool yet. But around a factory around a mine. Now you can do that right now. That's awesome. So you're saying like two years ago, it wasn't the case that we could do this? Now? It seems to be good. Where do you think it's going kind of projecting out? Say the next two years, next five years, 10 years, whatever it might be? Where do you think the community is going? Well, I think in general, I mean, I think in general, it's this is funny, because I actually asked a very similar question, when I interview people, for for jobs at our company send out the tables have turned, I think, I think in two years, we're definitely gonna see areas specific markets, like us ours, for example, factories, where putting robots in, or mobile robots and, or collaborative robots in is going to be a pure business decision. We have already seen this now, where people have stopped asking us to prove that the localization is going to work, for example, right? Like we either say localization works, and it does, or we say that it's probably not going to work, and we won't sell them, or we won't try to sell it, right. Like that's, but more and more, it's going to be the same calculation of always, it's just like, it's just gonna be a business decision, right? Like, robots are just something that I buy off the shelf. just purely pragmatic, which to a lesser extent, it's kind of that way with robot arms still. But in both cases, the other thing that makes that possible, is the significant decrease in integration time. And cost. And robot arms will still have a little bit of a challenge there, although many companies are starting to are starting to work around that. But you're going to see very low integration times. And because of that, it's going to be a lot more commodity like to just put some robots out in the out in the world. Five years, I mean, five years, I think we're really going to start to see robots in earnest within the general public. Again, maybe not, you know, maybe not in the daycare, but you are going to see them out in earnest in the general public doing useful things. So I don't mean in science fair projects, I don't mean little things like that. And I know that there are various trials, there are various trials of delivery robots and things like this, but I think it'll be a there'll be much more pervasive, much, much more pervasive. And we will be able to say that in certain cases, and probably not on road self driving, sorry to say, but concerning cases, the safety story is effectively salt, I think we'll be able to say that. And when you're there, you know, what that means is over that, that two to five year timeframe. That's where the real app like that, you know, the ecosystem becomes really powerful. Because you know, now we're at the point where you can write apps for our robots in a factory right now. But first, you need to get into the factory, but you can write it right now. And but now, you know, two years, five years from now, they're really going to be in front of everybody. So you'll be able to script against one of those robots. As as much as everything else, and maybe what I'll say is people might look at this and say, Well, you know, why are we saying this now? Right? Like, didn't people say this 10 years ago and 20 years ago, whatever it like, absolutely, they did, right. Like there was a glut of robot App Store companies. You know, eight years ago, nine years ago, roughly, around the time of the turtlebot, one launch. And then there was, you know, Microsoft heavily invested in robotics with Microsoft robotic studio. And that was, that was a 2000s. I think that was, but one of the things that's happening now is that all of those apps couldn't rely on safe, reliable autonomy. And then, and sort of think about his assault kind of thing. That's the that's the bad word, right? Like, yeah, you couldn't you couldn't, you need to have a target, which is useful, like a device target, which is useful. And now, that didn't used to be the case, right? Like when these app stores started, started up, you're just kind of assuming that a bunch of people on their own, we're each going to come up with safe, reliable autonomy. And that's, that's hard when you're moving, you know, hundreds of kilos around. But now now, no, I think we're in a much better place like that autonomy is well, I have, you know, many million reasons, million hours of operation to say that this is possible, right? Hmm. That is exciting. So wrapping up, do you have any? Do you have a Twitter or how can people find you or any any public links, you'd like to share anything? They can find me on LinkedIn? I unfortunately, do not. I am not part of the the Twitter, whatever, whatever, whatever it's being called. But yeah, I'm not on Twitter. I'm not on Instagram, but they can find me on LinkedIn. And in my email, which which will be needed to connect is Ryan at clear path.ai. So they can also find me on that email. But LinkedIn would be preferred, I believe. Awesome. Okay, I've enjoyed this. Thank you. Thank you very much. That's all we have for you today. I enjoyed Ryan's answer to where thinks robotics will be in two to five years. Where do you think robotics will be in two to five years? Let us know in the comments on sensing gackt.com or discourse. Thank you again to Open Robotics for being our founding sponsor. Bye, everyone.