Tech Brew Ride Home - (Bonus) Among the Amazon Robots With Matt Simon

Episode Date: June 15, 2019

That Wired longread that I suggested yesterday, about Amazon’s warehouse robots really stuck with me for personal reasons you’ll hear in a second. I talked to the author of the piece, Matt Simon, not only cause I wanted more flavor on what it was like to work with these things, but because he also raises interesting ideas about human/robot symbiosis. In short, the robot apocalypse might still be coming, but not today. And for the foreseeable future, that might be the growth industry for humans: robot baby sitters. Please enjoy. Matt's article: INSIDE THE AMAZON WAREHOUSE WHERE HUMANS AND MACHINES BECOME ONE Sponsors: Vistaprint.com/ride Stamps.com, click on the Microphone at the TOP of the homepage and type in RIDE Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to another weekend bonus episode of the TechMeme Right Home. I'm your host Brian McCullough. So that wired long read that I suggested yesterday about Amazon's warehouse robots really stuck with me,

for personal reasons, as you'll hear in a second. I decided to talk to the author of the piece, Matt Simon, not only because I wanted more flavor on what it was like to work with these things, but also because he raises some interesting ideas about human-slash-robot symbiosis. In short, the robot apocalypse might still be coming, but not today. And for the foreseeable future, that might be the growth industry for humans. Robot babysitters, please enjoy. That I wanted to talk about this story specifically because, you know,

people have gone inside Amazon warehouses and fulfillment centers before. But it's a little personal because the only real, quote unquote, real job I've ever had was for six months. I worked at FedEx basically at a place like you're describing where, you know, boxes are coming down a belt and we've got to sling them into the various containers to go to the airports. It's the whole reason why I know really obscure airport codes like MCO is for Orlando and things like that. Good skill to have. Yeah, well, mostly useless skill unless you're, you know, a travel agent or something. But so I was fascinated by this one. But let's clarify at first.

In your article, you didn't visit a fulfillment center. This was what, what do they call it, a sorting facility? It's a sorting facility. So it's kind of the next step up from a fulfillment center. So a fulfillment center, you have humans that are packing the boxes, you know, putting items into a box sending it to you. That box doesn't go straight to you from there, though. It actually goes to this place called a sorting center where you have lots and lots of boxes, a lot of these flat packages coming in that are then sorted into pallets that are then loaded

into the trucks that go out to your neighborhood. Okay, so explain that to me. Because the one where you visit, this was in Colorado, like outside of Denver, right? Right, right. So how far away would... Okay, so the fulfillment center is where they actually put the things in the boxes, and then they put them on trucks and then send them over to this other place? How far away is the fulfillment center?

That I'm not so sure of. So I'm not, it's positioned right next to the airport, so presumably they're coming in from maybe afar. But it's not like it was next door or something like that. No, no. So, yeah. So this is the sorting center pretty much on its own here in the Denver area. All right. So before we describe the new system that you got to play with, layout for me, as best you understand it, the way Amazon has been doing.

package sorting all this time because I think you said in the story like there there were still some humans over in the corner doing the sorting the traditional way, right? There were. So they still have a significant section of the floor in this warehouse dedicated to humans doing it the old way and it is extremely labor intensive. So what you have, the first step is this shoot where these packages are flowing like a waterfall. And at the bottom of that shoot, you have humans that are then picking up. them up looking at the label for specific code and then throwing it into the corresponding

shoot which then goes down various conveyor belts to people actually on the floor below and those people are physically picking up those packages scanning them and then stacking them on pallets and those pallets have then loaded into trucks that go out to neighborhoods so this is of course a really interesting problem in robotics is the manipulation that we're doing here so it's extremely intensive, picking up those packages, playing them around, but robots just aren't capable of doing, which is why humans are still very much in the loop in their new robotic system. But it's just that the work is very difficult. It's lifting heavy packages and having to manipulate them and get the correct codes or else

it's going to the wrong neighborhood. Well, and that's exactly what I'm familiar with. And we had to scan them, except 20 years ago, we scanned them with these little handheld devices, and then you had to put them into a bank and have it all be uploaded and things like that. Okay, so this distribution center with the new robotic process that we're about to describe, give me a broad sense. Like, how big is it? How high are the ceilings?

Is it one big, open, cavernous space, like I'm imagining? It is. It is a extremely cavernous space. The ceilings are very high. I don't have the feet for you, but extremely high. The field itself that the robots are scooting around on, is 125,000 square feet. And that just makes up about, seemingly about half of the facility,

because on the other half, there are those humans doing it the old way. It is a very, very large facility in this sort of industrial park outside of the airport. But what's interesting about it is that it's one of the few warehouses in the world where you have humans not only working alongside robots, working cooperatively with them. But you have this interesting contrast here, these two sides juxtapose. The humans doing it one way,

the old way without any robots. And then on the other way, this kind of slow transfusion of robots into that process to make it more efficient. Okay, so describe the new system. Does it still begin with the waterfall of boxes coming down a shoot? It does. But it's a more segmented,

waterfall, I guess you could say. So you have along the edge of the field, as they call it, where the robots are all scooting around, you have humans who are lined up in stations, and they have a shoot that has packages flowing through it, and they have essentially a pile of packages. They pick them up one at a time, scan them with a little red dot that comes down in front of them, and put them on top of the robot that is also in front of them. And when they scan it, that actually uploads the destination into the robots, So all the human does is press a button and the robot shoots off into the field and finds its destination, which is one of over 300 shoots in the floor that all fall down into boxes below. So once that happens, the packages accumulate below in these various, coordinated to zip codes, essentially. And it's the same principle as what was happening in the old way. So you have, instead of the humans scanning those at the end of the line, they are falling automatically into boxes. And those boxes just go straight onto trucks bound for your neighborhood. So it's important to note that while this is a really interesting robotics solution to this

issue of efficiency, humans are very much in the loop every step of the way. And it brings us to this really interesting time and history here where we have humans for the first time really working alongside very sophisticated machines. That brings up all kinds of really interesting ethical arguments and just the way that humans are adapting to machines as much as the machines have to adapt to us. Well, we'll come back to that in a second. But so essentially, these robots, the labor that they're eliminating is the physical walking or carting of the various boxes to the, I don't know, however many humans would be assigned

to like 30 different shoots or something. Like the robots themselves are on this sort of like chessboard grid and their program to deliver these boxes to the shoots. It takes out the actual walking the box over their part. Right. Yeah. So it's just the way that it ends up is more organized as opposed to, as I mentioned at the end of the line with a human old way, you have people physically taking those packages and stacking them on top of each other in these pallets that are then loaded into trucks. Down below the field of the robots where the shoots empty into, you have boxes that the packages automatically go into sometimes.

Other times they have humans at the bottom of those shoots, then loading the boxes, the packages into those larger boxes. But it's certainly alleviating much of the hard labor here. But at the same time, creating these new opportunities for human-robot interaction. So these little robots, I keep calling these types of robots like the ankle droids because they remind me of those floor droids and the Death Star, you know. They maneuver around their environment and they have, I think you described, they've got cameras on them and things like that, but they're not autonomous, right? There's like some sort of air traffic controller controlling them. Yeah, it's a really interesting system because robots, as we know them today, as they're becoming more sophisticated and escaping the lab, and rolling around and walking around in the real world.

They're autonomous, you know, largely on their own. They have vision systems and they can avoid obstacles and things like that. But these are, you know, give up actually some of their sovereignty to a cloud system that is coordinating up to 800 of these little robots scooting around on this 125,000 square foot field. So that comes with an incredible number of issues. It's about coordinating efficiencies. So getting a robot that has to go to its pick their destination, one shoot halfway across the field, but then has to figure out how to do that with all these other robots creating traffic problems.

And you're actually watching it on the field. You'll see robots come up to an intersection, stop, wait for another robot to go by and continue on. So it gets much more complicated when you imagine that accidents happen, and sometimes the package might fall off one of these robots. And that's when they're a little bit more of their autonomy kicks in. and they can use a camera on front of the robots to actually see obstacles. But then the system will kick in and actually start routing the other robots around that obstacle to keep the traffic flowing. There's literally lots of moving parts in this kind of system.

And it's just an extremely complicated problem in swarm robotics that Amazon has taken a really interesting approach to. If I were to walk out onto the field, as it were, would they dodge me or would I, would have to dodge them. Do they have that? Sorry. Yeah, they would see, ideally, yeah, they would see them. And be able to stop and route around you. And actually, they'll call breaks every once in a while because at the end of the day, these are humans working alongside robots and humans have to take breaks.

When that happens, it's a break for the humans there, but also an opportunity for people to actually go out in the field and pick up waivered packages if they've happened to have fallen off. So again, this really interesting interplay between machines and humans. We need them as much as they need us. I also found it fascinating that, like you described, you would think, oh, well, if we can just get these things. It's all about the efficiency and the moving stuff quickly. So if we just make the robots go as fast as possible, everything can be faster.

But it's actually, that's kind of counterproductive? It is. And, you know, one of the roboticists that was walking me through this compared it to having a Ferrari in San Francisco. It's going to be able to go fast for, what, half a block, maybe if you're lucky, but you're never going to get stuck in traffic. There's very little point to having a car like that in the city. I would argue very little point to having a car like that, generally speaking, but that's neither here nor there. So what they actually had to experiment with in simulation is the correct speed that these robots are. need to travel at in order to not get tangled up in each other there.

But also, it turned out more important was the acceleration and deceleration. So obviously, you need to break at these intersections for other robots to go by and be able to accelerate to improve the time that you can get your package to your destination, but not accelerate so much that you end up throwing the package off of your back. And again, all of this is coordinated among hundreds and hundreds of robots. having to figure out how one machine gets through the field, but how that one machine gets to the field of many, many other robots. Right.

And then also counterproductive, like, I think you said that they could deploy like 800 simultaneously. Yeah. But then you have, you know, you can't just build more roads. You build more roads. You get more traffic. So, like, I think they, like, limit it to, like, 400 or 500 at a time or something like that. Yeah, that's typical.

So, yeah, you know, you throw out 800 robots. It's going to get traffic jammy. Same thing that happens in cities if you put too many cars in there. And then adding yet another layer of complication, you have to have all these robots charged at any given time. If one runs out of battery in the middle of the field, it then becomes an obstacle and decreases the efficiency of the system as a whole. So what they actually do is they let their robots get down to maybe 25% battery at the lowest, but they'll have them peel off every once in a while to go to the side and dock. in charge for about three minutes and then pop back out on the field. So it's this exchange of robots that are sidelined and robots that are active and it's just constantly back

and forth. You can begin to imagine how monumentally complex the system is. Yeah, and so, you know, what we're familiar with is like the idea of like, like even in airports the baggage sorting thing is like you're imagining like, you know, you know, this maze of conveyor belts and then a scanner runs by and something hits it down one shoot and down the other or whatever. But like this is able to achieve that on like an order of magnitude, you know, more complexity than just there's 20 gates or something like that. Yeah, it's, you know, over 300 of these shoots on the field and extra complicating matters is that some particularly populous zip codes will have more than one shoot. And then that

the system then has to figure out, okay, well, in order to have packages evenly distributed on the field, which makes traffic more efficient, the robot needs to go to this shoot as opposed to this shoot, but also if one of those shoots starts filling up more,

that becomes a problem, and then the other robots will get routed to that other shoot for that zip code. There are almost infinite number of complicating factors to this system. All right, so let's talk about the the human angle here.

The humans are still involved, what, largely because of manual dexterity of, we don't have a robot arm yet that can pick a box up off of one shoot and put it on the robot itself, basically. That's exactly. It's a problem of manipulation, which is arguably the biggest problem outstanding in robotics. Robots just are not good at manipulating objects.

Just think about the number of, of products that Amazon sells and getting a robot to be able to manipulate all of those well is absolutely impossible at this point. So what Amazon has done is, I think, admitted to the fact that, well, we're not going to be able to get a robot that can pick products and put them in a box like they do in a fulfillment center. But here in a sorting center where you are working with boxes, which are flat, and a robot which they're experimenting with a robotic arm, has a vacuum manipulative.

where they're trying to get to be able to pick up packages and then put them on these little robots that are zipping around the field. But for the time being, these humans that are working in this sorting center are very much leveraging the unique human ability to be able to manipulate a range of objects. Not only that, but to make judgments, which robots just can't do. So I think a really interesting example would be something like a thing of laundry detergent, a clear laundry detergent. and it broke inside of a box and it started leaking. And, you know, human might be able to, first of all, smell that before they even see it. And if they pick up the box, they could feel that it's sticky before even seeing it because it's a clear fluid. That is a wildly complex problem for getting a robot to handle it.

It's just impossible, again, at this point. So a human can use that very unique human judgment to say, okay, that's wrong. I need to pick that up, sideline it because if I put it on a robot, it's going to spill longger detergent all. over the floor. I think I refer to it like a snail slamming the rest of the robotic floor, which is, of course, terrible for the system to get clogged up with laundry detergent. But, you know, that dovetails with the admission that we are just nowhere near having a robot that will be able to manipulate objects the way that a human does with the same speed and dexterity. So, yeah, what are the, to your mind, what are the implications here about, you know, the whole

idea that we're always hearing is that slowly but surely the robots are replacing the human stuff. But your piece almost describes like it's almost moving the humans up into like this sort of management, not management in the, you know, middle management sense, but in the like managing the overall operations of the machines, right? Exactly. I think one of the more interesting takeaways from this sort of system, this is coming up in other areas, is the idea of a robot babysitter. Not a robot that babysits kids, but a human that babysits robots. So you see this actually in delivery robots you've probably heard about that are scooting around sidewalks.

You actually have humans in a call center somewhere. If a robot gets stuck, it can call up to that call center and be remote controlled. Essentially, the robot admitting it doesn't know what to do because robots are kind of stupid still. So you have that kind of babysitting out in our cities, but also in the sorting center, you have humans. that have been actually promoted, Amazon says, from lower-level jobs to become essentially robot caretakers, where they're monitoring the health of the system to make sure that these robots, which, again, just aren't very smart yet, are doing their job appropriately. So you have, Amazon also refers to this as upskilling.

So what we have been arguing here at WIRED for quite some time is that the robot Apocalypse taking all of our jobs is not here, and I don't think it will be for quite some time. This is automation, same as it ever was. You have robots and machines that are able to do jobs better than humans, and we're, of course, in a system of capitalism that will give precedence to that sort of thing. So what we have developing, though, is a robotics system in Amazon's sorting center and elsewhere, where robots are taking over parts of jobs, not jobs. jobs entirely. So you can think of, you know, in a more office-y setting a long time ago, we got word processors, and that made our jobs not obsolete, but more efficient. That was automation,

in a sense. This is more of a physical system, but we are seeing, and some good studies are beginning to back this up, that that robot apocalypse isn't here. What's more happening is that we are kind of intertwining more with machines in our jobs. And this is very evident here. And this is very evident here at Amazon, where you have people working very closely with robots. And it's an admission that, first of all, our robots are nowhere near sophisticated enough to do these jobs, but also that if we're going to want to be successful as employees in this new ecosystem, we will have to adapt to the machines and admit that there are things that they can do better than us.

This is, of course, a sticky slope because robots will get ever more efficient and faster and better at many billion things like that, where they will be able to take over more jobs. But in the near term, we are seeing this really interesting dynamic, this interplay between people and machines. It's a man-machine symbiosis, especially as you're describing, where when the humans have to have a break time, the robots get the break time. So the robots are dependent on us for rest. So the last question for you.

I don't know how long you got to play with this little process, but was it a fun? job putting the feeding the robots as it were I I only did it for probably about 10 minutes I wouldn't know if I could make a judgment because it was very novel to me right for those 10 minutes but it is it's really interesting it's and that's another segment of this is so much of this can be automated as we again have done in centuries past do you automate out the boring parts of the human jobs. And again, we humans do that good judgment and fine manipulation that we humans can do. So, you know, standing there doing this job, I felt proud to be a human in a certain way, but also a little tense that, you know, machines are getting better. And if we

aren't able to adapt as employees in this new world, I think we're going to start to get left behind. But It brings up these just really interesting interactions between humans and machines that researchers are just starting to talk about, you know, developing bonds with machines and naming them, that sort of thing as co-workers. It'll be really interesting to see how this develops, first of all, in this Amazon warehouse in particular, but in the job market at large, how we, again, intertwine more with machines in the coming years.