Pioneering open source drones and robocars

Chris, we have lots of stuff that we could talk with you about. Of course, your history at WIRED, all the stuff you’re doing with drones… We wanna focus in on the drones. I read in your Amazon bio that you said you have a “hobby gone wrong.”

I have an Amazon bio?

You do, on your author page.

I had no idea.

Yeah, it’s a little bit outdated because it still calls you the editor of WIRED… But it does say you have a hobby gone wrong with aero-robotics and with 3D robotics, a company you founded based on a hobby gone wrong… So I’m curious what went wrong, if you could tell us that story.

Yeah, so I’ve described drones or 3DR as a hobby gone wrong, and what I meant by that is that the hobby was super-fun, it took off, it became kind of a movement, and it became a company, at which point it stopped being a hobby. So if you’re doing something for fun and it succeeds wildly, typically it’s not fun anymore. You’re spending all your time dealing with customer support, community management then as a company… And I don’t think I have flown a drone in the last year.

Bummer.

Oh, no. A sad ending to your amazing story. Well, tell us about the beginning of this, back when it was fun. Go back to the fun days. [unintelligible 00:02:55.10]

By the way, I still have fun doing other things, and it is a perfectly natural progression, and we’ll get into this probably later in the conversation… But because I miss the days of it being a hobby and fun, I decided to restart the whole thing with cars, not drones. So DIY Robocars is our autonomous car racing league. It’s super-fun, and remind me not to turn it into a company.

I was gonna say, when are you gonna transition that into a money-making organism?

I think that you and my wife are assigned to the task of answering that as never. [laughs] Yeah, so you were saying to talk about the transition from –

Yeah, go back to the fun days; tell us how this thing got started. I know a little bit of the story, because I got to hear some of this…

Can you share a little bit of the back-story, Jerod, like what interests in Chris? The open source summit/conference, maybe a little bit of back-story?

[03:57] Yeah, so we were at the Open Core Summit and Chris had a talk about drones and robotics. I think it was called “How I convinced the Pentagon to use open source drones.” Who wouldn’t be interested in that, right? So I was all-in on that talk, and got to have about an hour of Chris’ time just telling us this amazing story, and I wanted more people to hear the story, so I emailed him and said “Come on and tell us.” I think it started around the time of the iPhone, if I recall.

It did. It’s 2007. At the time I’m the editor of WIRED, I’ve got five kids, and my wife and I are scientists by training, and we’re always trying to get the kids interested in science and technology. We’d bring home projects, and they would invariably be disappointed and tell us that we’re nerds… Which is true. And this particular weekend, early 2007, two things came into the office at WIRED for review. One of them was a LEGO Mindstorms NXT kit, which is a robotics kit, and another was a radio-controlled plane. And we had a deal at WIRED, which is you were allowed to take a thing home on the weekend, as long as you agreed to review it. Typically, first come, first served, I grab these two, and I said “Okay kids, on Saturday we’re gonna build a robot, and on Sunday we’re gonna fly a plane.” And they’re like, “Okay… We’ll see.”

So on Saturday we diligently followed the instructions… And you know, when you build a LEGO robot, it’s got wheels, and sensors, and it takes all morning to put together and to program it, so it very slowly moves forward until it sees a wall, and then it sort of moves back. And they’re like “That’s stupid. We’ve seen Transformers… Where are the freakin’ lasers?”

“Boring…”

“Where are the lasers?!”

Yeah, exactly. And I was like “Okay, fine. I get it, little doll.” By the way, Hollywood has ruined robotics for kids, because you just can’t compete with CG.

Oh, my… Good point.

So then I’m like “Okay, tomorrow (Sunday) we’re gonna fly a plane.” So we go on YouTube and we see the videos of acrobatics, and cool stuff. They go to the field, to the park with me, I launch the plane, and it immediately flies into a tree. And they’re like “That also sucked.”

[laughs]

And I kind of had to agree with them on both counts. So I come back, and I had to bribe them with ice cream, and I come back and I thought “How could that have gone better?” Clearly, we needed a cooler robot and a better flying plane. And I thought “Well, what if the robot was the plane? What if it would just fly itself? That would be cooler.” So I googled “flying robot.” If you google “flying robot”, it turns out that you get drone. That’s what a drone is, it’s a flying robot. I’m like “Got it. Huh. What’s a drone?” If you google “drone”, it’s like “it’s an aircraft with an autopilot.” Like a plane with a brain. I’m like “Yeah, okay. What’s an autopilot?” If you google “autopilot”, it’s like “sensors, and compute, and software…” I’m like, “That’s kind of what’s in the LEGO Mindstorms box.” It came with gyroscopic sensors, accelerometers, magnetometers, a Bluetooth connection which would connect to a GPS… I’m like “I think it’s kind of the same stuff that’s in this box.” So around the dining table we stuck together the bits, and had a nine-year-old program the code… And we had a LEGO autopilot. And then we stuck it in the plane, now retrieved from the tree.

[laughs]

The next weekend we went out, and it worked. Not brilliantly well, but it flew…

Sure. Proof of concept.

…and it flew by itself. Yeah, exactly.

Any lasers this time?

No lasers. Still no lasers.

We’ll get to the lasers…

[07:43] They were interested for like ten more seconds, and that was it for them. And I was like “What just happened…?” That should not be possible. Because you know, at the time drones were like predators, and Global Hawks, and military-industrial, and they were classified, and super– ten billion dollars things… And we had just built a drone with LEGO pieces, around a dining room table, programmed by a nine-year-old. Putting aside that it was not a great drone, that should not be possible.

When a nine-year-old can do something that is classified, that literally export controlled is a munition with toy pieces, something important in this world has changed. And I wasn’t clear what it was, but I knew it was something. So I set up a website. Ten years earlier it would have been a blog, and ten years later it would have been a Twitter feed, but at the moment it was a social network. So I set up a social network I called DIY Drones, and it was largely so I could chronicle my own experiments, but also to ask dumb questions in public.

Learn, yeah.

Learn, exactly. And it just so happened that my discovery at that moment had kind of coincided with a lot of other people who were discovering the same thing, which is that this impossible thing could suddenly become easy. And some people had come via “advances in radio control airplanes”, with electric power and better radio, some of them had come via sort of the maker movement, with like 3D printing and Arduino… Some had started playing around with the components in the iPhone - which had come out that year - that also could have been found in like a Wii controller, like the accelerometer.

Right.

And everybody sort of had theirs. It was a glitch in the matrix. There was this disturbance in the force, that had gone out in 2007. And anybody who was paying attention to hardware sort of recognized it, but everybody saw it slightly differently. For example, the Fitbit founders had bought a Wii (video game console) and they were looking at the controller and they were like “What’s in here?” and it’s like “Oh, it’s an accelerometer chip. I wonder what else I could do with that”, and they made the Fitbit.

So everyone was recognizing there was this bounty of cool stuff that in retrospect we now call the peace dividend to the smartphone wars. Basically, the components of smartphones and the economies of scale of the Apples and the Googles of the world, with memsensors, Wi-Fi, battery technology, GPS technology, radio technology - all these bits had transformative effect in adjacent industries, and it was up to us to figure out how to take these bits and use them elsewhere. And my elsewhere happened to be drones. But you could see this explosion in hardware innovation around there, as everyone figured out some adjacent space to explore.

So yours was drones, and hobby at this phase. Tell us the next step in your process. You have a website, DIY Drones; I think at this point it’s growing maybe faster than you expected because, like you said, it was kind of a moment in time…

Yeah.

Was that a strong indicator for you, like “Holy cow, there’s something here, and I need to keep going”, or were you just continuing to tinker?

I was continuing to tinker. The very first three autopilots were ones that I had programmed… Which, you know, I’m not a great developer, so that tells you what level they were at. But two things happened that sort of tipped me off this is bigger. Number one is that people started contributing their own, better code, and electronics designs. So that was good. And then teams started to form; people started to work with each other. The way open source always works in my experience is that you do something to scratch your own itch, and you put it out there, and it might inspire other people to do their own thing. And at some point, they start to clump, and people say “You know, rather than doing my own thing, I think I’m gonna join up on your thing, or I’m gonna do a pull request, or a bug fix, or an issue report.” The moment they start working together, then you’ve got something.

So what we saw was that phase - I’m putting out stuff; our LEGO autopilot back in the day made the frontpage of Slashdot. So that was a thing back then.

[12:17] For sure.

And then everyone’s like “Oh, that’s super-cool, but your code sucks”, which is true; it was written by a nine-year-old. And by the way, that LEGO drone is in the LEGO museum in Billund, Denmark right now, as the world’s first–

What?! That’s awesome.

Yeah.

So did LEGO receive it? Did they care about it?

Oh, they loved it. I mean, I was the editor at WIRED, so it was kind of cool marketing for them…

That helps, yeah.

But yeah, so… I was also on the advisory board for LEGO Mindstorms later. So it all played well. By the way, a cool little bit about this - at the time (and actually still) autopilots were considered cruised missile controllers, and a weapon or a munition to use the phrase, so they were export-controlled. And if you export the technology for a cruise missile controller, you can go to Guantanamo Bay. It’s super-bad.

That’s bad.

And the act of putting something on the internet is considered exporting it. So technically, we weaponized LEGO.

You were on shaky ground there. [laughter]

And I was just like “Please, please, please subpoena my children, so that they have to testify in Congress about how they put a LEGO autopilot on the internet and committed some sort of grave crime.”

It’d be the best press of all time, wouldn’t it?

You really can’t do better than that. They sadly - wiser minds - realized that the problem was not us, it was the law.

Yeah, exactly.

When did the law change around that? At what point did they get wiser to it?

The law never changed. There are two things that the law accommodates. One was that there was an exemption for public domain. So if you open source this stuff, then it’s considered public domain and you’re exempted… On the assumption that no one would ever open source a cruise missile controller, but of course, that’s kind of what we did.

Apparently that’s a bad assumption.

Yeah, it was a bad assumption. And then the second exemption is that over time the definition of military-grade technology keeps evolving, as technology evolves. Remember, back in the day a Playstation 2 was considered export-controlled, because you could be designing nuclear weapons on it, or something like that.

So over time, as things become commercially available and sort of off-the-shelf, they fall off the list of munitions. Now, it took – I should tell you that it took to 2013 before the autopilot my children made fell off the list… Or that quality of autopilot fell off the list. It was only once you could literally buy it in Walmart that it was considered no longer eligible. It’s a constant struggle to educate the regulators that “Hey, you know what - GPS is like in your phone”, and they’re like “Huh. Okay. But what kind of GPS?” Is it GPS as that standalone module? Is that export-controlled? Anyways, it’s a long struggle, and I’m glad I’m not doing that anymore, but yeah, that’s how we got around it.

Did you actually have lots of meetings with them, where you have to go in and explain these things? Because there is this – like, you have the commoditization of all these different parts, like you said. There’s this ragtag team of open source people putting out into the world the plans for building defense mechanisms… Meanwhile, you have this huge industry of million-dollar – million is a small number; multi-million-dollar service contracts, and all these things, and it’s gonna quickly go by the wayside.

Yeah… So there’s logic; what you’ve just said is very logical.

Well, thank you.

[15:54] The way it actually works is that you have inspectors at the ground level, and their job is just to implement the laws as they understand them. So did we have lots of meetings? Yes. Did the FBI come to the office? Many times. And everybody was – I was gonna say everyone was super-nice; that isn’t actually true. The FBI was super-nice.

Most people. [laughter] Who wasn’t very nice?

The Export Control Administration is handled by the Department of Commerce, and they have these kind of regional inspectors, and the regional inspectors have quotas of like how many people do they put behind bars today… And they just didn’t wanna hear the logic. They were just like “It looks to me like you’re violating part whatever. Prove to me that you’re not.” And it’s just years of lawyers just to – literally, you just have to show them that this could be bought in Walmart.

Right.

And then you need to prove that it’s the same as the one you’re doing right now… Unfortunately, as a pioneer, we were the ones who had to go through it first.

So while you’re pioneering this and you’re taking lots of meetings with scary people, sometimes nice, sometimes not so nice people, you’re also at the same time this open source forum, and this open source code… You’re starting to build a hardware company, 3D Robotics. Tell the story of how that actually turned into a business. I know you have a co-founder, which is a very interesting tale… Unpack that for us.

Yeah, so we started with the community; the community starts taking off. People are sharing code and design files, and it’s pretty exciting. It’s clear we’re doing something and it’s accelerating really fast. It’s still very much kind of a hacker thing, but the word is getting out that this is now possible. And then the next generation of people come, and they’re like “Hey, we heard that you could put the letters DIY in front of a drone. I would totally like to have my own drone, but I don’t know how to program, I don’t know how to sodder, I don’t know how to compile, whatever. Can you just do it for me? Can we buy one? A kit, or something like that.” So we said “Well, I guess we need to turn it into a kit”, and we started really simple, with an autonomous blimp called Blimpduino, because it used the Arduino.

Nice.

My kids assembled it on the dining room table, in pizza boxes. We made 1,000 of them for Maker Faire. They sold out on Saturday. I came home and I said to the kids “Great news! Everyone loves our blimp kit. They sold out”, and they’re like “Whatever.”

[unintelligible 00:19:25.22]

Your kids are just completely nonplussed by this entire thing.

These are Silicon Valley kids. It’s impossible to impress them.

[laughs]

Some day, some day they’ll thank me. And I said “Well, we need to make more”, and they’re like “Not gonna happen. We’re not doing that again.”

“We’re out.”

[19:47] “We’re out. We’re out.” So I went onto the forums and I was like “I need someone to help put together these kits.” The smartest guy in the forums was this guy who was flying a remote-controlled helicopter with a Wii controller… He posted the code, and did YouTube videos; he was really good. His name is Jordi Muñoz. Never met him, just a guy on the forum… And he said that he had a little spare time and that he would be happy to help me.

I said “What do you need?” and he said “I need $500 for parts”, so I sent him a check, literally, a check in an envelope of $500, and he sent me back a picture of him in the garage, soddering together the parts. And I thought that was the end of it. They continued to be made, we set up a little e-commerce shop, and that was it. But he kept sending me more pictures, and just like “I’ve got some friends to help me. We’ve moved to a garage, we’ve moved to a bigger garage, we bought some pick-and-place machines, some low ovens…”

He’s scaling operations.

He is.

I still never met the guy at this point… And by 2012, Jordi had - with a full investment of $500 - we incorporated his company, which again, I did on the internet, and never met my co-founder, no capital… By 2012 we had factories in San Diego, Tijuana Mexico, we’re the biggest drone manufacturer in America by a factor of ten, making more drones per month than all of America’s aerospace companies combined. We’re on a ten-million-dollar run raid and I still haven’t met my co-founder.

Wow.

Maybe it was 2011, but it was early. I’m still the editor of WIRED. It’s a hobby gone wrong. So the venture capitalists, who have been saying “This is kind of a thing. You should do this.”

Do you think?

Yeah… I talked to my wife about it, and I said, you know, media is kind of getting a tougher industry… It’s still doing really well, but you could see that media was gonna get tougher. I’d been doing it for 12 years, running WIRED… And meanwhile this sort of side project had sprouted, and I said “I think I might wanna just raise capital and do it.” And she says “Well, you’ve gotta raise a lot of capital.” I’m like “How much?” She said “Well, a million dollars for every child.” We had five children, so I said “Okay, I’ll raise five million dollars”, which seemed like a lot at the time.

So we raised five million dollars, she let me quit my job, I met my co-founder… I actually met this during the fundraising, and it turns out that Jordi, when I actually started this – when he started this he was 19 years old. He’d just graduated from high school in Tijuana, Mexico. He was actually in the United States because he was having a child, and they wanted done in the U.S, and working on citizenship stuff… That’s why he had spare time. And I’d accidentally created a 21st-century aerospace company with a teenage high school graduate from Tijuana, who I met on the internet.

I love it.

Which is like perfect.

That almost never happens. Just once, maybe.

It doesn’t sound right, but… You know, as I say, if you wanna start Facebook, get a kid from a Harvard dorm room. Do you wanna start a manufacturing company? Get a kid from Tijuana.

Right. That’s competitive advantage, right?

That’s the Shenzhen of North America. I didn’t know what a pick-and-place machine was, and he just bought them used on eBay and downloaded the manual from the internet.

Wow.

If you grow up in Tijuana, electronics factories is just something people do.

Explain pick-and-place, just quickly.

Yeah, so an autopilot is basically a circuit board with lots of chips on it. And the circuit board is like a fiberglass board, you just have those made… And the pick-and-place is a robot that takes the little chips and puts them on the board, very precisely on top of a goopy grey paste, which is actually sodder. And then it goes into what’s called a reflow oven, and then that kind of melts the sodder and it all becomes electrically connected. There’s other elements of it, but it’s basically a robot that makes electronics.

[24:13] Wow.

Adam and I also met on the internet, and he had the Changelog going at the time. We’re business partners, but it took us years to build up trust, didn’t it Adam?

Yeah.

Anybody you meet, you have to build up trust, but especially when you meet them online. It seemed like you and Jordi - you didn’t mind it that much maybe because it was a hobby.

Yeah, trust has never been an issue for me, just because I’ve been lucky. I wasn’t putting much at risk. It was like literally $500. $500 and I expected nothing. I literally expected never to see it again.

Did it literally not take more than $500 to scale it to the ten million you mentioned? $500 in to invest…

They were cashflow-positive on day one.

Wow.

After we took venture capital, we raised like 140 million dollars.

Right. So eventually it grew up, but it was–

Yeah, eventually it grew up. But the first four years of the company… No, that’s not right; 2009 through – yeah, about three years of the company it was all organic cashflow.

Who’s buying the drones? DIYers, or was it…?!

Well, they weren’t even drones at that point.

Kits.

They were actually kits, yeah. They were DIYers… The whole drone community ended up being sort of two communities that came together, but never really culturally. Some of them were just like, you know, geeks who were fascinated by new technology, and robots, and software, that kind of stuff. That’s where I came from. No interest in the flying part, just kind of like “This is the hottest new thing. I’ve gotta understand it.”

The other community were like radio control hobbyists, who wanted to extend their hobby with autonomy, and going first-person view, and things like that. They were really into the flying thing. So I actually have no interest in the flying thing, which is why I don’t fly drones. I was really interested in the software and the hardware, and the data that drones could acquire.

So basically you have pilots and programmers. And the pilots - we’d go out in the field and they really liked the air time, and the programmers sometimes never even flew. So we had these two communities come together.

Today you still see the schism in the drone world between pilots and programmers. Programmers are about autonomy. These are robots, they should not have humans involved. No sticks.

And pilots wanna fly them.

And pilots wanna fly them. So basically we banned – not banned, but I have a policy “No sticks.” If there are sticks – sticks are the remote control sticks. If there are sticks involved, it’s not a drone. And we actually unscrewed the sticks. We still use it for FAA-compliance purposes; there still needs to be an ability for a human to take over… But we actually unscrew the sticks, just so no one accidentally touches them.

So I believe that humans should not be flying. I think the Wright brothers skeptics were right, that humans were not meant to fly. They’re meant to be flown, but not to fly… So I’m 100% autonomy. And the pilots still really like the human-centrism of the whole thing.

Today in the industry the pilots are called operators, and the programmers are called developers… And they’re still kind of in opposition to each other. Over time, as long as the FAA requires the so-called pilot in command, the operators still have to be there, but there’s nothing left for them to do. They’re literally just standing there as kind of a statue of compliance. But we don’t want them touching the sticks. It’s like if you fly a jetliner across the country - your pilot is not doing anything. You know that, right?

For sure.

[28:03] They are just in the cockpit; two of them are in the cockpit, just in case.

Just in case. There’s a social factor there as well, right?

Yeah, yeah. I call it flight theater. They wear uniform, they’ve got a hat, they talk to you over the – it’s all theater.

Somebody has to make the announcement, how many feet we’re flying at… “Everything’s gonna be just fine, folks. Everything’s gonna be just fine.”

Exactly. Now, they’ll still do the take-off and landing, because pilots [unintelligible 00:28:30.24] You actually can tell the difference between an autonomous landing and a manual landing.

How so?

The smooth ones that you applaud - the pilot had nothing to do with that. That was all the autopilot. The bumpy ones - that was the pilot.

Wow.

Okay…

That’s not fair, by the way…

[laughs]

But these days a new Airbus can take off and land, and even taxi by itself.

Slightly upstream, but are you bullish on the autonomous cars then, in terms of fourth-level self-driving? What are your thoughts on that in terms of autonomous cars?

Level four… Yeah, so my hobby on the side now is that we do autonomous car racing. These are sub-scale, so they’re one-tenth scale. Same technology, same software, LIDAR cameras, and other sensors…

Just smaller scale.

Just smaller scale, which means that we can do them indoors, that nobody gets hurt… They cost less than $400, and we can race, wheel-to-wheel; two cars on the track at the same time. It’s super-cool.

No sticks.

No sticks, literally. You get disqualified if you touch a stick. So we’re doing level four autonomy every day in our races. That said, they crash all the time. It is all about the crashes. It is freakin’ Demolition derby.

So what is it that makes the difference between a team then? Is it the ability to program, the ability to fine-tune the different components to –

All of the above. We basically don’t want it to be about the cars. A lot of people are like “I’m gonna have a bigger motor, better tires…”, and I’m like “That’s kind of not the point.” We actually have four different classes of cars. There’s two distinct classes, small and large, but within those classes we have four different technological threads. One of them is computer vision, and that’s sort of the Tesla; that’s standing in for Tesla. It’s all a proxy war, so we’re basically doing the Tesla versus Waymo everything, in small scale.

It’s all a proxy war. Love it.

So the computer vision teams are standing in for Tesla, and then there’s the deep learning teams. It’s not a perfect analogy, since Tesla uses some deep learning as well… But you have computer vision teams, that are kind of classic computer vision, and then deep learning teams fall to three classes. There’s reinforcement learning, which is the Amazon approach, where you give it goals, and rewards, and cost functions, and things like that.

There is a behavioral cloning team, where you drive around manually a few times, and it sort of sees a correlation between what the camera sees and what your input was, and it learns from that… And then there’s a [unintelligible 00:31:07.04] supervision team, which is the NVIDIA team. And when I say NVIDIA team, I mean based on the NVIDIA technology, with Jetson Nano and all that. That team and that approach is that basically you show it 100-200 photos, and you click on where you would drive if you were seeing that, and it learns from that.

There’s also things involving LIDAR, and SLAM, and fisheye lens that are looking at the ceiling patterns, and lights, and all sorts of clever techniques. But it’s basically algorithmic approaches, versus other [unintelligible 00:31:43.14]

Is there any methodology that’s pulling away from the pack, that’s proving to be more reliable?

Some of the custom ones. You can actually see the data we posted. They just beat the fastest human, which is saying something.

[32:01] I would say that some of the custom approaches that are really optimized for this course were indoors, at a place called Circuit Launch in Oakland. By the way, there’s hundreds of these communities around the world now, and about 10,000 people… But the main one is in Oakland, California.

One clever approach recognizes that the lights on the ceiling of this place are kind of a distinct pattern, so it localizes – it’s got a fisheye lens and it can see the entire room from above. One of the problems is when you’re looking forward or around you on the ground level, it’s super crowded; there’s spectators on the side, other cars etc. If you look up on the ceiling, it’s clean, so it localizes itself based on that. That one’s really fast.

There’s another one that’s doing the same thing, but we have cones on the key corners, and it localizes itself by spotting the cones with LIDAR. Those are the ones that are super-optimized for this course, and they’re the fastest. They’re not generalized. Well, obviously, if you’re looking at the ceiling lights, it’s not gonna work outdoors, so those are not generalizable approaches, but… Anyway. There’s the answer.

Okay, so to summarize on the feasibility of level four in the real world, are you bullish on that?

Not in the near term, no. By the way, we’re dealing with the same thing with drones. They’re not fully autonomous, for a couple reasons. The main reason is regulation. They’re not allowed. The only way we get efficiencies is when you break the one-to-one human-robot ratio. You’re autopilot, and your 777 is not achieving anything in terms of efficiency because you still have two pilots sitting on it. Now, it may make their life easier, it may be safer…

Yeah, the payload is a human. If the payload is not a human, then you do have some efficiencies there, right?

Right, but in our case, although the payload is not a human, there still has to be a human present, standing on the ground, doing nothing. So we haven’t actually achieved any kind of labor efficiency. And the reason we haven’t is, again, the FAA regulations have limited us in that respect, because of the pilot in command concept.

We’re about to break through that, and there’s a principle called type certification. If an aircraft is tested and considered safe by the FAA, it’s allowed to do riskier things. No drone to date - I think possibly one military exception, but no commercial drone to date has been type-certified. And we’re gonna have the first one in about a month’s time.

This has to do with the recognition that drones are a low-risk kind of aircraft, and that as aircraft have moved from a mechanical era in the ’50s to essentially a software era, the process of regulation has not kept up. The 737 MAX crisis is a perfect example of an outdated regulatory system that basically locked the software down in 1967, when the 737–

Wow.

Yeah. All the software editions since then have been patches on top of the core. So it’s like COBOL in the government. And because it’s too hard to recertify – because they basically say “We’re gonna look at every line of code, and you can’t change the code or you have to recertify”, it was very hard to improve the code.

Yeah. It’s like legacy code by way of dictate. Like, you have to keep it.

Exactly. I think everyone recognized that back in the day, when code was a tiny fraction of the vehicle, that was okay. But now, as a majority of the vehicle’s technology, it’s not okay anymore. So rather than just sort of swap to this new approach where it’s like “Hey, we’re gonna treat a plane like a phone…” Verizon doesn’t look at every line of code in Android. They say “Look, you guys do your job, build Android the way you think you should, and we’re gonna test it on our networks and we’re gonna look at its performance. We have a battery of tests, and if it performs well, fine. Update the code all you want, just don’t change the performance.” The FAA is going to be doing the same thing with drones, and it’s called performance-based certification.

[36:07] What they’re saying is like “You tell us how it performs, we test to make sure it really does perform that way, and you can continue to improve and fix the code as long as it doesn’t change these key performance criteria.” Now, this is super-good. It’s a 90-day process, rather than a nine-year process. It’s built on our code process, which we – after we got into hardware, we built a ton of drones and then got out, because the Chinese were doing such a good job. We put the code into the Linux Foundation as something called Dronecode, and it is today sort of the Android of the space; it runs a code stack called PX4.

The way we developed that code, and the professionalism of that consortium, and our testing - a simulation process is now becoming the standard that the FAA is adopting. And that’s great. They want to do this experiment at the lowest risk category. These are vehicles that weigh about 2,5 kilograms, they’re flying under 400 feet, not carrying people, and we’re gonna build up millions of hours of evidence that this is the right way. And as the FAA gets more and more statistical power to prove that this method works, they can expand it to larger drones, drones flying over people, maybe air taxis, urban air mobility, and maybe a generation from now, the next jetliners will use this certification process that had been tested on drones first.

So you mentioned Dronecode; it’s out there, Linux Foundation. We’ll link it up in the show notes for everybody to go click through and check out all the different subprojects, and bits and bobs. The DOD itself also seems to be adopting some of – is Dronecode the platform that the DOD at least in some capacity is interested in adopting, or they have adopted?

Yeah, they have adopted. DOD is giant, and there’s many different parts of it. But there’s one – basically, at the small drone size they can go to Walmart and they can see what you can get from DJI. But they’re not allowed to use DJI, because it’s Chinese, and all that kind of stuff. So they want basically – and this is a case throughout the military; they want what’s called dual use technologies. They wanna get the – I was gonna say bang for buck, but maybe that’s too on the nose for DOD… [laughter] But they want the same price performance, as consumer technology, but they want it to be more secure. So rather than just buying Chinese drones and using them, they want to work with an open platform that can create drones as good and as cheap as the Chinese ones, but is trusted… And is trusted because at least parts of the DOD now recognize that open source is verifiable, that it performs well… Linux has correctly evangelized that open source is a good thing.

[40:09] Paved the way.

Yeah. And so they’ve realized that the way they’re gonna get price for performance is with an open platform… And not only because it can be trusted and verified and checked, and they know where it comes from, but also it avoids vendor lock-in. An open platform means that they can get many different companies using this standard… Because they hate vendor lock-in. Vendor lock-in – it’s called a Program of Record. And once you have Program of Record, it’s a monopoly. And by dictating open standards, they’re essentially dictating competition.

So they’ve adopted elements of the Dronecode stack… There’s a communications layer called MAVLink, there’s an open source ground station called QGround Control, and there’s the underlying code called PX4… And there’s an entity in the Silicon Valley area called the Defense Innovation Unit. They’ve been sort of tasked with stimulating this kind of open ecosystem for defense needs, and they’re the ones who have mandated this codebase.

So where does 3D Robotics fit into this new, matured world, where there’s an open platform? Are you just one vendor amongst many, or do you guys still have something that sets you apart?

Yeah, so we’ve sort of accidentally got into hardware, then purposely got into hardware really big. Back in those days, in 2013-2014, only consumer use of drones was allowed, so recreational use, but commercial use was not allowed, so we had to go consumer… And it was hard. You’re talking Best Buy, hundreds of–

Small margins.

Yeah, small margins, a lot of price competition, a lot of risk… And we made a couple hundred thousand of these things. The Chinese did it better than us. Their economics were better, the scale was better etc. They were really innovating brilliantly, and they just beat us. So we got out of consumer, and just as we were doing so, the FAA allowed commercial. It’s like, why don’t we make cell phones in America today? The answer is China does a really good job of it, so instead we make cell phone software.

So that’s what we did. We said, “Look, the drones exist.” The only reason we did hardware and software was because we needed to make a vertically-integrated instantiation of a drone that works. Once you could do that, once somebody else is making the hardware that’s good, you can just focus on the software side and the experience is still good. So we moved into software, we focused on the data - construction, geospatial - and today we’re a big company in that space… And that would have been the end of it. We went from the DIY phase, the components, to the full drones, to just the software. Now we’re a SaaS company. SaaS is good business, it’s better margins… All good. And that would have been the end of it, were it not for this current freak-out about China, the Huawei stuff and all that.

Yeah.

And you know, what happened is that the U.S. government essentially banned the purchase of DJI vehicles. It’s kind of a soft ban, but it’s pretty clear the way the writing’s on – DOD made a hard; the Pentagon made a hard ban. So the government has essentially shut it down, which is really bad, because among the things that they were doing is they were fighting forest fires the Department of the Interior in the Pacific West, California primarily. They used drones for that, to look through the smoke, spot hotspots, things like that… And they were at risk of having this important program shut down. Public safety, fire, police… It was basically a problem. The fleets got grounded.

[43:58] So we realized that this was the best thing for Dronecode. Android only succeeded because Google was there to stick it into hardware and get it out there. Before that, it just didn’t have critical mass. We don’t have a Google for Dronecode, but what we do have now is a market. There’s a vacuum in the market, where DJI is not allowed to participate… And there was demand for an open alternative. The question is who’s gonna turn that into hardware.

We did not wanna get into the full hardware business at all, but we did see that there was demand for it, so what we did is we took one of our Chinese members of Dronecode - a company called [unintelligible 00:44:40.17] in Shanghai - and we commissioned a custom vehicle from them. They were already using the Dronecode software, but we commissioned one that was appropriate for government use, and then we had them ship it over at an incomplete stage. So with the commodity elements - the motors, the batteries…

And you assemble it…

And then we assemble it, we provision it with software in the United States, we test it… And that hybrid of Chinese commodity hardware, and then American software, and ultimately cameras and autopilots etc. so the smart stuff being made in the United States and the commodity stuff being made outside the United States, built around an open platform - that hybrid approach is what we’re promoting. Right now, some parts of the government like it, other parts of the government say “Did that atom touch China? No atom that touched China shall ever touch us.” We’re right in the middle of this silly political battle… But we think we’re on the right side of it, and we think that the hybrid approach is the way that we handle the geopolitical wars of the future. Check back in a year or two and you’ll find out whether we’re right.

Well, if you have the components from China, or assemble them yourselves, and you’re controlling all the software, you kind of control the security, I suppose, except for maybe potentially sneaky software hanging out somewhere… But again, if you control the components you’re putting together, you have a control of the rough security of things.

I think so, too. There was a Businessweek cover a year ago that got turned into – it was called Chipgate, where there was this allegation that circuit boards from China had a component that was not specified on the board, and that component was like a backdoor, sneaky, whatever thing… I don’t think it’s – first of all, it’s certainly possible. I don’t think that particular story was true. I think it sort of played to the paranoia of those who wanted it to be true.

[46:37] Now there is a contingent within the U.S. government who basically says that no circuit boards can touch China. That’s fine, we can make circuit boards in the United States. If you tell me that no plastics can touch China, no batteries can touch China, no wires can touch China, then I think we’re in silly territory, and you might as well just pay Lockheed Martin a million dollars and do it the old-fashioned way… But we’re trying to find out where that line is, and right now there’s a bill actually in Congress that defines the line as the autopilot, the radio, the camera and the gimbal. If those parts – and by the way, all the software that goes into them. If those parts are made in the United States, everything else can be made elsewhere… And I think that’s appropriate.

Awesome, Chris. Well, we’re running out of time here… Tell us real quick on your new hobby, DIY Robocars, for folks interested, where is the place to go? You said there’s a meetup in Oakland… What’s the waypoint?

For sure. We have a site at diyrobocars.com. What you’ll see is there’s links to all the local meetups. There’s meetups around the world. I guess we have about almost 100 at this point. In Europe… I think this last weekend there was a meetup in Stuttgart, and one in Helsinki, there was one in Cambridge, in the U.K. etc. So there’s a list of local meetups.

There’s a number of projects that you can use to get started. One of the popular ones is called Donkey Car (donkeycar.com).

Amazon has just released – or actually, it’s not clear whether Amazon has released it. Actually, I’m gonna check it; this is gonna be a real, live check. Amazon announced something called DeepRacer, which is their autonomous car… They announced it earlier this year, and it’s been delayed and delayed and delayed. I’m gonna right now find out, in real-time, whether it shipped today, as it was meant to. The answer is…

Pins and needles…

Pins and needles… And the answer is – track package, track package… It has not shipped.

Aww…

It has not shipped. It’s literally showing as arriving Tuesday, but it has not crossed the ship line, and we’re now slightly afraid it’s gonna get delayed to next year. NVIDIA has a number of racers; one is called JetRacer and one is called JetBot. These use the new NVIDIA Jetson, which is really cool… And there will be more, but go to DIY Robocars and they’re all linked there.

I love the tagline you have there: “Fast. Cheap. Out of control.” I just love that.

Yeah… They’re all equally important. You know, fast and expensive? Not the point. Slow and expensive? Not the point.

No fun.

In control? Absolutely not the point. [laughter] If you’ve watched any of the videos of our races, the crashes are half the fun. And by the way, it used to be they crashed because the software sucked. Now they crash because they’re going so fast and they’re so competitive; they’re hitting each other as they jockey for position, and they tumble… And nobody’s been hurt.

Good.

That’s progress right there.

We’ll have to have you back on, Chris, so we can talk more at length at some point on especially where this is going. It seems like we could have camped out just on DIY Robocars alone, and had a lot of fun.

Chris might have preferred it. It seems like you get really lit up when we talk about Robocars.

I know. Well, it’s because it’s the hot, new thing. The thing is that with drones it was technically solved, all the technical problems, like five years ago. And now it’s business problems. With cars, we’re still right in the middle of solving technical problems… So I’m a nerd, I like solving technical problems.

We really appreciate you joining us today. It was lots of fun. Thanks a lot, Chris.

Thank you!