Paul Clarke takes RoboBriefings on a look-see behind the scenes at the robotics, software and automation powering Ocado’s Smart Platform
In this RoboBriefings podcast, editor in chief Tom Green talks with Paul Clarke, director of Ocado Technology, the technology development division of Ocado Group Plc, the U.K.’s largest online-only food retailer ($948 million (2014) OCDO: LONDON).
Clarke discusses the technology breakthroughs and technical challenges Ocado has had to face in building out the automated warehousing/ ordering//delivery systems as well as the launching of the next phase of development: The Ocado Smart Platform, together with its 3-way business model.
Ocado Technology is 600 engineers pioneering a first-of-its-kind grocery delivery platform where all or most of the technology is developed in-house with little to no outside software/hardware intervetion.
“Disruptive technology like no other,” he calls the build-out of the platform; a platform that Ocado has plans to license to other grocery retailers worldwide.
From inbound automation to highly automated warehouses to a mobile delivery network that guarantees home delivery in one hour from time of order, Clarke takes RoboBriefings on a look-see behind the scenes at the robotics, software and automation powering the Ocado mega-business.
Clarke also details Ocado’s prototype-building of SecondHands,a humanoid co-robot designed to assist Ocado’s in-house maintenance personnel, in conjunction with the EU’s Horizon 2020.
Podcast Transcript
Tom Green: Welcome to another edition of RoboBriefings, the online podcast series from Robotics Business Review. I’m Tom Green, editor and chief at Robotics Business Review.
And today we have a special treat. We have Paul Clarke. He’s the director of technology at Ocado, the world’s largest online grocery retailer. He’ll get into it a little bit more about that, himself, but what he says generally about Ocado is, “It’s the world’s largest online grocery retailer, and we do it like nobody else. Our customers’ orders are picked and packed in huge automated warehouses, the largest of their kind in the world, before being delivered to their kitchen tables in one-hour delivery slots by our dedicated delivery fleet.”
To give you an idea of what goes on there, this gentleman has warehouses full of goodies, and there’s about 176,000 items in some of thesefacilities. And they currently reach about 72 percent of British households, shipping over 1.3 million items a day or 170,000 orders a week. So that’s a lot, and they happen to compete in a very tight space with organizations like Tesco, Sainsbury’s, and Morrisons. In fact, they’re doing Morrisons deliveries now. That’s a new wrinkle, and we’ll get into that with Paul in a moment.
Paul Clarke: Hi there.
Tom Green: Paul is an engineer and he runs what’s called Ocado Technology. There’s a difference between [Ocado Ltd. and] Ocado Technology [is they] are two organizations. And he’s just about built it. He’s got about 600 software engineers and IT specialists whom he’s responsible for, and they are rebuilding and re-platforming this magnificent automated system today, and beginning in the near future. He has not only done that, but he’s also going to create together with Horizon 2020 a humanoid robot. And we’ll get into that robot, which is called SecondHands. So welcome, Paul. How are you?
Paul Clarke: I’m good, thank you. Thank you for the introduction.
Tom Green: You must be a very busy man with all these things. You’ve got 600 cherubs over there to monitor and watch and make sure they do everything. And you’ve got a gigantic organization. Which—catch me if I’m wrong—we began writing about you folks a couple of years ago, and if memory serves me, three Goldman Sachs bankers began right after the millennium [with] this idea of delivering groceries directly to people’s homes without having brick-and-mortar places. And they’re going to automate it, and it’s going to be fantastic and wonderful.
And here we are, it’s 2015, 15 years later. You came aboard about 2006, I believe. And I was going famously. So what are you guys doing that’s so great over there, and how do you get it done? I know you’ve got like Swisslog, auto stores and things—all kind of stuff. Can you give us a little idea of the Ocado machine?
Paul Clarke:As you said in your introduction, we do it like nobody else in terms of the scale of automation that we bring to bear. And that was always the kind of vision of the founders from the early stage, that rather than pick and store or in dark stores, that to both deliver all the kind of unique elements of our model. The only way to do it was to build these huge centralized fulfilment facilities.
As you say, we have the two biggest in the world of their kind, and in fact, we’re building the next two now, so we’ll soon have the biggest four. And really from an early age, because we’re breaking the mold here in terms of how people do this, we have no option really but to create the solution of the technologies ourselves, and that’s the responsibility, in part, of my division at Ocado Technology.
And so we build almost all of this kind of iceberg or Aladdin’s cave of technology that lies below the surface, that very few people see, and we build it all in-house. We buy almost nothing, to the constant annoyance of software companies who ring me up. But it’s served us very well, and we’re going to continue to do that. So it’s a very, very broad and deep technology estate. So it ranges on one side from where customers place their orders on our websites or mobile apps through supply chain, automated warehousing, and then the whole last-mile operation, which we can get to if you want to in terms of all the telemetry and optimization that goes into that.
And then underneath it, there’s an even another whole ecosystem of things that people wouldn’t expect a retailer to have. But we’ve had to build very complex simulation and emulation, vision systems, robotics, data science, analytics, and so on. In part, it’s because groceries are very, very different from non-grocers. We call it “non-food,” and if you can do grocery, you can do non-food, but the reverse is not true. So a lot of our difference comes from the differences to do with grocery in terms of the size of the order and how people place their orders and all that stuff. But so hopefully that gives you a sense of some of what we do.
Tom Green: There’s some really good things about pioneering, though, as you are right now, because it’s a big world. Everybody eats, there are a lot of chains all over the world feeding people, and they may take a liking to what you’re doing there in Ocado. They may be licensing your technology, your techniques, whatever software you happen to be developing. I know Morrison’s just sort of hopped aboard, they just said,“Well, take it over for us, please.” So this may be a trend.
Paul Clarke: Indeed. Really, if you go back to when we were founded, it was always the idea that we were effectively going to evolve a solution first for ourselves. But given the scale of investment of hundreds of millions of pounds over the years, well over the other side of several billion dollars, that we’ve invested in building this solution for ourselves, that we weren’t doing that to distribute groceries inside the U.K., but when the time was right, we would make that technology available to other retailers.
And as you said, we did it for Morrisons last year in the U.K. who are a big bricks-and-mortar retailer, and then literally the day after we put them live, we set about a very ambitious replatforming exercise. In fact, “replatforming” is not really quite the right word. It’s really an end-to-end rewrite of all of our systems to run in the cloud and to take advantage of cloud technologies, so that’s a very unusual cloud development.
It’s very different from doing an Uber or a Snapchat that’s quite monochromatic in terms that it may be very big, but it’s much narrower in terms of the breaths of of use cases. Whereas for us, it’s that whole spectrum I talked about internally that has to be rewritten for the cloud, and that certainly has got the attention of technology companies like Google and Amazon or AWS, with whom working with closely. So that’s one half of it.
And then the other half is building a new kind of set of technologies for creating these highly automated warehouses that we operate. Because although in a mature online grocery market like the U.K., you can justify spending you know, £230 million in one go to build one of these warehouses like the last one we built. But that’s quite a tall order in a new or emerging market, and so we wanted a way to build them in a more modular and scalable way without sacrificing any of the optimization and productivity that is at the heart of our model, and that’s what we’ve created. And we’re using that actually to build our next two warehouses in the U.K.
But sandwiching that software in the cloud and that new technology for warehousing together, you come up with what we call the Ocado Smart Platform. And we are going to use that to put some of the largest incumbent bricks-and-mortar retailers from around the world online, using our disruptive business model. And indeed, we’ve been talking to many of them over the past six to nine months, and we’ve said publicly that we intend to sign our first one this year. And that’s very exciting. So that’s going to turn us from a fusion of a technology company and a retailer as we are now to this three-way combination of a technology company and a retailer and a platform business. So that’s a very exciting new chapter for Ocado.
Tom Green: Now, OK, so you have totally automated goods going to people’s kitchen tables—what about goods coming into the warehouse? Are those as automated as well or…?
Paul Clarke: So obviously, we’re still reliant on the normal kind of supply chain in terms of using the road network. Although, there it is different because the goods come pretty much direct to us from suppliers. That’s very important because that’s one of the other big things that we gain through our model. It’s much more of the available shelf life it’s actually available to the customer because it’s come in one hop to us rather than through a normal distribution chain. And that allows us to pass much more of that freshness and shelf life onto our customers. And the same would be true for our platform customers too.
But at the point of receipt after that, it’s all very different. So then that highly automated where it all gets stripped down, and the goods get put into different kinds of storage media, or depending on their size and shape and how many we sell per hour and so forth. That’s when all the kinds of automation take over in terms of cranes and robots and other kinds of things to store the goods away into a point at which we actually need to pick some of them into a customer order. Then they’ll be brought out into different kinds of picking stations within our automated warehouses and will be picked into customers’ orders. So yes, the inbound side is heavily automated just like the outbound side is.
Tom Green: Now I know you’re very big on vision systems, and vision technology seems to be something that has really taken off. Last year, the U.S. Patent Office recorded the most amount of patents for vision systems yet. So there’s really been an explosion there, and I think that would be great if I was getting groceries from Ocado because I don’t like to have my apples bruised. And maybe you’ll come along someday with a vision system that can save me from the torment of getting bruised apples delivered from you folks. So, what are you guys doing that’s so special about vision systems?
Paul Clarke: Well this kind of goes back to what I said about that groceries is different. I mean we’re picking on average, about 50 items into a customer order, and that’s across three temperature regimes. So frozen for what we call “ambient and chilled”—actually there are four in all. There’s another one called “produce,” where your apples would sit. But unlike picking toner cartridges, CD, or whatever into an order, for food, there all sorts of soft and hard constraints to do with what can sit on top of what, what will fit with what because food comes in irregular shapes and sizes. It has short shelf life.
There are all sorts of food-separation rules … you can’t put bleach in with sausages and so forth. So we have to obey all of those rules when picking as many groceries as will fit in a box for a customer as possible. At the moment, that’s the process by which we decide what the optimal fit, which includes the 3D orientation by the way, is all done. It’s all automated. But often, a human being is doing the actual final pick.
The ultimate goal is to have robots involved in some of that. And unlike fitting a wind screen into a vehicle on a production line, which is quite a repetitive process, because of the 45,000 different types of items that we have in our catalog, which is growing all the time, how you pick a bag of oranges is very different from a pack of diapers or a soft fruit. And therefore, it’s complicated in terms of how you pick up and manipulate those products, which is the subject of one of our Horizon 2020 programs actually.
But also, it’s much, much more complicated from a vision point of view because it’s not a repetitive task. You’re going to have to take account of what has already been picked, and you’re going to have to take account of how you position the item you’re picking on top of what’s already there in the carrier bag. And the actual robotics side is much more straightforward. It’s not easy, but it’s not so straightforward. The complicated bit is the vision systems, and there are no commercially available vision systems to do what I’ve just described. So we’ve been having to work on creating our own. So we have teams of people with more Ph.D.s than I can poke a stick at. But [they] work on those things, and they share between them, vision systems is a common kind of glue, but then we will use on many different sorts of applications.
Tom Green:Well, Mother Nature had a 5 million year head start on us.
Paul Clarke: That’s true.
Tom Green: So we’re doing a pretty good job of it, so far. We’ve got a bit of a ways to go. The vision system seems like a really tough problem to solve, but it’s solvable. There are all sorts of solutions popping up here and there in research labs. So if we can get those commercialized, and who knows, you folks may come out with the winner in vision systems, which will be good for the rest of us.
I really like this philosophical point you make here in your preparation for your talk at RoboBusiness: “Staying disruptive is all about first or second derivative. It’s about acceleration, not velocity. It’s about how you get better at getting better.” I really love that part. “How you disrupt yourself first and being your own worst enemy.” So that’s what you really have to be when you’re ahead like that, don’t you?
Paul Clarke: Absolutely. There’s no place for complacency if you’re a disruptor because you will get knocked off the top. I mean we already have the world-leading solution for doing online grocery, and that’s not just in terms of the scale of automation, that’s also in terms of productivity and other metrics. But we are not at all complacent about that, and hence why we’ve been working actively on how to disrupt ourselves at many points in our solution. And that’s what this new warehouse technology, for example, will do. It will leapfrog where we already are. So that’s kind of a great way of staying ahead. But it’s also why, if you look at our different innovation streams, we have, as a disruptor, a lot of our day-to-day innovation is just businesses as usual. It is what we do and it includes a lot of things that many companies would consider research and development.
Then we have what we call “research and development” that sits beyond that. Which it would include things like the robotics and the vision systems, simulation, data science, and so forth. Then beyond that, we have what we call our “10x streams.” So that’s the kind of you know, based on the kind of Google type idea of a 10 times leap rather than a 10 percent improvement. And that’s where we try and disrupt ourselves. That’s where we try and be our own worst enemy, and it’s a constant kind of way to make sure that we never feel “good enough” is good enough. However far ahead you are of the competition, it’s never far enough. So that’s where we try to make sure that we stay disruptive.
Tom Green:Well, you definitely have stayed disruptive because you would never think, in that ecosystem that you’ve built there, that you’d try to introduce a humanoid robot to help engineers fix things around the plant. It’s a different way of looking at things, and I was just shocked when I saw it in Wired, and I said, “These people are really pioneering a lot of areas. Are they doing this just too sort of help out with Horizon 2020, or is this something that’s going to be very viable within the organization itself?” So could you take a minute and tell us about the genesis of SecondHand and how you got into it and where you intend to go because it’s a fascinating project?
Paul Clarke: Sure. So this is a little different from our kind of mainstream because, as you say, this is about creating a humanoid maintenance robot that will provide a second pair of hands to a human engineer, whether it be to help them do things that they can’t do, like places they can’t reach or taking temperature measurements or things that would require extra tooling that we could build into the robot.
But it’s also about how to keep them safe, and once again, rather than teaching the robot a set of kind of, learnt operations that they can repeat. What’s very ambitious about this project is the level of inference that we want the robot to get involved in. So the robot is going to actually learn by observing the humans, and it’s going to learn about what is normal and what might be planned exceptions.
What if there’s a sign of something has gone wrong? Or where might the human have missed out a step? Whether it be taking screws out of a side guard on a piece of conveyer, and the human has missed a step out, which actually means that something might fall on them or whetever. And the robot actually needs to step in and warn them all.
So there are lots of parts to this. There’s vision, there’s dexterity, and then this whole inference piece, and this is why it’s a collaborative project with a number of different research institutions across Europe. It’s over five years, and it’s really going to push the bounds on many of those different fronts. We’re very excited about it, not just about creating a maintenance robot, which will be great, but it’s all the other things that it will teach us along the way that we may be able to use for other purposes.
And ultimately, as I talk about in the Wired article, although we are aiming at maintenance, ultimately, if we’re going to pepper the planet with these automated warehouses, we also need to think about how we automate the process of building them in the first place.
Tom Green: From maintenance robot, to robot builder. I like that. Thank you Paul. We’ve been visiting with Paul Clarke, director of technology at Ocado. Thank you very much; have a great day.