Operating autonomous cleaning robots at scale with AWS IoT

Hi, everyone. So I'm really excited today for two reasons.

The first one is I'm going to talk about cloud robotics with AWS and K is also going to fill the concepts and all of the theories with their insights. So they'll add some color to it and especially, yeah, bring their view to cloud robotics.

And the second point, I cannot really tell you yet. For those here in the room, you might already guess what's going to happen. But let's see.

Alright, first of all, let's talk about the agenda quickly. So I'm going to start about with the basics of robotics, talk about some challenges that our customers typically see and then move to the evolution of robotics, how we got to the current point and also talk about how that could be done on AWS with the services that we offer in the foundations of cloud robotics. Then I'll hand over also to Carter to bring in their view. They are going to talk about the journey on AWS and also their next generation of robotics.

Alright today with me and by the way, I'm solutions architect with AWS today with me is also Leonard. He is the vice president of IT and digital transformation at K. But yeah, let you introduce yourself.

So thank you very much. Thanks for the introduction. Also from my side. Very nice to be here. I'm excited to showcase what we did last year in cloud, but especially with robotics.

Um many of you might not know our company, especially if you are from the US. I learned also from the US colleagues. We are not too proud from the logo. Um so therefore, I wanted to say some words about what we are doing. We are a family owned company founded in 1935. Um so quite old um and founded by Alfred Kasha and his wife. Um he was the inventor um but died quite early um end of the fifties and his wife took over and expanded the company during the years starting in, in France and Switzerland to a multinational company. In the meantime, in more than 70 countries around the world. So quite global and more than three billions of revenue last year, we are 50,500 employees. Um quite big. In the meantime, we doubled the last 10 years. There was a lot of growth and our main product we got famous with in Europe is the high pressure washer. So this is where most of the people know us about. Um but also I hope today I can give you more insights, what we do else. There's a lot of what we do and this is what I want to share.

We have a purpose in our company and we tried to find a good background to our presentation, um which should fit to the, to the story and it's renewed to sustain together. We make a powerful impact towards a clean world and how we do care about the clean world. It's about our products. So basically, we have two different business areas. On the one hand side, we have the consumer area where we have high pressure washers, vacuum cleaning, steam cleaners, window vacuums, really consumer products. It's about 50% of our revenue roughly. Um and this is also what most of the people know in the households. On the other hand side, we have the professional area. I tend to say this is for it guys, the 24 by seven products. So these have to run in very hard circumstances usually and also long time. So it's not like the consumer side. And we also have here high pressure washers, but also we have scrubbers, sweepers and also municipal. You see a big municipal device we have. Um and yeah, it's a quite a variety of products we have there renew to sustain.

Also, this is a part of our DNA and this is what we show here. So we are um cleaning um monuments like the Mount Rushmore or the stairs in Odessa for good cause to not only have it clean but to keep it for our next generations. And this is what our DNA is about in the company, to really strive for the future and to see what's next and what's coming next.

And with us, with that introduction, I give it back to you.

Alright. Thank you. Great initiative.

So then I want to step back a bit and talk about the view of AWS on robotics and its current challenges. And we start with a few in the field of the internet of robotic things. Our customers are typically struggling with the nature of robotics or robot has multiple sensors, sensors for motion images, video, and you need to combine all of those and process those in one single place and one single place is also a good, a good hint for the next one. You have typically a robot in an environment that is embedded in the physical world with the buildings and and yeah, optimally, it would interact with the building and it also is embedded into the system. So in your back end systems for planning the next route of cleaning and so on. So um that's that's one of the challenges that our customers are also facing.

Now, we have all of that data and customers are struggling with the two speeds of data. So the first speed would be real time data that you have to capture on the device and act accordingly. And also, yeah, the second one with the batch analytics, the slower time, the slower speed in the data processing. And for each of those, you need to set up different pipelines, different ways of processing and also different ways of acting on the data.

Then um yeah, we we we're going to get into an area which is more in the field of of what culture is also going to present today, more and more. So one is the over the air updates and building that mechanism to securely push out new updates and operate in the future is one of the key things that our customers are also looking to get solutions for.

Then when you have multiple robots in the field, you also need to look at their their status, you need to look at the health of each robot and sometimes you don't have that control anymore. It's out in the field, but you still want to understand its, its processes, its status. That's why you need a fleet management and promote access in a secure way.

Lastly, our customers are also looking into developing the next generation of robots and that is highly based on data and machine learning. So enabling that robot to also operate in in environments that it has never, ever has never seen before, is crucial and is is what you want to. Um yeah, also achieve.

Alright. So since we come from a more engineering background, software engineering background, we started with the challenges but obviously there are also some opportunities when you get that right. Robotics will drive the transformation and we will hear more and more stories from culture today. But other industries, for example, logistics or construction, they are actually benefiting from, from the um more and more usage of robotics and in oil and gas. We also have customers that are applying robotics in fields and areas that are so yeah, um extreme where a human would not be able to operate and getting that right, getting the deployments right in those environments is is key.

Alright. So how did we get here? How did we get to the current generation of robotics? Let's start with the first generation of robotics. And in the first generation of robotics, it was more like a process oriented way where you would um statically say, move your robot arm for example to one place to another or it would even be directed directly by the by the human itself so that the tasks would be more in the field of like guided and preprogrammed tasks.

Now, moving to the current generation, we see that robots are more working together with, with humans in their environments. It's not as easy to predict anymore. And all of that would be based on, on data and data analytics from the past. And for that, um yeah, kata also has a great story to share and I'll hand it over to you, you know, also we had some experience with the first generation of robotics.

Um we started to invent a robo vacuum cleaner in 2002 already. And the interesting thing is that that product was not sold like a million times. And um 10 years later, competitors came in and sold a lot of products. And there was a good reason for it because when we started in 2002, there was no smartphones, there was no connectivity, there was no mobile apps. So the people were not wanted to afford like 1000 bucks for for a hovering machine for at home, they have to program, they have to care about etcetera. And they that the product went not well to the market and we stopped in that market. But then years later, the competitors came in when mobile apps were there, when the connectivity were there, the robot had the interaction of them, the products got successful.

So from the first generation to then a connected generation, there was a big step in technology which enabled to have advantages for the customers. And this is also what we, what we saw in the meantime, we have the next generation released. We started again in 2017. But there we see if you are too early to market or miss some specific thing, it could be difficult. In the meantime, there's a lot of um connected products and robots. We have um we have a robot mower for, for the outside we have the vacuum cleaner, but also we have robots we show also here in the professional field.

Ok, thank you. So the key takeaway of of that section is you need data, you need data to then analyze and base your your next uh generation of robotics and the development of of the robots um on on the previous experiences that the device made and to get that done, AWS offers a wide portfolio of like analytic services, machine learning services. But also you will see later, it will offer or offers the edge pieces of the solution. So it does not have to be running in the cloud all the time, but also locally.

Ok. Now talking about the foundations of cloud robotics. So what can you actually do with those services or with the use cases that you can think of? So typically our customers, they start with the ingestion of data. So getting the data into the cloud, gathering the data from the field and store it, analyze it from there. What what other customers are also doing is building a machine learning model so that it can operate on the edge. And um yeah, do the inference at the edge after that.

Um if you have already gathered more and more data, what you can then also start doing is set up a simulation and test your next generation of robot robots in that simulated environment. And once you have your next generation, for example, out there, you also need to operate and get the support to the robots in the field. So that's where the video streaming also comes in, for example, or the real time metrics for the robot in the field.

Now we have 212 other points that are in the end here, the software provisioning and also the fleet monitoring and culture will share another story

So that's what I mean by adding some color to the concepts. And so, in general, software provisioning over the over the air updates - that's one area where customers are looking to continuously roll out more and more features and continuously keep it secure, even when the robot is somewhere else and not right next to you where it can connect. And fleet monitoring - yeah, I'll, I'll let you take that away. You have a great example for that.

This is where we started in 2013. We started to connect our professional products - so the bigger scrubbers, sweepers - to connect it to the cloud, to collect data, to do software updates, which was a kind of a challenge I have to say. But also to provide insights to the customer about the status of their machine - how many machines they have, so many have more than one machine. They could do a fleet overview. They could see in which status are the machines - which is healthy, which is not healthy.

And we did it quite early so there was not, there was no AWS IoT available which we use today. But we still hosted the whole application, the whole IoT application on AWS and connected via 2G. Our machines with a telematics control unit to the cloud. And with that, we started to get into a connected environment and we learned also about our customers.

Okay, thank you. So the next overview is how, how you would get to the next generation, how would you use all of the different services in the field of AWS IoT? And how would you then also use the edge piece of the cloud?

So we start typically with connecting the devices, connecting the robots and getting the sensor data into the cloud. And that also happens not just in the cloud itself but locally on the devices. So that's what you see with that edge continuum there.

Then, next step, we already heard about that - managing your fleet, managing your devices out there, it's key. And there are ways to do that with, for example, AWS IoT Core.

And what you get then from the field from the robots is the data that you can use to learn and actuate. So learn and actuate means you can develop your next version, you can develop this behavior of the robot that would move and act in situations that are not yet there yet or that are not predictable. And you would enable behavior that goes beyond what you've seen before.

And all of that is what you always have at AWS - it's secured with, for example, encryption with mutual TLS, so certificate based authentication, and the configuration that you get.

Alright, so next up, this is going to get interesting. He handed out the safety vests, right? Everyone has a safety vest. Okay, no worries. It's not going to get too dangerous.

Let's get started. Hi, my name is Kiera Kiera B50. I'm looking for a home. Most people have already finished work. I clean big pools, narrow ones and corridors or irregularly shaped warehouses. I am always ready to go. I prefer to clean early in the morning or late at night fully autonomously. When I run out of energy, I recharge my batteries at the docking station and exchange the water. I am also very maneuverable and agile. I went over the dirty ground with my powerful rotor brushes. I must confess, I am a cleaning fanatic. I even clean right up to the wall. Nobody is perfect. Interested? Get to know me, the first fully autonomous scrubber dryer from Kia.

Some of you might already realize there's one of these products here. We will showcase it later. Before we get into that, I want to give you some insights on our journey on AWS and the cloud.

Basically, we started quite early in 2012. At least in Germany, there was no AWS region available during that time. And our issue was that we wanted to host our website more scalable, more globally scalable, and to provide ecommerce solutions to our customers. And we started to search how we could do it.

And AWS and the CDN kicked in and we started to try to work with AWS and it was really satisfying our needs for the ecommerce systems and the website.

And then in 2013, we started, as I already said, to connect all our professional machines to the cloud with the IoT gateway we had there. We had roughly 20,000 machines connected during the years to that platform. And this was also one of the cornerstones where more and more applications got into AWS over time.

In 2016, we had a global cloud first strategy for the company. So new applications were naturally built on the cloud if we developed it ourselves. But also commercial off the shelf applications were moved to the cloud or installed on the cloud if they were there, but not all of them because we still had a lot of on premise systems.

And in 2019, we decided to do a bigger migration of all our on premise systems to the cloud. There was also some hard deadline we had because we had to move out of our colocation. I have to say. So we migrated roughly 120 applications and roughly 200 terabytes of data in six months. And that was kind of interesting, I have to say, because as you see on the timeline, we started growing over the years in AWS.

So we started when there was no German region, we started only with one account to say it that way. And we were just growing in. But when we knew we will migrate our MES system, for example, or CRM whatever, now to AWS, we had to rethink and look how we do it. And that was also part of that short timeframe where we had a lot of workshops on how we do the architecture, how we adapt to it.

And for sure we did then an adoption of all our accounts to have a good segmentation between the accounts and follow the best practices. But still that was a very exciting project.

Now, we started this year to also migrate our SAP systems to AWS. That's going to be a longer lasting project. We have several big systems, a lot of servers, around 200 servers we have to migrate. So that's going to be a longer project. But still, it's interesting how we adapt with the systems from on premise to the cloud, and which kind of automation comes in.

And one milestone I don't have in here, but I would like to share - we will go live with our first factory in Vietnam next year without any local infrastructure like servers. We will directly deliver it out of the AWS cloud. So all systems which are needed to produce our high pressure washers will get their information out of the cloud, which is also a key milestone we will reach next year.

And this is also interesting - we entered in 2018 back in the world of connected products in the consumer field. There was a longer break. You saw that earlier, we started to connect our high pressure washers to the cloud. And many of you might now think, okay, but what's the use case behind? So what's the benefit if I'm in my garden and want to clean the floor or whatever, my terrace?

And the point behind this is that often people buy a high pressure washer, they unpack it at home, they are all very excited and they start cleaning with the highest pressure available. And if you have a terrace made out of wood, you break the wood, so it will get porous and breaks.

And with the connectivity with the mobile app, you can really tell - okay, I want to clean my terrace floor, it's made out of wood - and we directly send the right pressure to the device. So there's a configuration for the device that we have the right use cases or the right pressure for the customers.

Also before winter times, because most of our products come into service around winter times because people leave the water in and it gets cold and frozen and then things break, we push notifications to the customer that they think about the weather is going cold, be careful, there might be some issues if you don't care about your high pressure washer.

But also we have other connected systems in the consumer area. We have a water irrigation system where you can ask Alexa to take care about the watering of your flowers and so on.

So we connected a lot of consumer products to AWS and we completely rebuilt that former IoT solutions to a serverless architecture. So we used the AWS IoT gateway, data streams, and put it all to a complete serverless architecture to be able to scale because in the consumer field, different to our professional products, we have a high number of connected products.

In the meantime there's many 100,000s, like 800,000 connected high pressure washers we have. And we need infrastructure behind which scales and we don't have to take care about the operation of it on a 24 by 7 level.

And then we have the next evolution of the connected product and we will also show that it's really working autonomously. So hopefully when we start it, it will move around and not crash anything. I'm quite sure that will work out. It might take a second until it boots up.

But for the professional field, it was clear what's already working in the consumer area - that you buy a device, it starts cleaning at home autonomously - that has also to come to enterprise customers. And for us that was also part of our story to renew, to renew how we built and invent products.

And we were working on the new robot. The interesting thing here is there's more than 10 cameras and 2 L sensors. So there's a lot of electronics and even more IT. So there's an Intel CPU inside that device. The display is not a usual Qt interface but a web browser running. So it's a usual web interface. And this also shows how such a product changes the culture more or less of the company because before, openly saying we had products with few meters of cable, not a lot of electronics, mostly mechanic products.

And here we have more than one mile of cable in that product.

Um we have, as I said, a lot of compute power in there and it's always connected. So there's a completely different approach to develop such a product. It's sensor based.

Um there's safety reasons we have to take care of and um the whole story doesn't end when we ship it to the customer, but um we keep on uh connecting with the product all the time.

Um this is, this is interesting, you see, it starts moving, it's not loud now because on the, on the carpet, um that's going to be difficult to, to, to let it clean. Um, it's now moving around, um just autonomously and this is also what customers are using it for. So usually it's really in bigger areas like production facility but also airports, grocery stores where people are around, it's not a problem. So it stops when someone is getting in front of it.

And this is also some of the key features we have well like um connecting cables in the device, electronic cables, sorry. Um so the question was what, what is meant with more than one mile of of cabling, but it's like um cable harness in the, in the in the machine itself um to connect the sensors and so on.

Um so there's really a lot of electronics in that, in that machine. Um so for us, it was definitely as a company, a complete change to develop that product. We had to think from the customer backwards to what the product needs, which regulatory or laws are needed to fulfill an autonomous cleaning robot. Like we have it here to let it drive around. But also we had to take the challenge to think of how to get the right technology, the right sensors into the product, how we get the connectivity, how we can ensure service, for example.

And this is this is also 11 example i can share in the meantime, if there's an issue with the machine, we can log into the machine, we can see what's on the display. We can really do proactive support to the customer and help and we have a support team supporting our customers proactively with that.

But also one of the challenges were the firmware updates because you have for sure. Um you have to take care that if you are pushing more and more firmware updates, if you have regular dev ops processes, then with the products that these processes are stable to say that this is also where we started to rethink how we do the firmware updates to our machines, how we ship it.

Um and we use aws appsync for, for that one. So um we have the, the robot to have always um the the latest firmware on it. And um the customer can see on the dashboard, which status is there? Is there issues? Um with the firmware also you can see on the dashboard geo fences.

Um but there's clearly um one of the topics we have because without the automated build process, you will end up at some stage to have broken devices at the customer, the customer will not be satisfied and the robot, this is also one thing you can imagine it's not cheap. So there's, it needs to be reliable, it needs to be reliable during day and night time. And as it goes autonomously, um, the customer asks us to have the highest availability to the product to give you a feeling about what the device is, cleaning or what the devices are cleaning. It's in kilometers, it's 24,000 kilometers. We cleaned this year already and a wide range of square meters.

And also this is something where we have big opportunities with the new technology, not only with the autonomous robot going around, but for our business model, this might get bigger changes or a bigger opportunity to change because we could now charge for clean square meters instead of selling a machine. Um where's opportunities are also doing rental business, for example.

So we have always insights about how the customer is using the machine, um what's necessary um but also how we further develop the machine. And this is what we found out during the years when we connected our products. And we gathered the data that we could clearly see that. For example, customers were not using our products as long as we thought all the time.

So we were able to resize the batteries or we could change the battery lifetime. Also, for the kira, for example, the usage of the battery is completely different between a production facility and a grocery store because in a grocery store people are moving around. So we have, we have to adapt and we have to see where the charging stations are and this is data we get we hadn't had before and we have to take that insights to bring it back to the customer in the end.

And um what's around the corner? I think there's, there's much more to come to say it that way. Um so with the data, we collect, we will improve machines, we will bring new features to the machines. Um but we will also be always at the customer site um with seeing what's going on there.

Um we will bring more robots in the professional field also to the customer, for example, like vacuum cleaners, bigger ones and we expect that the robots will start talking to each other by interfaces for sure. Um so that's uh vacuum cleaner, for example, is first vacuum cleaning something and then the kira is going on and doing the wet cleaning, for example. And then we have more intelligence, but also for the kira, we have already examples where the kira interconnects with facilities and with standards and facilities.

So if we have bigger areas, usually there's fire doors to shut down when there's a fire in a building and we are already able to control fire doors if there's enough electronics. So the device gives them a signal to the to the building control system to open the fire door. Go through th the cleaning job comes back, fire door closes again.

And also here we expect that there will be much more interconnection with building systems or even other products in the future. And this is only possible if you have in the background, a scalable infrastructure um open to connect with others um and bring new features and inventions.

All right, thank you, lena. So, um, i guess we, this is called wrap up, but it's more like an open discussion for, for more points because um when we first sat down together, thought about what we want to tell you today. My first idea is let's talk about connectivity. And luckily kata already has a very strong partner also, which is uh the the aws premier partner, zoe and they also work together on the connectivity part and a lot of other parts.

So um going beyond the connectivity is what was really interesting to find out and you might have heard some of the points already here and there, but i would like to take the chance and um yeah, talk to you more about, for example, the first one. So component, right? Sizing. Um how did you find out about the battery sizes? For example, mr stallone? I hope so.

So basically, we gather the data um over many years i already shared that we started in 2013 to to gather data from the machines about their usage. Um and when we had some years of data we brought it to our, our data analyst team and they, they were looking into their data and try to find out how our machines used, how our usual running times. And in the past before having the data, we sold the machine, we heard from the customer when there was an issue or we wanted to buy a new machine or needed spare parts. But here, then we had insights about the usage. It was new for us and it also took a while to have enough data. This is also something which is important in the in the professional area. We don't sell like millions of kira. Sadly, if you want one, you can have one, but we don't sell millions of them for sure not. And then it takes a while until you have enough data to do some analysis and to prove we can we can improve on, on batteries or on other other things we have.

Great then um the predictive maintenance part, i think that's also an interesting one because i i mentioned machine learning a lot, right? You heard you heard that previously, but you can, you also had a, a great example for an application of machine learning there.

Yeah, basically um with a with a simple noise sensor, we were able to, to learn when a brush is over more or less. Um so the noise sensor really, really points out a different curve of a acoustics. And with that, we were able to see, ok, we could ship the customer in advance um a new brush um to, to really show that it might be soon time to have a new brush there. But also for other predictive maintenance topics, it could help. And also with a simple sound sensor, you could find out about the usage of the machine because you hear when it's running. So also this one is quite easy if you have the sensor in um you can learn from the data as well. Despite that, it's not like um sensor data or whatever, but acoustic data. Also this is possibilities. We see and we can transmit, it's not like heaps of data like videos or so

Great um firmware updates. So we've heard a lot about firmware updates already before. Um it it is a crucial part. Um but i also found out firmware updates on like the application layer is rather simple with like um services like o green grass. But if you want to go further down down the level to the, the, the, the image itself or the os, it gets a bit trickier. And yeah, yeah, definitely. That was the updates was a challenge from the first day when we started to connect the products. And as i mentioned before, we don't want to break it. It was the highest priority. So for sure, we we checked if we have like two partitions then on the on the device and how we ensure that the firmware update is getting down. But especially now on the robot like that with a lot of compute power and with normal computers, a linux running on that device, um you sometimes want to just update a kernel, for example. And there we still struggle to say it that way because there's no service. Usually it stops with green grass to get it easily done. This is where we also had to, to bring it and electronic knowledge together so that there's things merging in the in the capabilities to be able to do that and also processes like de ops, which is um i would say that was not there 10 years ago in the electronics, but it was already there in the it. And now one could learn from the other and build with that stable and robust products. But in the end, i have to say firmware updates, it's i mean, from the consumer side, we are used to it and also in the in the professional side in the enterprise world, it's expected to have that. So that's a, that's a crucial part of the product. So it's not only to talk about the product, but it's also to talk about the software and how you do the maintenance, how do you improve over years. So it's not stopping with selling the products anymore. That's really a big changer, cultural change also in our company.

Yeah, i think that's a great segue to the next one. So you produce a product, you produce a robot and you sell it to your customers and it's out in the field, but you have no visibility into what's happening there actually. So one of the things that you also looked into was the the after point of sale visibility.

Yes, this is where also we had to adopt, adopt our organization. I mentioned the example with the service. So you have to build a complete new service architecture to be able to react within hours or minutes instead of someone sending out. So there's a complete different approach to, to sell such a device because usually you sell service contract with the with the device to to have a fix or replacement within days or next day. Um here they want to have if the machine breaks, they want to have someone logging on to machine checking what's going on um and so on. So basically there's there's really a change also in the organization to to be able and to be yeah, to, to get the workforce to understand the whole cue. So it's not only the product, it's a cloud, it's a connectivity and they have to find out where's the issue, how can i help and so on. So it's a complex environment in the meantime, not a simple product with a few meters of electronic in it or cabling in it. And this is i would say for a company like we are manufacturer um since, since years, one of the challenges we face all the time, but also we, we solve and we find ways with our teams to really get it working and get it run also on a global scale.

Ok. All right. Um at this point, i would say we, we can take further questions also next to kira outside there. Um and i would like to, first of all, thank carter and also leonard and the team to, to make that happen to bring their robot to re invent. And also thank you for your attention. And um as i said, we can also continue with further discussions.