Latest / EDGE AI POD / From Arduino to AI: A 20-Year Journey with David Cuartielles

Transcript

From Arduino to AI: A 20-Year Journey with David Cuartielles · EDGE AI POD

• 0:07
  Speaker 1: David, welcome.
• 0:07
  Speaker 1: Welcome to our partner recording AGI partner.
• 0:14
  Speaker 1: I don't know what we're going to call this thing, but thanks, good to see you.
• 0:17
  Speaker 1: I think I saw you last in Austin, texas, at our multi-day event down there.
• 0:28
  Speaker 2: Did you enjoy Austin?
• 0:29
  Speaker 2: Yeah, you know, I lived there for a while before going to this event, so I I took the opportunity to visit some friends, you know.
• 0:34
  Speaker 2: Nice, yeah, I mean I think it's a good place, it's a fun place to be.
• 0:38
  Speaker 1: It's cool.
• 0:39
  Speaker 1: Yeah, there's like 5 000 tech companies in Austin now, so it's really become kind of an interesting little epicenter.
• 0:46
  Speaker 2: Yeah, we have our own office there.
• 0:48
  Speaker 2: It's more focused on sales and especially like dealing with corporate customers, you know like pre-sale support and these kind of things.
• 0:57
  Speaker 2: But yeah, I mean it's good to be there.
• 1:02
  Speaker 2: I spent six months there.
• 1:03
  Speaker 2: I got a chance to visit a lot of people and get to know, uh, the ecosystem and what's happening and so on.
• 1:09
  Speaker 1: But yeah, clearly it's a good live music.
• 1:12
  Speaker 2: Some good live music as well yeah, yeah, yeah, I'm saying pretty good oh, cycling, yeah, that's true.
• 1:21
  Speaker 1: So, uh, let's talk about ar, arduino, and you're probably one of the best people to talk about Arduino, since you're a co-founder of Arduino, and I'm sure everyone who's listening to this has heard of Arduino, because you guys have been around for like what?
• 1:35
  Speaker 1: 20 years.
• 1:35
  Speaker 1: Now.
• 1:35
  Speaker 1: Is it your 20th?
• 1:37
  Speaker 2: anniversary coming.
• 1:37
  Speaker 2: Yeah, 2025 is 20th anniversary.
• 1:40
  Speaker 2: Actually, this morning I was reading it's also the 20th anniversary of JIT, so it's like JIT and Arduino were born actually with like a month's difference.
• 1:48
  Speaker 1: Interesting 2005 was a special year.
• 1:54
  Speaker 2: It looked like I guess we had no big international conflict so we could focus on development instead.
• 1:59
  Speaker 2: That's true, yeah.
• 2:03
  Speaker 1: But yes, go ahead, go ahead, you can tell no.
• 2:06
  Speaker 2: No, it was like yeah, we're turning 20 years.
• 2:10
  Speaker 2: Our community has grown a lot.
• 2:12
  Speaker 2: Now we hire people that studied with us at high school.
• 2:17
  Speaker 2: So, it makes me feel pretty old, I have to say, but it's a great feeling.
• 2:23
  Speaker 2: It's a great feeling when you interview people and you realize that they were into the philosophy of our project when it was just an open source project before it even became a company, and that's, I think, quite special to grow with this.
• 2:39
  Speaker 1: ALEX LANGSHURSTENBACHEVILLE-.
• 2:40
  Speaker 1: Right, well, arduino has educated a generation of developers, basically, which is pretty cool, and you guys started, as I understand it, so Arduino is named after a cafe that you guys used to hang out at in Ivrea.
• 2:54
  Speaker 2: Yeah, well, long story Coffee shop.
• 2:56
  Speaker 2: So actually, arduino was the first king of Italy, right, that's right In the year 1000 or so?
• 3:01
  Speaker 2: Right, that's right About the year 1000 or so, yeah, yeah, and that named half the places in this village, or well, as soon as I say village, it will be upset with me, but it's like the small city of Ibrea in Italy.
• 3:11
  Speaker 2: And then, yeah, there is an Arduino square and there's the Arduino cafe and we used to sit there, and so, when we're trying to name the project, we actually we were not super creative, we were just trying to have a name to raise an open source project and Arduino was not taken and we used it.
• 3:33
  Speaker 2: Yeah, it was a coincidence you know, there you go.
• 3:37
  Speaker 1: Sometimes that's the way the best names come from, just a happenstance, you know, and they stick.
• 3:41
  Speaker 1: But yeah, no, that's pretty cool.
• 3:44
  Speaker 1: And um, actually, um, yeah, so I, actually I was going to say we're going to be, we're going to have our event we just talked about our event in austin.
• 3:52
  Speaker 1: We're going to be in back in italy, in milan, in july, which will be exciting to go back there too as well.
• 3:58
  Speaker 1: But, um, so, yeah, so you started I'm interrupting you guys.
• 4:06
  Speaker 2: No, go for it no what I'm saying.
• 4:08
  Speaker 2: Like a minute ago, before jumping into this recording, I was just talking to some of the people that was in a DyniML or HEI workshop last week in Malawi like sponsored by the foundation, and we know yeah.
• 4:20
  Speaker 1: So you were in Malawi.
• 4:21
  Speaker 2: Yeah, yeah, yeah, I was there.
• 4:24
  Speaker 2: It was actually really good.
• 4:25
  Speaker 2: The students that participated in the workshop, and some professionals as well, were exceptionally good.
• 4:31
  Speaker 2: They came with really great ideas and we can see how the content that's been made for the workshop by Marco Senaro from ACTP and Marcelo from Chile and Brazil and the guys from Slovenia, joao and John it's really starting to crystallize into something really good, like it was really well orchestrated so the students could really follow from, like, the origins of neural networks and how they vanish and they came back and how that then suddenly became, you know, the origin of HAI, and then we could do a lot of very practical examples where the students could really work with voice and images and so on, and that turned out really really well.
• 5:17
  Speaker 2: And, yeah, so a minute ago actually, because it was so good that the teachers were just keep on chatting on WhatsApp Somebody said like, oh, you have to come to Italy in July and I was like, yeah, I don't know, I might.
• 5:30
  Speaker 2: You never know, I have an office there?
• 5:33
  Speaker 1: That's true?
• 5:33
  Speaker 1: Yeah, no, and I think that's the so.
• 5:40
  Speaker 1: For those that don't know, there was a.
• 5:42
  Speaker 1: There was a program in Malawi in East Africa recently where there was kind of a whole symposium of learning there for educators and students.
• 5:47
  Speaker 1: And, and you know, one of the key things I think about the foundation and I think that kind of snaps into sort of the arduino philosophy is kind of empowerment through knowledge and just getting educated, and sometimes you're educated as a student I think we all were but sometimes you're also educated as a professional.
• 6:04
  Speaker 1: You're always learning new things and I guess with tiny ml and edge ai, I mean there's.
• 6:09
  Speaker 1: I was going to ask you like how do you think things have changed in the past few years?
• 6:13
  Speaker 1: I mean, arduino has been around for 20 years so but the Edge AI, tinyml stuff maybe in the past five years has become a little more prevalent.
• 6:20
  Speaker 1: How have things changed for Arduino as AI has become more of a thing, yeah, I mean.
• 6:27
  Speaker 2: Well, as you know, first of all, we started work with microcontrollers.
• 6:30
  Speaker 2: I mean, the only thing we were looking for was an easy way for people to access the technology.
• 6:37
  Speaker 2: It felt almost ridiculous that it was so expensive and that it was so complicated to just do something that would detect pressing a button and turning on a light.
• 6:46
  Speaker 2: Sure, yeah, if you think about it, one of the biggest successes in the IoT business is the interactive digital lamps you know, so is the one thing that says the most.
• 7:00
  Speaker 2: You know, like the biggest IoT device is connected lamps.
• 7:03
  Speaker 2: So from that perspective, why should it be hard to create that?
• 7:09
  Speaker 2: That was one of our goals and we went from there to you know, now we have full Linux computers with eight cores and they can be updated over the air, right.
• 7:18
  Speaker 2: So with that increase of functionality we can also do a lot of the from the small AI kind of things from the tiny, really tiny ML, like small systems for like a couple of tens of kilobytes that can do small inference to check temperature or respond to gestures or whatever to actually systems where you can load your AI with a Docker container over the air.
• 7:46
  Speaker 2: So, yeah, we have grown a lot at Arduino and definitely being part of the growth of the AGI Foundation has helped us to also shape our development, because when we started it was my partner, massimo, started to collaborate with BJ and Pete Worden and trying to figure out, oh, do we have a processor that can actually compute the smallest AI you can have?
• 8:10
  Speaker 2: And oh, it turns out we have.
• 8:12
  Speaker 2: Okay, let's try to do something with this.
• 8:13
  Speaker 2: Became the first book on O'Reilly on digital and ML and it became kind of like de facto, our words became the de facto words for teaching computer vision with AI and so on.
• 8:24
  Speaker 2: So yeah, it definitely is.
• 8:26
  Speaker 2: This journey helped Arduino become what it is now.
• 8:29
  Speaker 2: You know, it's this interaction with the foundation has helped us really do these kind of things.
• 8:34
  Speaker 1: Yeah, no, that's great and I think, yeah, like you said, ai on the edge has come a long way, even in the past five years.
• 8:40
  Speaker 1: I mean, have you seen any inklings yet of people trying to do generative AI on some of your platforms?
• 8:49
  Speaker 1: I mean, you started with sort of the 8-bit Atmel, so that's not gonna work.
• 8:54
  Speaker 1: Where is this going?
• 8:58
  Speaker 2: Yeah, well, I can imagine.
• 8:59
  Speaker 2: Well, right now our most powerful computer is a headless computer that is not really meant to use in home environments or, like more, it runs in Yocto Linux that was really meant to be embedded in a machine somewhere, maybe operating a machine that flies, for example.
• 9:13
  Speaker 2: So in those kind of situations generative AI doesn't, as of today, doesn't make any sense.
• 9:19
  Speaker 2: I think we will see generative AI working in other ways in the future that will go beyond just creativity kind of tasks, more like looking at solving complex problems when they're happening.
• 9:31
  Speaker 2: Like looking at solving complex problems when they're happening.
• 9:36
  Speaker 2: So, but as of now, we haven't seen.
• 9:36
  Speaker 2: I can't really report because of our portfolio allows us to do what it allows us to do, right, right, like, like.
• 9:42
  Speaker 2: On the other hand, I have to say like just yesterday, actually, one of our, one of our designers in Italy, he connected Claude to Arduino CLI as a demo and he shared with our internal Slack and it's beautiful because he has managed to like by giving access to the command line in his computer.
• 10:01
  Speaker 2: He would say like oh, make me a program where I have a potentiometer connected to whatever and this thing on this other pin on the system would automatically create it, upload the code and the world would just work.
• 10:11
  Speaker 2: Right, so I see multiple ways where AI and machine learning and tiny machine learning can interact with the Arduino ecosystem in multiple ways.
• 10:29
  Speaker 2: Right, right, so, as you might know, I'm also teaching at malmo university, um, here in sweden, and um I'm the head of the masters interaction design, and we actually teach our students how to train their own llms, and the first thing they do because you want to teach them arduino the first thing they do is to have the llm program for them so they teach them with, like, whatever code they can find.
• 10:52
  Speaker 2: So I think it's opening up, for right now it's very wild, but I think we're gonna see more and more not just useful, but also creative uses of things that we haven't imagined, and I think that's one of the features of technology.
• 11:07
  Speaker 2: We put a platform that enables our people to do something we never thought about.
• 11:12
  Speaker 2: We we started to see some of those results.
• 11:15
  Speaker 1: Yeah, no, I think the I mean, you could argue one of the biggest impacts of generative AI right now on technology is just in the developer space, you know.
• 11:22
  Speaker 1: So the ability, you know they call it vibe coding now and like the ability to use, you know, cursorai and some of these startups that are really revolutionizing the way code is being developed and I guess, in these embedded systems and other things, using generative AI and training language models to help, you know, accelerate that development is probably a huge, huge opportunity, especially, I think, for people that are learning, learning the space to get someone to sort of help figure that out.
• 11:50
  Speaker 1: So that is pretty cool.
• 11:52
  Speaker 1: And the other thing I think that you mentioned was interesting and people don't always realize is sometimes the simplest implementations are the most effective.
• 11:59
  Speaker 1: So you mentioned about smart lamps.
• 12:02
  Speaker 1: I was at a conference for physical AI recently and I reminded everyone, the most successful robot in history is the Roomba, the vacuum-cleaning robot, and so you don't necessarily need to build a full humanoid that does backflips, you know, to be effective.
• 12:17
  Speaker 1: And so Arduino is kind of an interesting platform in that it provides, you know, that real kind of foundational, you know, implementations that can do sometimes simpler things but much more effectively, rather than doing something much more complex that could cost a lot of money and and not be as effective.
• 12:34
  Speaker 1: So so, yeah, sometimes you know small is the new big when it comes to some of this stuff, right?
• 12:41
  Speaker 2: Yeah, that's the funny we saw in Arduino, especially when we look at our potential and existing industrial customers.
• 12:50
  Speaker 2: You know they're looking at incremental changes in their processes.
• 12:54
  Speaker 2: Because I can probably kind of really embrace like what AI can really bring all at once, because I will require changing so many processes but maybe like automating something here or there, and for those kinds of things actually Dynamo is great.
• 13:07
  Speaker 2: For example, I mean, imagine a power drill where you, for example, sense how the current is administered to the motor and that allow you to know whether you chose the right bit for the hole you have to drill.
• 13:22
  Speaker 2: You know, so it's very very simple things where you can apply very simple sensing to produce a great impact for a couple of cents, because that's really the thing.
• 13:33
  Speaker 2: Like if AI is gonna make your drill be costing like 10% more, people might not buy it.
• 13:40
  Speaker 2: We need to figure out how to do these incremental changes, and those small, small changes have a huge effect, a lot more than we can anticipate.
• 13:49
  Speaker 2: So I think there's a big opportunity there, and actually what our CEO, fabio violante, has been doing is to to go meet a lot of these companies who are, uh, looking at these like small changes and and discuss with them okay, why don't you try this thing while you try that thing, what can we do for you?
• 14:12
  Speaker 2: And so on.
• 14:12
  Speaker 2: Obviously we want our boards, right?
• 14:15
  Speaker 1: yeah, I mean, I think one of the big areas maybe people don't realize too is like in terms of there's brownfield and there's greenfield implementations, and so greenfield usually means like ripping everything out and starting over again, or building something brand new, which is very rare because it's very expensive.
• 14:30
  Speaker 1: And you know, many times, especially in the industrial manufacturing space, all the CapEx has been depreciated and people have working factories and working processes, like you said, and so those are what we would call brownfield deployments, where you're adding capabilities to an existing deployment and maybe adding AI capabilities or telemetry in the field.
• 14:51
  Speaker 1: And that's kind of like the vast majority of business out there these days is you know, how do you add to something that's already a sunk cost and a working process?
• 15:01
  Speaker 1: Add more efficiency, more intelligence to it, and then you know, at some point, that the equipment will be replaced and then then there'll be some sort of upgrade but, yeah, a lot of people don't realize the brownfield is kind of where, where the action is.
• 15:16
  Speaker 2: I think, I think it's a bread and butter today, because, let's say, you have like a production chain for example, you use plc's to control the speed and decide whether something has to be replaced, and you are using an old computer with some old computer vision.
• 15:28
  Speaker 2: Now you can put everything in a camera this size and I need to trigger an on off like the current.
• 15:33
  Speaker 2: The current thing has to be removed from the lamp because it's not okay, you know, yeah, so, and that goes a fraction of the price as well.
• 15:41
  Speaker 2: That's crazy, but right now, when these cameras cost a hundred dollars and they can do hundreds of frames per second a lot more than traditional systems based on other cameras with computers dedicated.
• 15:50
  Speaker 2: So so that's that's the interesting thing.
• 15:53
  Speaker 2: And and um, that also minimizes or let's say modularizes the production line a lot more, in a sense and in a way that makes it very easy to add new things to it and make it very operative also from a replacement perspective, and it allows you to also have an ecosystem where you can think about parts from different vendors, so you're not locked into a vendor, because a lot of times your sensor if it talks over a modbus or canvas or whatever other thing.
• 16:23
  Speaker 2: I just sense you're like, oh, I detected the mistake.
• 16:25
  Speaker 2: And then your system just knows oh, when I get a signal like this, I have to push the robot arm, I have to move the next thing.
• 16:32
  Speaker 2: You know, I think it's okay, I'm gonna make a wild.
• 16:37
  Speaker 2: I'm gonna make a wild statement.
• 16:39
  Speaker 2: This is kind of like when ibm standardized the design of computers you know, like, in a sense, now we can have systems that are smart at sensor level and they can trigger things, uh, through standardized communication, uh lines like mode buses, I square2Cs and whatever, so that they don't need to actually know.
• 16:59
  Speaker 2: The intelligence doesn't need to know what is at the other side, it needs to know.
• 17:03
  Speaker 2: This is a signal that triggers me.
• 17:05
  Speaker 2: Period, right, right.
• 17:07
  Speaker 1: Yeah, no, exactly I think.
• 17:09
  Speaker 1: Yeah, you know, sort of having the intelligence at the sensor level enables the kind of the metadata to be communicated over standard protocols, which can be super efficient.
• 17:18
  Speaker 1: Like you said, it also enables, you know, a lot more flexibility of vendors and partners to put equipment together and stuff like that we had talked about.
• 17:28
  Speaker 1: So we recently started talking about Austin.
• 17:30
  Speaker 1: You know, Arduino was founded in Italy.
• 17:33
  Speaker 1: You were just in africa, so arduino is really a worldwide uh company these days.
• 17:38
  Speaker 1: I mean, are you what?
• 17:40
  Speaker 1: What kind of uh?
• 17:40
  Speaker 1: You know geographies?
• 17:43
  Speaker 1: Are you seeing the most activity these days for, for, for your kind of stuff?
• 17:48
  Speaker 2: is there?
• 17:48
  Speaker 2: Is there one or is it just like?
• 17:49
  Speaker 2: happening everywhere well, yeah, okay, I think it depends on the product line.
• 17:55
  Speaker 2: As you can imagine, the US plays a very important role for us, has always played an important role for us.
• 18:03
  Speaker 2: The US has a different approach to, let's say, experimenting with technology.
• 18:09
  Speaker 2: A company could say, oh, let's just invest half a million dollars and replace all of our plc's.
• 18:14
  Speaker 2: You know, you, you see that very seldom at other geographies.
• 18:19
  Speaker 2: People plan differently, and so on, um, so, so that's interesting.
• 18:25
  Speaker 2: Um, but then at the same time, as you know, europe is 50 something percent, uh, small and medium-sized companies.
• 18:32
  Speaker 2: So it's like, uh, so, uh, we've seen a of our industrial products.
• 18:36
  Speaker 2: We decided to be used in this context.
• 18:39
  Speaker 2: But then, if you move to, for example, countries like India, in India we have open manufacturing in India, which is the first time we manufacture as a sort of Italy, because there was a huge demand of our educational boards, I see, so yeah, basically that allows us to cut down the cost because we, otherwise we are obviously we are not competitive if we, if we manufacture in Italy and then have to export there, and so on and so forth.
• 19:03
  Speaker 2: So so for us it's very interesting the fact that I always make say this as a joke.
• 19:09
  Speaker 2: But Arduino is a software company and we have 60 hardware products, right, but it's because we can basically plug our systems into any existing factory.
• 19:18
  Speaker 2: I think we are the best example of a retrofitting system.
• 19:22
  Speaker 2: Like, we have our testing machines, we have our designs.
• 19:25
  Speaker 2: We come to a factory and put on the table this is what we want to do.
• 19:28
  Speaker 2: And at the end we put our machines and say, well, the next thing you need to do when it comes out, you have to put it in here and make some of our automatic testing.
• 19:35
  Speaker 2: And we that's why we have a hundred percent yield in our productions, because every single voice says it one by one, which is not very common for a medium-sized operation.
• 19:43
  Speaker 2: That card, we know you know right, like, uh, making sure that 100% of the words, when they reach the end user, are functional, so that the experience of programming them and putting them into work is very relieved.
• 19:58
  Speaker 2: So, yes, that's our approach.
• 20:02
  Speaker 2: And again, ai also places we grow there.
• 20:05
  Speaker 2: You know testing processes and so on.
• 20:09
  Speaker 2: So yeah, it's.
• 20:11
  Speaker 1: Yeah, sounds like there's, you guys.
• 20:13
  Speaker 1: Yeah, a lot of activity, and I'm sure you're racking up a lot of miles too, you know, going from Africa to Austin, to Sweden and Italy and a lot of other stuff too.
• 20:24
  Speaker 2: But it's a worldwide phenomenon.
• 20:27
  Speaker 2: Like in the foundation, we have so many different companies from different parts of the geography, different parts of the world, so uh, yeah, yeah, and I want to see actually some of this, what we are doing with, like, um africa and someone, exactly thanks to the foundation and to ictp, uh, on the field, who is very much pushing for it, uh, with the help of of UNESCO and some other entities.
• 20:51
  Speaker 2: So I think it's important to understand how the dissemination of knowledge around technology needs to have all of these different actors collaborating, because I don't think we can put everything on a single person or a single entity.
• 21:04
  Speaker 2: It's very hard.
• 21:05
  Speaker 2: So, unless we all collaborate, this technology will not reach out to certain places, and I cannot state how important I think it is for reaching out.
• 21:15
  Speaker 2: And again, I'm still in shock on how good the guys were in Malawi and the workshop we ran.
• 21:22
  Speaker 2: I know, seriously, incredible, incredible professionality and level.
• 21:30
  Speaker 2: I understand that also, people who applied were a bit more senior like people already had their bachelor's, for example than we would be teaching at other places, but at the same time, I mean I think it's a combination of the level of interest that this is awakening and happening at other places, at different places, yeah, so, yeah, I think, I think it's an effort we should continue to pursue.
• 21:56
  Speaker 1: You know, yeah, yeah definitely we're excited about it and you know, agif foundation has a scholarship fund and that's you know.
• 22:04
  Speaker 1: We use the scholarship fund for activities like that, as well as fellowships and travel grants and things like that.
• 22:09
  Speaker 1: So we're going to continue to fund that, because what it's doing is it's educating the next generation of leadership and also empowering folks from all these geographies to understand how to use this stuff locally, which is pretty cool and kind of democratizing it a little bit.
• 22:25
  Speaker 1: So that's good, good stuff.
• 22:28
  Speaker 1: Well, david, it's been great catching up again.
• 22:30
  Speaker 1: I'm sure I'll see you in person in the next few months somewhere, who knows, maybe an airport somewhere.
• 22:36
  Speaker 1: But yeah, thanks, yes, thanks again.
• 22:42
  Speaker 2: Yeah, thank you so much.
• 22:46
  Speaker 1: All right, take care, david Thanks.
• 22:47
  Speaker 2: Thanks for having us Take care Bye.