Problems Solved: Drones, Bananas and Real Estate*

Published Mar 16, 2023, 4:05 AM

It's our first anniversary and—almost 50 episodes in—Jacob Goldstein checks in with three past guests. 
Drone delivery guy Keenan Wyrobek thinks he has solved a big problem holding back commercial drone delivery in America. Fruit-ripening maven Katherine Sizov is figuring out bananas. And Glenn Kelman of Redfin has some deep insights from a tough year in the real estate business.
*The problems in real estate weren't so much solved as left behind.

Pushkin. It's been a year since we started this show, and we started it because I wanted to talk to people who are trying to make technological progress, people who are struggling with big, interesting problems, who go to work every day and try to figure out how to do things that no one in the world knows how to do. And now enough time has passed since we started the show that some of the people we've talked to have in fact figured some things out, they have solved some problems. I'm Jacob Goldstein, and this is What's Your Problem? On today's show, we're following up with three people who we've spoken to over the past year. First Keenan Wirobek. He may have solved the key problem that has been preventing drone delivery from taking off in the United States. Then Katherine Sizov, who seems to have figured out how to get ripe bananas onto grocery store shelves at the right time. And finally Glenn Kelman. He is the CEO of the real estate company Redfinn and since we talked to him last he, like a lot of people in the real estate industry, has had something of a tough time. What I've been through just personally has sometimes felt like a spiritual death of the past six months, but he has come through it with some really illuminating insights. We'll hear those later in the show. First up is Keenan Wirobek, co founder and CTO of Zipline. Zip Line is a drone delivery company. They uperated in a bunch of countries around the world. They've done hundreds of thousands of flights in Rwanda, where their drones deliver medicines across the entire country. But when I talked with Keenan last year, he was up against this kind of combination technical slash regulatory problem that was making it hard for him and other drone delivery companies to expand in the United States. Here's what the problem was. In other countries, there are strict rules that require planes to have transponders, these devices that allow drones and planes to detect and avoid each other. But rules about transponders are more lax in the US, and as a result, commercial drone companies have to figure out how to detect and avoid planes that are flying without transponders. Today, drone delivery companies like zip Line in the US solve for this in a kind of ridiculously low tech and labor intensive way. They actually pay people to sit on the ground and stare at the sky watching for planes. This obviously is not going to scale. We're not going to have a drone delivery business when you have to pay people to sit on the ground and stare at the sky for drone delivery to be a real economically viable service in this country, someone needs to figure out how drones can automatically detect and avoid planes. When I talked to Keenan last year, that was ziplines big problem. Now, after years of trial and error, he thinks they've solved it. Keenan told me the story of how they figured it out. So we went on a journey here. Um we started. We actually started with radar because radar was the thing that sort of the holy grail of FA regulators understand radar. It has the benefit of being able to see through clouds and things like this. Okay, so why didn't radar work. The regulators want drones to be able to sense aircraft coming from any direction and avoid them, and this is a big challenge. And so if you put enough radar on an aircraft, to see in all directions up, down, you know, left, right, front, back. It weighs something like five times more than our drone does. Okay, it's a massive radar system, and it's it's basically there's just no way to fly it. Okay, So it's too heavy. The simple answer is too heavy. Won't work. It's too heavy, too heavy, And so we went from there to cameras. It's like, okay, can you just look now? Cameras have some really interesting challenges, and I think the first challenge is easy to If you ever tried to take out your cell phone and like look for a plane in the sky, you've seen the first challenge. You almost it's almost impossible to see it on your phone. Right to our eyes, we can kind of figure out, oh, there's a plane there, but it's literally a couple of pixels to a camera, even a very high resolution camera, and so being able to tell the difference between you know, smudge on the lens a bird and a plane very difficult to do that reliably. Okay, So radar won't work, cameras won't work. What do you try? So this is sort of a back to the drawing board moment of like what on Earth could work. So if you remember the early hearing aids, that were these cones that you would stick in your ear and use still hear something, So personally remember them. But I have seen photos. Yes, you see photos. Okay, yes, the So before radar was invented the Brits in the United States, we built each huge cones, like twenty foot size cones that humans would stick their ear onto the end of the twenty foot size cone and this they point the cone at the horizon to try to hear aircraft coming before you could see them. Wow. So when was that in use? World War One and then early World War Two until radar was invented, which was mid World War two. So just be a dude with his ear on a giant cone and if he heard a plane, he'd say whatever, everybody into the bunker, planes coming, scramble the planes intercept this kind of thing. Yeah, and you know it kind of makes sense, right, Like as humans were pretty good, we're walking around, you hear a plane, you look up, you're like, oh, there's the plane. Right, Um, we don't see a plane first. We hear a plane first. And that was the inspiration and it was like, okay, well, we've tried the other the things that seemed more obvious, let's try this. And that this was one in turn was working with me, and it was basically saying, hey, let's you know, the philosophy is fail fast, right, It's basically, let's try to understand the technical reasons why this is impossible as quickly as possible. Let's rule it out. Yeah. So the number one reason to be skeptical is it you've ever been, like, you know, on a bike or you know, in a motorcycle or a car and it's quiet in the car, But on a motorcycle you put your head, you're out the window, and you hear all this noise. You can barely hear the person, you know, biking next to you. That's called aero acoustic noise. It's basically the noise made by the rush of air over your well. Yeah, when you roll down the window in the car when you're on the freeway, it's super loud. It's super loud, right, And so these planes, the microphones are in the rushing wind created by the plane in the sky, by the drone in the sky exactly, So is there Basically the question was, with all that noise from that rushing wind, would we still hear those aircraft far enough away to be able to avoid them? That was the biggest thing that we were worried about. Um, but we we we found a way using a microphone array. So basically, you know, a set of microphones on the aircraft to be able to literally listen for where aircraft are and use that to avoid them. So so you're saying it works, You figured it out, Tell me how it works. Yeah, So so the final system across the wing, um, there's eight microphones and they're on. They're on, and they're on little they're on little polls. They look like basically like a you know, I don't know the length of a straw, like a drinking straw coming out the front of the of the micro of the wing. There's eight microphones and now you need a bunch of microphones to do basically what your ears do, which is beam form, which is basically you figure out how the sound hits one microphone a little sooner than the other one, and it helps you understand, oh, the noise is coming from over there versus you know, coming from your right side versus your left side, so that that's what it looks like on the aircraft. And then okay, so this is the hardware piece. Is there a sort of software layer. I've got the hardware piece in my mind. You're gathering this sound. I mean, I mean, I know everything sounds like an AI problem now, but this one really does sound like an A problem. Right. You're pattern matching because you want to say it, does this sound like a plane? Let's test it against a billion other sounds? And I don't know how what's the software piece work? Like, Yeah, there's a big software piece. The software has been far and away the biggest effort here. There's a there's a tiny team of hardware engineers and a big team of software engineers working on this. So how does this software work? So the software has layers, and the first layer is a layer called beam forming, and beam forming is this technique very similar to what our ears used to figure out? Okay, where are the sound sources coming from? Right now? That's the first layer. Okay, that's cool. So now you know where the sound is coming from, but you don't know what it is yet exactly so, and then the next layer is an AI layer where we've we've done thousands of flights with different aircraft, different sound sources, and we've used that data to train AI model to actually figure out basically it reinforces where it's coming from and to figure out, hey, this is probably an aircraft, this is probably not an aircraft. And then the AI just says, yes, it's an aircraft. I've detected and now avoid avoid avoid or what what's the last So there's two there's two more layers here. So the next layer is it's called the tracking layer, and the tracking layer is you kind of think of that that layer is the layer that remembers over time, right, So this takes the information from this AI layer and over time says, oh, you think something's kind of over in that direction, I'm gonna I'm gonna keep track of that, and every every you know, every fraction of a second, it gets an update and it kind of updates that track of sort of where it thinks that sound is coming from. And then the final layer takes those tracks of where uh these these and these tracks are estimates, all right, you kind of think of them as like, hey, I'm there's you know, there's an there's an aircraft in that direction kind of over there. And the next layer is the layer the planning layer, the layer that takes that information and says, okay, there's an aircraft over there, I better go over here to say away, I better go in the other direction, exactly exactly. So okay. So so now you have solved this non trivial technical problem. There is also a regulatory problem, right, you have to get approval to fly your commercial drones without people on the ground looking at the drone the whole way. The FAA has to say, yes, we believe that this works and you can fly your drones. Do you have that approval? Yes, But let me explain it because, as in all things with from a regulatory perspective, it's subtle. So what's really exciting is we just got yeah, yes, Asterius. So there, basically we're about to start this walk crawl run sort of progression, which is how you should start a new technology like this. So we just got approved to fly this in operations in the United States. So in the crawl walk run, the crawl phase which we're about to start, which is basically this technology turned on, so doing the avoid les, sensing and avoidance, but with still these visual observers standing on the ground staring at the sky and then later this year we go through a process with the FA to remove the visual observers and keep operating. And that's the holy grail moment that's coming soon. So the holy grail moment is flying a commercial drone flight in the United States without someone watching from the ground exactly. And when do you think that will happen? If I'm being optimistic, it's about three months from now. If I'm being more realistic, three to six months from now. It seems like this has been the big underappreciated by the general public problem that has prevented commercial drone delivery from taking off. Excuse my fun in this country? Is that right? And like you think you've solved it, and the FA thinks you've probably solved it. Yes, this is Yes, it has been a long few years to get through this journey, but I am over the moon excited about this because this is this is the big unlock. Great, good to talk to you. Can we talk in a year, Let's do it, okay, twenty twenty four, back to you then, excellent. One other piece of zip Line news that Tenan told me about. The company is working on this new kind of drone. It's this wild two part system where the drone itself hovers like three hundred feet above the delivery site, and then it lowers down this little box on some kind of wire. The box goes all the way down to the ground, opens up, delivers the package, and goes back up into the drone. They hope to have that in operation by early twenty twenty four, so that'll be one more thing for me and Keenan to talk about next year. Still to come on today's very special episode of What's Your Problem, The Banana Ripening Frontier also how higher interest rates are profoundly changing not only the real estate business, but lots and lots of businesses. Next up, Katherine says Off, founder and CEO of a company called Strella Biotech. Catherine and Strella, they developed these sensors that detect when apples and pears are getting ripe. They work not like in the grocery store in your house, but in these giant storage rooms where apples and pears sit and ripen. The idea behind the product was to reduce waste, increase efficiency, that kind of thing, and the sensors are based on this thing that happens in nature. There are lots of kinds of fruit where when they get ripe, they emit ethylene gas, and that gas is like a signal that causes nearby pieces of fruit to ripen more quickly. When Catherine and I talked last year, she was trying to solve ripening problems for other kinds of fruit besides apples and pears. When I followed up with her earlier this month, I asked her, is there some new kind of fruit that you've cracked since we last talked? Bananas? Actually bananas. Yeah, it's a big one. Yeah, it is a one. Tell me about bananas. What happened and when did it happen? Yeah? I guess my first question is how do you like your bananas? Because, says a heated topic, I prefer them a little too ripe as opposed to a little bit unripe, Like if it's still kind of green, if it's still hard to peel, if when you peel it it's hard to peel, and then you eat it it's a little chalky. That's that's not my favorite banana. I'm one of those freaks that like, say, a little bit underripe, actually, But I think you're in the majority. So what have you figured out about bananas, Like what was the problem and then what did you figure out, Like what was the status quo whatever a year ago? What was the banana problem? So the banana problem is that every time we go to the grocery store, the bananas tend to be a little bit on the green side. I don't know if you have this from personal experience, but they tend to actually edge towards green and then sometimes it goes to the other direction where they're way too ripe and no one wants to buy them. Right. But basically what happens is that they're brought into North America totally under ripe, so they haven't they're not mature whatsoever, and they're green. When you peel them, it's very hard to peel them. They ooze latex, so they're like if you try to peel it, it's it's kind of almost like bleeding latex. And then what happens is these bananas are put into what are called ripening rooms. So basically bananas go into these rooms and then they're dosed with ethylene gas, which is what makes them ripen. And this is just to restate when we talked about last time, like in the wild a ripening piece of fruit or ripening banana emits ethylene gas and that is a signal to other surrounding bananas that they should also ripen. Right, So the dosing with ethylene gas is just doing what happens in nature. That's exactly right. So they're put into what's called a ripening room and they're dosed with ethylene gas. And then comes the art of it, which is that every day a ripener that's a job will go into that room and say, hey, or these bananas more green or less green than what I was expecting, And they just look at them visually and they say, mm, doesn't look quite yellow enough, and then they adjust the conditions of the ripening room base on that. So it's a superhuman, super subjective way to do it. And we thought, you know what, this is probably the cause of a lot of inconsistencies. Every single ripener has a different idea of what a stage seven banana looks like or stage four banana looks like. And so let's take the human out of this, let's take the art out of it. We know about fruit physiology, let's turn this into a science. So what we did was we put our sensors inside these ripening rooms, and we can tell how the bananas are reacting to their conditions and how they're ripening. And then we throw some machine learning on top of that and we say, hey, this banana is going to be under ripe if you keep going the way that you're going. So we're basically going to the retailer and helping them ripen their bananas for the store shelf. Oh huh, so how does that work? So a grocery store can you name it? Are you working with some grocery store chain? I would have heard of a big one in Canada. Okay, can you say the name? I can't, sorry, like hundreds of stores across Canada or something. Yes, So, so how do you go into the grocery store? So you go talk to a ripener and they say, you know what, You're totally right. This is a huge problem for me and I have a million other things to do, and I will Oh. I wouldn't have expect that. I would have expected them to say I'm a genius at ripening. Go away. That happens too sometimes. But I think the ones that you know in general, when you're when you're good at your job and a new tool comes along, You're like, hell, yeah, that's awesome. So I think most people are really interested. So okay, so you go, you talk to the ripeners. Keep going. So we talk to the ripeners. They're like, I have a million things on my plate. Ripening in bananas is one of them. I don't know if I'm doing a good job. Would love to see what's up. So then we put up We set up shop in a couple of ripening rooms. We work with the ripener to understand what they're doing, why they're doing it, and train our model on it. And then we expand and how long did it take you for your sensors to be better than a ripener? It's we're still going. So we're kind of well, let me ask you, are you better than a human ripener? Now? Well, it depends, right, So if what we can do is we can basically give a tablet to a person off the street and you can go ripen a bunch of bananas and not do a terrible job at it. So that's pretty important because it usually it takes three to five years to train a ripener, yeah, in the art of banana ripening, But now we can Now we can kind of nagate all that in terms of like the ogs of banana ripening, I don't necessarily think so, but with a little bit more time than I think, we'll get there. So basically you can make anybody as good as the sort of average five years of training banana ripener. That's right now, yep. How long did it take you to figure to do that? Oh? My god, well, I mean I have. I didn't talk to you too long ago, but to me it feels like basically an eternity. It was a year. It was a lot of year for you. You did a lot of very long year. When you think you'll be doing it in the US, when do you think some US grocery store will be buying your service? We're in trials with several grocery stores and distributors, now, okay, in the US, yes, anybody you can name. No, okay. Automating the task of figuring out how to ripen bananas is not some world changing technology, But you can zoom out a little bit here. Creating machines to automate one task or another has been at the very core of technological progress for hundreds of years. In the long run, little automations like this are like the fundamental efficiency game, getting more and more efficient at lots and lots of little things. That is how over the long run we have raised living standards for most of the people on Earth. In a minute, we'll talk tech and real estate giant macro economic fluctuations with Glenn Kelman, the CEO of Redfinn. About a decade ago, this new kind of thing popped up in the real estate world. It was called ie buying. If you wanted to sell your house, you could go on to a few sites on the internet, enter some information about your house, and a big company like say Zillo, would actually buy your house for cash. Then they would fix it up a little turn around, and try and sell it for a profit. Zillo famously lost a ton of money on this business and they got out in late twenty twenty one. Another big eyebuyer was Redfinn. Redfinn is a big brokerage operates all around the country. And when I talked to Glenn Kelman, the company CEO, last year, Redfinn was still in the eye buying business. But Glenn knew that if house prices started falling, it would be tough to stay in that business, and he told me, so in very dramatic terms. I brought this up with him when I talked to him again earlier this year. So, look, the last time we talked was in the first half of last year, and we talked about iBuying, and you said that I buying would be good on the way up and on the way down you said, in your words, it would be a roller coaster ride to hell. Did say that? Remarkable prescience, remarkable prescience. So tell me what happened with Redfin's I buying business since the last time we talked in the first half of last year. Actually, we closed the business, and we did that for two reasons. One is very intuitive, which is just that home prices fell dramatically between May and the end of twenty twenty two, and we were left holding the bags of homes that we had to sell by hook or by crook. But the other factor is just the increasing cost of capital. So you're basically fronting the money to people who own a home so that they can move on. And when capital was nearly free, we were able to front that money at very low costs, so the offers we were making to people were very competitive. The idea being that that very low interest that almost free money you were able to borrow was core to that I buying business. It is because you're giving someone five hundred thousand dollars for the three or four months that it takes you to rehabilitate the house, put it back on the market, and get someone else to move in. It's a wildly capital intensive business. It's a wildly capital intensive business. To actually be the principle in a transaction where you're the source of the money yourself, that was something that was new. We saw that with we Work, we saw it with ibuy, we saw it with all used car marketplaces. And so that was all driven by this ultra low interest rate environment that drove so much of the tech boom right correct. And so we felt that even if housing prices stabilized again where we could have a reasonable expectation that we could buy a house for five hundred thousand dollars and sell it for five hundred and thirty thousand dollars, what had really changed in a permanent way is we had gone back to a world where capital would be expensive and the cost of fronting that money would be so high that we would have to offer much less for the house to offset the borrowing cost, and that has been the way most businesses operated for most of time, that if you're fronting money and taking risk, you're going to charge the consumer a hefty premium for that. And that suddenly was an offer that I wouldn't take if I myself were the consumer. And so if you're selling a product you don't believe is good for your customer, you've got to stop. Ellen, I want to talk sort of beyond housing, Like you were the one who told me a long time ago that I buying existed in part because companies like Redfinn had suddenly had access to just an incredible kind of unlimited amount of really cheap capital. And that was one of those things that when you told me that, it helped me understand the whole economic era we were living through. It really did, and I thought about it a lot. And now you're telling me that that era is over, and you know, obviously the world is telling me that as well. And so I'm curious what you think it means, not just for housing, but for the broader economy, for you know, technology companies, for innovation, for the economy more gendure. What's it mean? Well, what I've been through just personally has sometimes felt like a spiritual death of the past six months, because if you pride yourself in part on your creativity, as many people in technology do, your basic attitude toward almost every idea is yes, And even if that idea takes ten or twenty years to pay off, if the cost capital is nearly free, a dollar twenty years from now is almost worth the same as a dollar today, so you might as well make the investment. And so I've had to shift my own approach to running Redfin from one that's very entrepreneurial I hope that's always still part of Redfin, to something that's a more bead eyed allocation of capital, where people come up with a hundred ideas, all of them are very exciting, but if it were your own money, you'd mostly say no. And even though I always prided myself on spending Redfence money as if it were my own, there was such a premium on growth that you took more risks. And sometimes I took the risk because I wanted to do it, and at other times I took the risk because I felt like I had to. It was like one am at a casino table. You've got a lot of chips, you think maybe I should just stand up and walk away, and just everyone in the room is yelling at you to keep pushing the money out to the middle of the table. That wasn't just with eye buying. It was a whole bunch of different initiatives, and so that can be a pretty stressful way to live. But it's also creative and exciting and exhilarating. And now I just think this is a good way to live too. I love my job more than ever, even though we've been through hell over the past twelve months. But it's more important for me to say no than to say yes. And that isn't always creative, but it's sober and responsible. It's about time. I'm fifty two years old, Jacob, It's time to grow up. Yeah. So, if the metaphor of the teens when money was free was, you know, be at being at the table at the casino at midnight, the pile follow chips and everybody's yelling at you to go all in, keep keep betting, what's the metaphor. Now you're in your hotel room. Only you have the code to the safe. Nobody uses that safe, But now your balance sheet that's safe really matters, like how did you do in the end on I buying, Well, we can't call it a win. It was it a profitable business, but we didn't get our fanny handed. Two is either you own any houses. We've got like a few houses to sell, but not many, and obviously there's always a sting in the tail. So the last house you sell is the ugliest one. So you know, never count yourself lucky until you're dead, I think, is some saying of the ancient Romans. We'll see how we do on that last property, but we think we've mostly gotten out alive to wisdom of Glenn Kellman, CEO of Rickfin. To everyone who has listened to this show over the past year, thank you, thank you very much. If you want to help us grow into our second year of What's Your Problem as I see it, you have three options. Option one, recommend the show to a friend, Option two, leave us a nice review on your podcast s app or option three. So just a guest for the coming year by emailing us at problem at Pushkin dot fm, or by letting me know on Twitter. I'm at Jacob Goldstein. By the way, options one through three are not mutually exclusive have at It. Today's show was produced by Edith Russlo, edited by Sarah Knicks, and engineered by Amanda k Wong. I'm Jacob Goldstein, and we'll be back next week with another episode of What's Your Problem.

What's Your Problem?

Every week on What’s Your Problem, entrepreneurs and engineers talk about the future they’re trying  
Social links
Follow podcast
Recent clips
Browse 143 clip(s)