Ben Dart on 2023 Trends in Robotics at ICRA

Introduction [00:05]

Roland Meertens: Welcome everybody to the InfoQ podcast. Today is a special episode where I discuss the trends I see in robotics with Ben Dart. However, to discover what the trends are in robotics, we decided to visit ICRA, the International Conference on Robotics and Automation. ICRA is an academic general robotics conference, which attracts over 4,000 people who are working in both research and the industry. The conference is very large and has both an academic part, where researchers are publishing and discussing their research, as well as a more commercial part, where robotics companies try to sell everything from complete robotic solutions, robotic parts and special software. Or maybe they are just trying to recruit the best of the best of the robotics engineers who are walking around on this conference.

As I already mentioned, I visited the conference together with Ben Dart. He's the Team Lead of Robotic Perception at the company Emesent in Brisbane. He has a lot of work experience in writing software for everything from mining apparatus to self-driving cars, and is now working on autonomous drones and walking robots. He's always working on the cutting edge of software for robots, and due to his wide range of experience, he has a good view on what the state of the field is and where the field of robotics is going. Without further ado, let's listen to what his impressions of the conference were. Ben, welcome to the InfoQ podcast.

Ben Dart: Thank you, Roland.

Roland Meertens: We just went to ICRA a couple of days, the International Conference for Robotics and Automation. What were your overall impressions?

Ben Dart: I have not been to a conference of this scale before, so the impressions were, "Holy moly, there's a lot of very cool robots that I didn't know about." We spent somewhere in the region of five hours just walking around trying different robots for a variety of different purposes and they were all so unbelievably cool with a huge amount of unbelievably passionate people driving these booths. And largely engineers that just wanted to talk to other engineers rather than... It wasn't a sales event, which was also really very cool.

Telepresence robots [02:20]

Roland Meertens: It was definitely a really cool experience of us trying all the robots together. I think one of the things we tried a lot and which was definitely new compared to... Or the amount of companies working on telepresence robots this year was far higher than last year. Which ones did you like the best, or which ones did we try?

Ben Dart: We tried a couple of different ones. So the Reachy Robot by Pollen, the robot... A very cool concept, so doesn't need to walk or do anything like that, has a mobile base using omnidirectional wheels and then arms built simply from a series of servo motors. It's super cool because their purpose is to deliver it as a research engine. And so when I was talking to the woman, she was like, "Yes, so for that nature you can buy it from us, which is the simplest way to do it, and we provide it for a very reasonable price." But the whole thing is also just open source. The mechanical design, the software, everything for that robot was open source.

Roland Meertens: Oh, so you can build your own telepresence robot at home?

Ben Dart: Yes, you can, which is crazy. What a cool idea. I mean that's quite enabling. And then it worked quite well. You used a VR headset, the remotes would track your hands, and then you would basically manipulate the arms using that and then you could close and open the grippers using a button. And that was just so much fun. I don't know where you would use it, outside of research purposes or for fun, but the concept is quite cool.

Roland Meertens: We also tried Shadow Robot, which was basically two hands which you would control individually through your fingers. What did you think of that?

Ben Dart: Yes, it was a super cool concept. The motion capture system that they use was excellent, so it really... I guess the fidelity with how when you moved your hands and you moved your fingers, it tracked really, really well. And when I was talking to the guy, I'm like, "So how good is it? Can you pinch things and pick things up and do things as if they're your real hands?" And he said, "Yes."

At the moment in the state that we used it, which was already incredibly impressive, it was fully uncalibrated, he said that there's significant effort to make sure that it matches the size of your hands so that when you do movements, they track really closely to your hands. And that wasn't done because it would be a huge waste of time to do that for a demo at a conference. But even without that, you could pinch things and pick them up and move them around.

Roland Meertens: We could take these very tiny cubes and just build a whole tower out of them. And I think you even managed to grasp three cubes at once with two fingers.

Ben Dart: Thank you. Yes, I did achieve this, which is very impressive. It was quite intuitive. You think about this sort of thing and how long would it take you to learn how to use a robot hand to pick up some stuff? But I don't know, we were doing all of this 60 seconds after we started playing with it.

Roland Meertens: Yes, I think that that's one of the more impressive things is that we managed to... In the past, whenever you want to try to use a robot remotely, someone would always have to tell you, "These are the buttons, these are locations," et cetera, et cetera. And we are now just strapped into some sets and immediately understood how to use it, which is a big leap from how it was in the past. You also said something about the collaborative nature of these robots, so we could just walk around them, which might sound very unimpressive to people who are not into robotics.

Ben Dart: Well, Yes. I used to work for an industrial robotics company, and that was six or seven years ago. And at that time, fully collaborative robots were very few and far between and not particularly well-thought-out. At that time, if you walked into a room and saw people just swarming around these robot arms, we're talking about the Shadow Robot with the hands. The hands were connected to Universal Robotics, six-degree-of-freedom, industrial robot arms. And there was a time where it's inconceivable that you would just be walking around and interacting with these robots without safety cages and cutoff switches and all this sort of stuff. And now almost every stand, well not almost, a huge percentage of the stands had just running collaborative robots, which is super cool.

Roland Meertens: Yes, there were so many robots just walking also through the audience. I think we saw one incident where robots just somehow didn't respond anymore to things and just kept walking.

Ben Dart: Yes, it cleaned up a catering stand.

Roland Meertens: Yes.

Ben Dart: Took it with it. It was hungry, hungry robot.

Free roaming robots [04:40]

Roland Meertens: But it was also very tiny robot, so it couldn't really do a lot of damage. But besides that, the larger robots, they're all just roaming freely through the audience.

Ben Dart: Yes, couple of wheeled robotics, as well. We played with a balancing wheeled robotic that can shift and balance its weight, and that was also extremely cool.

Roland Meertens: And also there we asked them, "Can we try it? Can you hand us the controller?" And they just immediately gave it to us and we didn't get a whole safety briefing. They just told us "This is forward, this is backwards."

Ben Dart: Takes a little bit of trust, but I think that you've got a lot of people that are enthusiastic but serious about-

Roland Meertens: But does that mean that we will see these robots now more often on the work floor. Do you think your colleagues will start using telepresence robots all day or?

Ben Dart: I think there's a long way to go before we get to that. I talked to each of these prospective telepresence robotics companies and the business case was commonly for giving to research organizations rather than for companies to use for specific teleoperation purposes. It was quite interesting because, at least for me, I didn't realize there was such a huge market for research organizations, but there is. The people are doing some crazy things with these robots. When I was talking with the Shadow Robotics guy about this, he was basically talking about a large quantity of his customers, they provide an API endpoint, want to try and integrate some ML into this, some sort of AI process to do something. We didn't elaborate that on too much, but I think that that's going to be interesting with whether or not they go deeper into this with having these machines rather than being telepresence-controlled, being controlled to do some repetitive tasks or something like that.

Roland Meertens: And even for a telepresence, I can imagine that it will be a thing where, for some simple jobs or if you want to talk to a coworker at work, it'll be easier to just drive some robots to them instead of sending them a Slack message. You can immediately physically poke them.

Ben Dart: Now you can tap them on the shoulder or accidentally crush their collarbone.

Roland Meertens: That could happen depending on which robots you're using. Yes, I think then the smaller robots we tried are definitely better than the massive robot we tried. I think actually last year at ICRA, a friend of me couldn't attend, so she logged into a telepresence robot and then drove behind me so she could also talk to the people at their conference floor. And at some point, I logged into a different telepresence robot so we could talk to each other over two different telepresence robots. That was quite a good moment.

Ben Dart: What an interesting concept. We didn't see too many of the traditional... Because there are several companies doing traditional telepresence where it's just a robot with a screen on wheels that you can pilot from a web browser or something.

Roland Meertens: But I have the feeling that she didn't get a lot of replies from people. It took a bit of convincing for people to try to talk to an iPad. I have the feeling that as soon as you have hands, that would change a lot because then you're really feeling like a physical presence, for other people as well.

Ben Dart: The X prize winner was I guess a telepresence robot that you sit in a chair and you control the hands. And from looking at the stand and watching the person operating it, it appears that you use your feet to turn a wheel and then you have a pedal or something like that and that allows you to drive the robot and rotate it. And then you have arms that copy your motion and you have a VR headset and the robot has a camera and a head that moves around and a screen with your face on it. And that would give you quite a high ability to move around inside an environment and interact with that environment that companies, like Anybots and BEAM and some of these other ones that produce the screen, maybe will want to adopt something along those lines.

Drones [10:07]

Roland Meertens: Yes, I can imagine. Anyways, your line of work is more working on drones. What did you think of the presence of drones?

Ben Dart: The drones that we saw, I think we only saw one major drone stand and it was super-duper cool for a variety of reasons. And it was a bit of a shame at this conference that they don't help to provide an area where they can do drone exhibits. From talking to the one drone company that was there, they were talking about how it was quite an effort to get even their tiny, little, essentially harmless drones to operate inside the arena. And they had a net and everything, so it's perfectly safe, but I mean a net... There's extremely safe drone nets and you can have tethered drones and all that sort of stuff. There could be a place there where they have more drone exhibits because they're both super cool and it's a large part of the sector. But the one that we did talk to, they had two things that I think were unbelievably cool. They had the Flappy drone.

Roland Meertens: Maybe quickly explain where the Flappy drone is.

Ben Dart: Traditionally, all of these little drones are often quad rotors or multi rotor vertical motion drones. This guy was essentially an ornithopter. It had wings that it flapped back and forth and that.

Roland Meertens: So it flies like an insect.

Ben Dart: It flies just like an insect, Yes. Again, a bigger drone arena would've... Because when I saw it in motion, it was flying with the wings facing down and it's able to create thrust vertically so it can take off and land. What I didn't know until I talked to some of the people afterwards was that it can actually transition to horizontal flight and fly extremely fast horizontally, which would've been an awesome thing to see. That would've been quite cool. We'll have to look up some videos about that after. Company is called...

Roland Meertens: Bitcraze.

Ben Dart: Bitcraze. The company that I work on does drones, but they're much bigger. Usually a 50 centimeter diameter or bigger. And we carry a payload that does high-density mapping of environments and we have a considerable amount of processing power. I don't necessarily think it's a lot. You and I have both worked in self-driving and in that industry you have a insane amount of processing power available to you because you can put a lot of computers in the back of a car.

On a drone, it's really, really hard because you have to lift everything that you use and if you have a big graphics card, it draws power and it's heavy. So you're in a super constrained environment. We thought we were constrained. We're doing some pretty big algorithms, doing a lot of path planning through unknown environments, high quality slam to make these maps, all of this stuff. But then when we saw this drone company, these guys have an eight-bit microprocessor that they're doing machine learning on. There's a couple of papers that we saw of students that are doing inference engines to find other drones and do optical flow, but on a computer that it's almost an Arduino, it's so small.

Roland Meertens: The students would then distill their neural networks and quantize their neural networks because their processor couldn't do floating point operations and they can still get a lot of cool features onto a tiny drone like this.

Ben Dart: Which is really cool. And for me personally, a little inspirational because we don't tend to do so much ML because we don't have access to big graphics, like graphics cards make running inference engines a lot faster and we don't have access.

Roland Meertens: Normally people use technology, or this is built on technology such as CUDA or TensorRT, which are fast ways to do inference of large neural networks. And these people manage to get similar performances with very small neural networks.

Ben Dart: Which is super encouraging for us specifically because we don't have access to a massive graphics card. We have a little one and we've got a pretty good processor, but if they can do this sort of stuff on that tiny little computer, then we can hopefully do some pretty cool stuff on ours, as well.

Roland Meertens: That's the thing, that even if you're not working in robotics, you can still use the same technology to really reduce the amount of servers you need in your server as a normal server company. Yes, this technology is definitely... It's very inspiring to see this.

Ben Dart: And then if you are doing that sort of thing, AWS is money. Every time you run something on AWS, it's money. If you reduce it, you can essentially save a bunch of money as well, which would be quite an interesting path for this to take.

Legged robots [14:18] Roland Meertens: What did you think of the legged robots we saw? I think most people know the Spot by Boston Dynamics, but on this conference there's a lot more companies displaying their legged and balancing robots.

Ben Dart: Apart from the one that destroyed the catering sign, they all look quite good. This is of interest to me because we also do some stuff with Spot and it is a really good platform, but it is quite expensive. It's not a cheap platform. A good question is will there come up applications that the Spot is good for, but you maybe don't need something as robust and as technologically advances the spot to achieve. And then maybe it's something that's more achievable for companies that are smaller and can't afford to invest in robots like the Spot.

We also saw some that are combination of quadrupeds with wheels, which I think is really cool. I really think the quadrupeds are super cool because the problem of foot placement and how to walk around is a crazy hard problem to solve. And Boston Dynamics have done an amazing job, but wheels are awesome. We have invented a mode of transport that is crazy good compared to walking around. And so you see these quadrupeds that are both quadrupeds, then they have wheels. And that would be an awesome combo because I'd love to see you can transport around on your wheels in most situations and then if you need to change elevation using stairs or go up a slope or something, then can you retract the wheels and walk or just walk on top of the wheels. Some of these combo robots are quite cool.

Roland Meertens: We asked Boston Dynamics and they told us that they had about a thousand or over a thousand, maybe 1500 or something, robots in the real world. Do you think the adoption will be way higher once they get views? Or what do you think is hindering the adoption of this by everyone? Is it now mostly robotics enthusiasts using this? Is this some early adopters?

Ben Dart: I think it's mostly research organizations. Maybe I can talk about our personal use case and then talk about some of the stuff that we talk to people about. But we use Spot because we're interested in doing environments that are highly uncontrolled. An example would be underground mining. The Spot represents the ability to walk around on ground that you don't necessarily know about. There's like uneven surfaces or rocks or stuff like that, which a wheeled robot won't do. In our case, something like a wheeled one would only be useful for transiting to and from the areas that you can only pass with your legged form.

Herding the adoption is... I would say there's two reasons. There's the price point and there's the use case, what exactly is this useful for? It seems like what's coming out of this is that the Spot is incredibly useful for routine inspections and that sort of stuff. There's examples of Spot being used in hydro plants in dams or in nuclear facilities where you basically have a Spot that's docked, it has a camera on the top, you program into it because it has the API gives you a native API for Waypoint following and this sort of stuff. You basically walk it around your factory and you take pictures of specifically interesting things. A valve that has to be off all the time or something that you've noticed in a previous inspection you want to check again.

And this is a really cool concept for the Spot because you can just deploy this one robot inside a much larger area. It can walk upstairs, it can move around and do lots of stuff, but if it's a routine inspection, you can solve the same problem by just having cameras everywhere. And putting 500 cameras in is still cheaper than buying one Spot and potentially simpler. It's quite difficult to come up with a use case where that very expensive robot is better than some of the solutions that you have, especially in controlled environments where you can change everything about it.

Roland Meertens: They told to us that the biggest competitor is people just directly changing the environments, just directly placing a camera.

Ben Dart: That's why the search and rescue stuff, going into uncontrolled environments like caves or in forest areas or stuff like that, it's very interesting because controlling a forest is a bad idea. You just have to find the right reason to use it and then I think it'll be a super successful type of robot.

Roland Meertens: Nice. Maybe talking about AI, machine learning algorithms that we see. Any companies who were selling anything specifically? What did we see people do with software and emerging technologies such as this?

Ben Dart: I was mostly focused on some of the hardware aspects. Because there's a lot of people selling platforms, but there weren't too many people selling direct software algorithms. We did talk to the people doing the software calibrations.

Roland Meertens: I think the calibration is indeed very important that one thing which is different from maybe the normal software industry is that in the normal software industry it is still hard to build data sets, but at least your sensors are always good. In robotics, you also have to make sure that your sensors aligned to each other and that seems to be a difficult thing. What I did notice is that the robotics still has a really interesting mix between people using deep learning methods and people using traditional algorithms, which just work a bit faster on the low processing power they have.

Ben Dart: Yes, Yes. In terms of companies, we feel like we didn't see many, but there were plenty of students that were struggling with this. We talked with a couple of different of the poster presentations that were doing machine learning algorithms and had some really cool concepts of things to do with them. And then the one that stuck out for me was there was a poster about finding emergency landing sites, which is relevant to me. I work in the drones. The drones go into places where you can get into a region where you have critical battery or some error and you need to land, but you don't know where to land. And at the moment, the approach for most drone companies like us, DJI, everyone is just fly down. But if you're above a tree like a canopy, then you're going to fly down into trees, which is bad for your drone.

Roland Meertens: I know that feeling.

Ben Dart: There's plenty of papers that are dealing with these sorts of things with using LiDAR data to calculate areas that are safe to land. And the one that we talked to and specifically was doing it with a traditional approach, finding an estimating normals and trying to calculate ground planes. But Yes, I can see that there's slowly plenty of people that will be working on similar things using ML approaches as well, because it can give you very high performance and also be quite fast. Some of the traditional methods are quite hard, computationally. In these drone platforms you don't have an RTX 3080, some crazy graphics card, just sitting on there, so you need to be able to do it fast and with small amount of hardware.

Self-driving cars [20:37]

Roland Meertens: How about self-driving cars? One thing I noticed in the last years I went to this conference was that there were a lot of self-driving cars companies advertising, from Waymo to Tesla to Argo. How was it this year?

Ben Dart: I think we only saw Waymo and Wayve. As with the industry itself, feels like it's dropping off a little bit. The Wayve one was quite good. When we talked to the people at Wayve, they had quite an interesting demo about you could drive inside of their simulator, which was cool. And they had basically used a nerf to generate three-dimensional visualizations from real drives using just a camera.

Roland Meertens: They drove through a part of London with their real car and then trained a neural network to create these neuroradiance fields to regenerate an environment around you so you could drive through the regeneration of the learned environment.

Ben Dart: Which is quite a cool concept. And then talking with them, it was super cool because there was a bunch of them there and they were super keen to answer questions and we got into technical discussions about how they're trying to solve some of the safety issues and stuff. And they were really upfront with this is work in progress and they weren't trying to spruik up or lie about anything or try and cover up that this is an insanely hard problem and there's going to be challenges, which was good. We didn't visit the Waymo stand. It's there, but they didn't have a demo of any sort.

Roland Meertens: It was a very small stand. The stand was very small compared to previous years, I feel.

Ben Dart: Yes, maybe they're at capacity, don't have any hiring in the pipeline or something like that.

Roland Meertens: Yes, I don't know. I don't know if the reduction in self-driving car companies here is due to money or because the problem is harder than I expected or because they solve the problem. We don't know.

Ben Dart: If they solve the problem, I feel like that will be an exciting ICRA 2024. Everyone gets their self-driving cars. I don't know. From what I understand, it's still an incredibly hard problem, even with the constrained operational environments that people are trying to operate in. It's a space that we've both moved out of, but I'm very excited about it still. I just hope that more cool things happen rather than the whole industry collapsing.

Roland Meertens: Indeed. I hope the same. It also looks like there's still progress. We still saw a lot of students working on the same problems. There still is progress in being better at detecting what's happening in the world around you. Maybe then going to a more easier or a more directly applicable program? Agricultural robots?

Ben Dart: Agricultural robots is something that's growing really quickly. I've talked to some people at a company called SwarmFarm. They were not present at the conference, but I've talked to them separately. And agricultural robotics is an incredibly interesting problem. At the moment, you have a lot of people from big companies working on it, John Deere and this sort of stuff. And they present a lot of really, really cool technology to do with farming, but using people-driven tractors and all that sort of stuff, but using technology enhancing that. Potentially they have an agricultural robotics program, I'm not sure, but they were the only people present at the conference and they didn't seem to have information about that. But it's an incredibly cool robotics problem. It's like a semi-constrained environment, so you have some of the difficulties of stuff like self-driving cars where you have to navigate in terrain where there might be obstacles or the traversability of things is really hard.

It's a farm, mow every day. Sometimes you get really long patches of very long grass and it's like, "Can I drive through that or not?" Is a very, very hard problem to solve. And then on top of that, some of these robotics companies are working on solutions that should hopefully be extremely environmentally friendly. An example that I talked to a SwarmFarm is that they basically working on, or have worked on a system where... Currently, how they weed or do pest control and weed a crop is either fly over the top and dump pesticides and weed killer or they will drive through the middle with these big booms that come off the side of a tractor and spray everything.

There's a lot of companies now working on can we detect weeds and then spray them individually or even kill them with a laser. There's a company that has managed to kill weeds just using laser, which is incredible. That means zero pesticides. And if you can't use the laser and you can just spray an individual weed, you might save hundreds of tons of pesticides and that's all stuff that's not then going into the topsoil and ruining it.

As far as I can tell, a lot of agricultural robots are aiming for small, modular robots. Instead of these massive tractors that just tear up the ground and cause a lot of damage, you have these little robots that just can navigate through the crop, doing minimal damage, not turning over the topsoil, super accurate. And this is something that when I've talked to people, there's always the concern, "Will this replace people's jobs?" But the farmer sits in the tractor all day and does this and then has to go home and do all of the rest of his work for her work. The concept, I hope, is quite adoptable and will be more environmentally friendly and save us a lot of pesticide use and all this sort of stuff. There wasn't that much representation of this stuff at ICRA, as far as I can tell. It would be cool to see some of that next year.

Roland Meertens: What I did see on one screen, but I don't think we really talked to these people, was a company which also try to monitor the individual health of the plants. If you have a crop, you could drive over your fields, then map every plant you have and then also monitor that. You could also identify some areas where you want to have a bit more water so that these plants also grow faster.

Ben Dart: You can find diseased ones and then get rid of that part of the crop before it spreads out.

Roland Meertens: You can very well see in your massive fields where do you have which problems. That also seems to be an up and coming fields where I don't know how far we are from commercialization and this being used every day.

Ben Dart: Yes, I think that several companies have pilot programs with farms, at least in Australia. Companies like John Deere own crazy amount of the stake in farming equipment. It's like 70% are owned by three companies or something of the market share in the world. Don't quote me on that statistic, but it's a big number. There's a lot of competition that can come out of these little robots and if they're effective, it'll be a really cool space to be in. The path to commercialisation [26:54]

Roland Meertens: Yes, indeed. ICRA is more a research conference, which also attracts a lot of companies. How far away are we from commercializing this? Where are we on the chasm? Where are we on the bell curve? Is this purely robotic enthusiasts? Is this early adopters? What technologies we saw, do you think, are most ready for being used by everyone?

Ben Dart: There's a couple of straight-up companies that are there, Dyson and John Deere and stuff, that just sell a product and maybe they're there for recruitment purposes or something, and so that's quite interesting. But a very large portion of the other bits, I would say, are very low on the bulk of some of these teleoperation robots and some of the drone robotics companies and a variety of others. When you talk to them and you're like, "Who do you sell these to?" Even Boston Dynamics, a large portion of their robots get sold to the research organizations, which I guess are at the bottom end of the bell curve. They're absolutely on trying to develop these technologies into something.

I couldn't put a time on it, but I don't know how long it takes. Usually, if you come up with a really good idea... And it's a hardware startup, so it always takes longer, but something like these hands we're starting to see... Maybe there was four or five companies that are producing robotic hands for researchers? So they're selling to researchers. At some point, someone in a research program will come up with an incredibly good idea. They'll spin that out and then that will become popular. But that's got to be many, many years away. I'd say Yes, early on the...

Roland Meertens: Was there anything where you thought, "This can totally be something which we can all buy in the next five years?"

Ben Dart: Something that we can all buy? Yes, as a company we could potentially buy some... A lot of the legged robots companies are starting to investigate what they can do with it. Certainly, if you are a hydro dam owner or something, there's going to be applications that Spots and legged robots are used for in, I think, the near future. Certainly, there's a lot of companies investing a lot of money into trying to find the right use case to be able to sell lots of those.

Roland Meertens: Yes, the legged robots are definitely... We see them improve every year. It's amazing to see how fast that goes. And the amount of spinoffs and competition, which is starting, is also incredible to see.

Ben Dart: Yes, Yes. They're actually also now just running student competitions where we saw some of the... There's an obstacle course for legged robots where I guess you pass playing through it.

Roland Meertens: I think the National Institutes for Standard and technology has set this up, so they are now trying to figure out, can we find a good metric for how useful these robots are for search and rescue? They set up a whole competition where a lot of universities can compete with their legged robots to see how useful is. But the fact that people are thinking about standardizing the technology helps a lot I feel for the adoption. Then maybe last thing, if I want to work with robots or if I want to get started with something, what are the entry level robots? What should I play with? What should I buy? What should I try to use in my work?

Ben Dart: We went and saw several dev kit robots. There was a company, Viam, I hope I'm saying that right, which develops a platform, a more sleek version of the turtle bot, but they provide a really cool API for controlling the robot over a web interface and this sort of stuff. There's plenty of the drone ones that produce the little hobby drones. It's hard to say. As a robot hobbyist, some of these were out of reach for me, personally. Unless I had a really good idea for something that I would go and do, I wouldn't go and pay the four or $500 Australian dollar, convert that in your own terms, for a robot just to mess around with at home. I would want to have some good idea. But certainly for companies investigating interesting solutions, quite a good little platform there.

But Yes, plenty of people do invest in that. I talked with one of the people about, there's a platform called Teleo, which makes drones that you can program with a scratch language to fly around. And this sort of thing is a super cool idea because there's plenty of robots like this, as well, for driving robots. You can teach younger people about robotics and teach them that, how to make logic and make code in the simplest format to do really cool things. I think maybe Parrot does something similar. But that sort of thing, we did see that around a little bit as well, and that would be super cool.

Roland Meertens: I absolutely agree with you on that price point for entering this technology that if you want to learn programming, you can do it using any laptop, just starting with JavaScript. If you want to play around with robots, it will easily set you back multiple hundreds of euros before you can do anything very interesting.

Ben Dart: I guess it's a steep entry point. It's hard, right? Because it's hardware, you have to buy something. There's plenty of stuff that people do in simulation, you can easily get something like Gazebo going and do robotics inside simulation, but it's not...

Roland Meertens: It's not what makes the field exciting.

Ben Dart: Exactly. When it's driving around your house or flying into your walls or whatever that that's when you've got the excitement.

Getting started with robotics [31:50]

Roland Meertens: I also think the one thing you mentioned before, the turtle bots, they are kind of beginner robots. The turtle bots are often one of the first programs you write where you can just tell a robot, "Go forward, go backwards, read what is happening with your sensor data and based on that make a decision." You have this whole perceive-and-act loop. I think those are the closest things we have in robotics to a standardized learning platform. But for computers, people can nowadays just buy a Chromebook or buy an iPad and get started with that. So you have a standardized platform. Would there be a need for something like this in robotics?

Ben Dart: Do you mean for software or for the hardware platform?

Roland Meertens: Both.

Ben Dart: I think ROS makes up the software platform, if you're going to do robot. I don't think I've ever met anyone that in this field that hasn't heard of ROS. So if you Google write a robotic program, you'll end up at ROS. And they provide pretty good tutorials for how to do a lot of simple things, and they provide a huge tool set to be able to do all that. That feels like if you're starting off in the software world, you'll end up going through there, at some point. For the hardware, it's really hard because there's so many different types of robots. You can have your dried wheel robots or your omnidirectional ones, and then arms or drones or cars.

Roland Meertens: Yes, indeed. For ROS, it always is very easy to get started with something. But in my experience, if you have your own kind of cheaper custom hardware, that's also when you have to start your own modules in ROS, which is a bit of a shame.

Ben Dart: ROS provides a huge number of drivers for some of these, and a lot of the companies that sell introductory robots provide, I'm guessing, a ROS API. It's a clever way to do it. But Yes, you are right. If you go to eBay and buy some extremely cheap robot arm that's made of servos, then you're going to have to write your own CANBus interface to it, which is a stepping stone. It is hard to find that combination of really cheap hardware that you can easily start with without spending $600 as someone that's never played with one before, as well as getting the easy to introduce software.

I would say as well, it depends on your level. You always need to start somewhere, but if you start with just trying to develop ROS nodes, you have to program in usually Python or C++ or learn how to interface to ROS through one of the packages that provides the API to other languages like Rust or whatever. As far as I know, I don't know if there's a scratch interface that allows you to write a ROS node. There's also the whole, you have to learn how to program a little bit before you can start using ROS, which I guess is a little barrier.

Roland Meertens: In the past we also had the Lego NXT robots. Those were really amazing, but I think that Lego killed this project a couple of years ago.

Ben Dart: Oh, really?

Roland Meertens: Yes.

Ben Dart: The Mindstorm?

Roland Meertens: Yes, the Mindstorm.

Ben Dart: No way. That was so cool. We even used that at university. It's such a good concept because you can just build a robot arm and then control it and then break it down by the end of the lesson.

Roland Meertens: Yes, those things were really amazing and also had a relatively good price point, but I think they're, they're not there anymore, but I could be wrong. Yes, I think they stopped the division, which is a shame.

Ben Dart: If you want to talk about the Hello Robot one as well, the other cool thing about that was just simplicity. Some of the robots that we saw with the big arms and six-degree-of-freedom robots and stuff are complex and hard to get into. But there was at least one demonstration from a robot that's the simplest format of robot you can have. A diff drive base, two linear actuators, a couple of servos. It's robust, provides a really easy API, and I think they have a right-to-repair sort of deal as well. They don't stuff you around if it breaks, you just fix it yourself.

Roland Meertens: So basically their robot was a stick, which could go up and down and left and right. And what more do you need to grab a cup? That's literally everything you want. It has a very tiny arm which can open and close, so you just... You could go anywhere and reach any heights and it was all very simple, just perfect.

Ben Dart: When I talked to the guy... Well, maybe they had it advertised, They have a ROS API for that, I believe.

Roland Meertens: I think so.

Ben Dart: Pretty easy to get into.

Roland Meertens: Yep, all right. Then thank you Ben for visiting ICRA with me, and thank you for joining the InfoQ podcast.

Ben Dart: No worries. Thanks, Roland.

Roland Meertens: That was the interview with Ben Dart. I hope you enjoyed this special episode, and I hope you have a better view on what the trends in robotics are. The field of robotics is still small but very much emerging when looking at what is happening at conferences like this. I would like to create more content for InfoQ around the field of robotics, so please show me whether you liked or dislike this episode and what you want to know more of. That said, thank you very much for listening, and thanks again to Ben for joining.