Cloud OnAir: Google Cloud IOT from devices to cloud – enabling business outcomes in a secure way

[MUSIC PLAYING] PUSHKAR SHARMA: Welcome to Cloud on Air There are live webinars from Google Cloud every Tuesday My name is Pushkar Sharma I’m a product manager in Google Cloud IoT Today we’ll be talking about Cloud IoT, and talking about the business use case, and how it applies to IoT all the way from Edge to the cloud You can ask questions anytime There are Googlers on standby And we’ll be taking those questions at the end of the presentation But let’s get started So today, first we’ll start off with measuring the business outcomes with IoT Then we’ll talk about the Cloud IoT, in general, how we see it Then we’ll talk about our new offerings, Cloud IoT Edge, and Edge TPU And then we’ll move on to talk about security and Cloud IoT provisioning So let’s talk about business outcomes with IoT So there was a survey done by McKinsey And it turns out that 84% of the companies were stuck in pilot phase for over a year, and at 28% for over two years And about 30% of these pilots only could make it to the scale So about 70% of these pilots never made it to the production So what’s really going on? I mean, we call this as a pilot purgatory So essentially what’s going on is you started with pilots, and then you never have success with it So what are the key success metrics which we found? It comes down to really the business alignment Are you aligned with your senior executives? Do you have support? Is there a strong business case or vision and how it will impact the business? Or is there internal alignment in how do we execute, and a long term perspective on that? So really, it comes down to an ROI If you can establish ROI, you can easily get these alignments And that’s where you can move from pilot to the next stage So what we focus on here is the why and not the how Because many times, you start off with the technology You’re proving out POC, whether the technology works And it starts off as a very grassroots thing, but you want to really try to go down to the making a business impact And that’s how you’re going to move from a pilot to the full scale So when we think about business outcomes and how you’re going to really start thinking about, it all starts with digitization And what is really digitization? It’s really a lot of things which are analog today, they’re your physical blind spot These are things which are not connected Or even if there is data, they are in silos So what you really want to do is you want to connect these things– customers, staff, processes– in a unified manner, in a unified system So now you have all this data coming together in a single platform And from here on, now you can start to analyze and start creating these insights And this is your big data analytics On your past data, historical data, you can start to see what things were going on, what was the process, how things were happening in your business environment But then you want to also start to look at, can you create prediction models? Can you use machine learning and advanced artificial intelligence to create predictive models for your business, which is really indicative of the key performance indicators? And then you want to use all of this information, which is the real time information coming from the business, the past behavior, past data and insights, the predictive model, the what’s going to happen in the future, to really start to think about workflows And that’s very important, because there are workflows happening today But they are not connected They are in isolation You want to think about, how can I take that process, which is in physical form happening today, can we make it more digitized? Can we make it all connected in a seamless way? And that’s where you can start to start optimizing and get benefit from IoT So when we look at IoT ROI, so it really comes down to that you want to be measuring ROI, Return on Investment So there is a direct ROI, which is essentially revenues and costs So you want to increase your revenues Can you sell more?

Or can you reduce your cost? Are there things you can do to reduce it? So that’s direct And there are things you can do using IoT because, with IoT, you have more data, more insights So you can enable features, enable sales, service enablement You can do improved pricing, measured demand and elasticity You can also look at the costs and can we improve efficiencies, uptime, reduce the risk, and improve quality? But then there are other benefits of IoT which you can expand to For example, marketing And this is where you want to know your end users better Can you use the data to understand their behavior? If they are connected through IoT, you actually can And you can then upsell and cross-sell products And you can do personalization based on that customer You can also build better products because now you understand the customers You understand their behaviors And in general, you have a better relationship with the customer They connect with you more You have a better brand On the engagement side– so when it comes to service delivery especially, you want to have that deep customer knowledge of how they use their product Can we proactively reach out to them? Can we help them? Can we make their experience very delightful? And also, with all the data which is coming through, you actually start to understand a lot of opportunities which were in your blind spot And you start to see how can you use that data to mine new businesses and create new business outcomes which were not possible before So let’s go into a few examples of our different verticals Hopefully, this will resonate with you and give you an idea of how you want to start to think about different use cases in IoT So for example, manufacturing So if you think about manufacturing, what’s really important? I mean, the important thing in manufacturing is productivity Factories are operational 24 by 7 Can we keep the uptime? Can it be more efficient? Can it be more safe? Because many of these manufacturing heavy industries inherently have challenges there, because it’s a live environment and a very high temperature and extreme environment there So you could use predictive maintenance We’ll talk about some of that later on But essentially use artificial intelligence and machine learning to create predictive models for your machines So now you have a better understanding of when these machines is going to fail so you can take action up front You can also understand what are the factors which are driving the failures so you can work on that and extend the life of your machine So essentially, you’re not only improving the asset utilization, but you’re reducing the downtime The second aspect of manufacturing is the process There could be inventory supply chain or different processes For example, in a car, it goes through paint It goes through a retrofitting, testing And you want to understand how much time it’s taking between each process You want to optimize for that And if there is an inventory which has excess piling up, you want to reduce that So these are all different processes you can understand by using IoT and getting real time insight of what’s going on And then not only in the factory, but also especially in just-in-time manufacturing, it goes beyond the factory It’s also connected to your vendors and upstream and downstream supply chains And you want to get a very good sense of how the supply chain is functioning Is there going to be any shortages of components that may impact your production and also, maybe, customer experience in terms of delivery of these products? And in general, you want to understand how your employees are doing, how your machines are doing, and can you improve your productivity and reduce your failure rates and have a high ROI for your customers? There are many partners, like SpringML, Relayr, Tellmeplus, Clear Object These are different partners which are actually using Google Cloud Platform to deliver that And clearly, if you have use cases around that, you can also reach out to us And we can help on that too Likewise, on oil and gas, it’s slightly different The same extreme use cases, but here the business outcomes are really around, first of all, can I get better resource mapping? Can I understand where the next set of oil is going to be? Can I use machine learning, and can I use satellite image data to get that understanding? Also I want to optimize the machines,

and their failure is important, and where different assets are, the trucks are, the real machines And even employees are your assets And this is where you want to understand, are they safe? Are they in a safe environment? If there is going to be a problem, you want to predict that And if there is a problem, you want to evacuate And the safety is very important So these are the key use cases in oil and gas There are partners like Cisco, Losant, again, Relayr, FogHorn They operate within the whole IoT architecture Some are more in Edge, some are in Cloud But there are a lot of partners Google has which can help in these use cases Let’s talk about logistics Primarily, when we say smart transportation, it’s really about logistics And that could be a vertical by itself, but it also feeds– we talked about manufacturing And even in retail, it feeds the rest of the industries, as well And here, the asset tracking, clearly– like these trucks need to be tracked So asset tracking is important But many times, there is also the product on those trucks, which is important And you want to understand, for example, if it’s a cold storage, you want to understand how the temperature conditions are If you’re transporting something which needs to be maintained at a certain temperature, then the telematics of these vehicles, how fast they’re going, are they following the route, did they deviate from their route, needs to be notified in real time so you can take action on that And then also, just like any asset– we talked about predictive maintenance– you want to understand the predictions of when these assets are going to fail Can we do something to enhance their useful life? What are the key failure modes for these assets? So let’s talk about health care So we’ll talk about health care, and pharma, and medtech They’re two slightly different, but very similar use cases So health care, this is about, for example, hospitals This is hospitals or extended day cares where you’re dealing with patients And really, what the main goal for a lot of these hospitals and in the health care industry is really to serve patients And the patient experience is the most important thing Their health is the most important thing So you’re using IoT to, first of all, A, measure the assets So there are lots of assets, like where the wheelchairs are, where the beds are, where the IV pump is So when there is a need for it, the nurses don’t have to go look for it And they can address the patient as fast as possible Also what happens is, because there’s so much unpredictability around these assets, you end up buying more than you need And it increases your costs And these are expensive equipment And you end up having a lot more of inventory And that adds to your cost, overall And then we know about this connected health, use cases around monitoring of key vitals of the patients while they are at the facility, like mental health patients, or infants, et cetera, where you want to make sure either they don’t leave the facility, and the key vitals are there If they fall, you want to be able to detect fall detection and take appropriate action as quickly as possible Then let’s go into retail There are lots of use cases in retail We’re already familiar with many of them And we can all associate with it So one of the key ones, especially for the business, comes down to better engagement with the customer Because for retail, the experience in the store is the most important thing And can you use the technologies, like beacons, cameras, and geolocation based notifications to create these experiences, and location based notifications and marketing on-site You may want to also use image detection and video analysis to look at footfall Or you can use RFIDs for that and understand what parts of the store are high traffic areas And can they be used to improve your promotions and product placement, and in general the way you operate your business? Automated checkouts are also very important, because it’s part of the customer experience And can you do more with IoT with understanding who the customer is, what products they have, what account is associated with to really improve that analysis? And then there are a lot of use cases, not just around customer experience, but also the back store operations And a lot of that is about the best customer experience,

but also about inventory Do you have the right amount of inventory at hand? Is there a lack of inventory? Can you replenish quickly so, when customers come looking for something, it doesn’t run out? And then can you also have a better understanding of how the products are moving on your shelf? What are the fast mover products? What are slow movers? Can you optimize a placement on the shelf? Let’s talk about pharma and medtech So this is about pharmaceutical companies and medical technology companies who are in the business of helping customers from a medication standpoint And their main goal is to ensure that the customers or the patients are taking medicines appropriately And it’s not only good for the customers, but also good for the business where they can improve the adherence of the medicines That way, they can improve their revenues But also having a better patient engagement not only improves their brand, but it also allows them to do automated interventions So if a patient is not taking medicines, they can notify them Or if it’s a connected insulin meter where you could measure the insulin levels and inject appropriate amount of insulin to the patient, that can be done automatedly as well And then there’s a big part about pharma and medical is that they spend a lot of time in the FDA process, in the regulatory process And they have to go through these clinical trials, which is very expensive And oftentimes, a lot of these clinical trials fail because they just don’t have enough sample points But with IoT, you can actually improve and accelerate that Because now, you can start to get a lot of the data live And that will improve the way you go to market And you can launch your products faster OK So let’s talk about Google Cloud IoT What do we mean by that? So when we look at the Google Cloud IoT Platform, you want to think from both devices all the way to the Cloud So on the very extreme is the devices And these are the devices, which could be Edge device, for example And we have a product that we’ll talk about shortly is the Cloud IoT Edge, which allows you to do three things One is it allows you to connect to the Google Cloud This is a product called Edge IoT core, which does it securely, and then makes sure the connection is strong and easy for you to develop as a product And second is the Edge ML, which is how can you do inference on the device and machine learning on the device and create those outcomes without relying on the cloud? We do support Android Things This is a Google OS for IoT devices, which is hardened and built specifically for IoT for the best security But we also support other operating systems, like embedded OSes and Linux OS as well And finally, these are the CPUs or the processor units So there’s CPU, or the general purpose or MCUs can be GPUs for advanced machine learning capabilities And then we have a dedicated product we just recently launched called Edge TPU, which is for machine learning inferences on the chip itself And it’s a hardware accelerator We’ll talk about it shortly, too And on the right side is the Cloud IoT Core, which is the entrance point for all the IoT data into the cloud And this is basically your gateway to inside the cloud for the rest of the products It basically pumps all this data into Cloud Pub/Sub, which is essentially your message queue And from thereon, different products within Cloud IoT can actually take up this data For example, you can use Cloud Functions to create your own applications You can also use Cloud Data Flow to do ETL operations and transform this data and store it, for example, in BigQuery or BigTable And then use machine learning to get this data from these databases, create machine learning models, and then do either inference on the cloud, or you can send these models back to the Edge ML to create these inferences at the Edge, as well And then you obviously have the visualization layer on top of that That’s the last thing You could use Cloud Data Lab, Data Studio, or maybe build your own insights on top of the Cloud platform to create those visualization charts

and reporting, et cetera So that sort of, hopefully, gives you an idea We also have a Google Maps platform, which is also a Google product, which allows you to get better location updates And you can feed that information back into the Google Cloud Platform to create, for example, asset tracking use cases on top of that So let’s talk about why Google Cloud IoT, and what’s so differentiated about it? So first of all, Android Things, we have a unified end-to-end integration where seamlessly it connects with Android Things and other OS and connects to the cloud So it makes the development of the hardware and IoT devices easy It’s a global IoT service It’s purely a serverless service where you don’t have to maintain different hubs You simply spin up your devices You only care about device’s data coming through You don’t have to worry about scale And it has intelligence built-in So it applies machine learning and AI in the cloud It allows you to create IoT which is not simply basic use cases, for example, temperature monitoring, but also create prediction models around temperature As I mentioned, it’s serverless by design And this is true, just like BigQuery and many of these products So it basically is a global network backbone and a front end which is essentially powering the billion-plus user applications from Google And pretty much all Google applications are built on the Google Cloud Platform And then using the same network backbone– so this is what is powering the Google Cloud IoT And you can take the advantage of that network for your application and can build applications with the same level of reliability and scale So let’s talk about Cloud IoT Edge So if you step back, there is the Edge computing, which is along back when there was no Cloud It was all on-prem But there has been a lot of advantages to the cloud And people have been really migrating to the Cloud But there are some use cases where there is a little bit of Edge which is still there And that’s important from various use cases For example, it could be security It could be the fact is that you don’t want some data It’s mostly part of policy that you don’t want some part of your data to leave on-premises to the cloud So can you do processing on the Edge? There could be use cases, like robotics, where the latency is very important So you don’t want the round trip delay times to the cloud So you may want to take certain actions quickly, and hence, you want to do Edge computing Likewise, for predictive maintenance, we talked about a use case about machine learning inference You may actually be doing Cloud learning, ML learning in the cloud, but you maybe want to do the inference in the device itself Likewise, object recognition– so here if you want to be taking quick– for example, it’s a manufacturing line, and you want to be testing for defects And in order for you to have the maximum yield and the high throughput, you’re going to be doing the fast object recognition And again, it’s an inference on-prem on the device itself, and that’s something you want to do quickly Likewise, there are use cases– if you’re all familiar with smart vehicles– warehousing where there are AGVs And here, also, you want to be taking actions quickly Or maybe the WAN connectivity is not that reliable, and hence, you want to do Edge computing So what is Cloud IoT Edge? So as we talked about, it extends the cloud data processing down to the Edge It runs on Android Things and other operating systems It connects to the cloud using the Edge IoT core It stores, processes, filters, derives intelligence And then you can run inferences for the TensorFlow Lite models, locally And it allows you to do next-gen of machine learning at the Edge And this is how it really works You would start with the building and the training of the ML model in the cloud You aggregate the data coming from the devices You aggregate the data You use a machine learning model to train it, and then you send it back Essentially, what you would do is you would convert that machine learning model from TensorFlow to TensorFlow Lite And then you would compile it for the given target If it is the GPU, or general purpose CPU,

or if it is the Edge TPU, which I will talk next, you would want to compile it for that platform And then you would install it, essentially send it back to the device And you will start to do inference on the device itself So we talked about Edge TPU several times So what really is it? So it is a very tiny, high performance, inference chip set, which is not a general purpose CPU It is custom built for machine learning It can do fast iterations And it’s the first phase is to build for video and vision, because these are very high CPU-intensive And it can run concurrently at 30-plus frames per second This allows you to create high performance machine learning at the edge, with the low power consumption, lower footprint And also, because it works seamlessly with the cloud, it’s an end-to-end implementation, not isolated where you have to put it all together yourself So you basically use Cloud IoT core and Google Cloud for machine learning, and send these machine learning models back to the Edge TPU And you can do inferences And we are going to provide you with the tools and capabilities, so it’s very easy to use and deploy We’ll also be providing you with the Edge TPU development kit It will be coming in shortly Essentially, it would have an Edge TPU It would have an associated CPU, as well It would have a secure element We’ll talk about it shortly It’ll all be integrated on a system It’s called SOM, a System On Chip module, which you could use to create applications and develop these application before you go into the full production And this SOM would actually sit on, for example, Raspberry Pi And so that way, you could reuse the development boards you already have in an environment you have and just put in this new SOM, which is already secure It connects seamlessly to the cloud It has security built-in And that will allow you to improve and increase your go-to-market We will also have a USB connected Edge TPU These are the AIY boards which you can connect to your laptop or development environment And then you can do machine learning on these, as well OK So let’s talk about Cloud IoT core and security So Cloud IoT core I mentioned is the entrance point for all the IoT data But what really is it? So basically, there are two parts to it And one is the protocol bridge That’s the connectivity part And that’s essentially your connectivity through MQTT and HTTPS protocols It does automatic load balances, takes the data up from the devices and puts it into the Pub/Sub The second aspect to it is this entire device shadow, this [INAUDIBLE],, like what those devices are and how they are managed So essentially, all these devices are added into projects and registries And for each device, there is a device ID and additional attributes which are required for a connection– so for example, with a public key or a certificate And then there will be additional metadata and state information about the device, its health, and business attributes of these devices, but also the telemetry data which is coming through the devices For example, temperature, humidity, motion, and all these different sensors which are on the devices would be on that So one of the very important things about these devices and IoT in general is security And we formally, Google has a very strong opinion on security And we have been pushing the boundaries on that So the basic idea is that we use essentially asymmetric keys So private key is built into the device And the public key of that is in the cloud And the way it works is that the device, it’s over TLS, but the way it works is through JWT tokens So the device will create a JWT token and sign it using the private key And this private key is then sent over to the Cloud And that signature– the private key is not sent, the signed JWT is actually signed And the JWT signature is verified by the cloud

And that’s how the connection is established So there are problems with that One is that you have to make sure that private key is safe and secure And what you would end up doing is, and that’s what we recommend, is using secure element from microchip or NXB, which has a private key built into it at the manufacturing time And then you use this private key Obviously, you cannot use a private directly, so you would end up having the public key equal And you get that from the manufacturer And you would put that public key onto the cloud And when the device boots up, it does the same thing It does the JWT creation, signs it, and associates with the cloud Now, this is a very cumbersome process where you have to write a script and have these public keys sent to the Device Manager So what we have come up with and tried to solve for is with the Cloud IoT Provisioning This is a new service which basically allows you to manage this entire process easily So to recap, really, there are two problems, right? So one is this key manager You have to secure these keys You have to transfer these public keys from the SE vendor to the OEM And then there is the device management part of it is that, OK, now you have to make sure that these public keys and device IDs are preregistered onto the cloud And the same device ID and the target information, which project ID and registry ID needs to be actually on the firmware So that means that you have to pay a lot of attention It becomes a complex process So what we’re trying to do here is we’re working with silicon vendors, select silicon vendors to get these public keys from them which are stored You will simply use Reel ID or UDID, or additional information for security sake To get those and claim those devices, you will simply easily create your targets, which is simply your registries where you want to put those devices in And when the device wakes up, it simply asks for where it needs to go, and it goes there So this is exactly what’s happening in this slide As I said, it’s the same exact slide as the previous one, but the part here is that there is a cloud IoT Provisioning service which gets the public keys from the silicon vendors And it allows you to claim them, and it allows you to identify these devices and provide the related configuration and also provide information where these devices need to go It adds the target on to the Device Manager in the cloud And so when the device wakes up it– and all these devices are very generic You don’t have to worry about these devices They are very generic devices They go and basically say, OK, I’m X, Y, Z. Where am I supposed to go? It gets the new config, which tells it where to go And this config could be as elaborate as you want It could be a new firmware It could be any things you want this device to do And then it comes back as a new device and connects to Google Cloud And that’s how it gets authenticated and starts serving So let me do a quick demo of this service And this is available for early access, but let me give you a quick snapshot For example, when I log in as a customer, I’m going to be looking at a blank screen, obviously, because there are no elements I have made So I’m going to go in here In this use case I’m using Reel IDs So I’m going to have a Reel ID Reels are essentially all the secure elements are chips They have wound up on a tape And it’s wound up in a reel And it’s a package It’s essentially the package ID So I’m going to just put a package ID here, and I’m going to just claim it So it’s going to basically go through these reels And these reels now have shown up And these reels have been claimed, but they have not been targeted So now, I’m going to be targeting into a registry So I have a registry open here And you can see this is all empty right now, so I’m going to be putting these devices in this registry And there is a registry– -1 is a registry in this project ID And that’s exactly what I’m going to be doing here So I’m going to target it to this particular test target here And I’m going to assign it So essentially what it’s going to do is move these devices from this place to the registry So basically what it’s doing right now is it’s going through these in the loop, checking for these devices And these devices are now actually moving

into the registry So when I’m going to try to refresh here, I should be able to see shortly The devices should show up But if you want to, this is available as an early access And you can sign up for this early access by clicking onto this link here And then we will get back to you We want to understand your use cases, as well And then we’ll provide you an early access So hopefully, it would have come up by now OK So now, these devices have come up And you can see all these devices which were empty before And then it’s very easy for us to move from these devices And here, these devices have been empty So next time you have a next batch of devices, you come in and do the exact same process So you don’t have to worry about where the devices went These all devices are very generic When they come up, they get their configuration, and then they connect With that, we’re going to end the session here This is going to be our last slide And then stay tuned for Q&A. And we’ll be right back in a few minutes– maybe less than a minute And then we’ll take up the questions which you may have So feel free to give us any questions, any feedback And then we’ll be right back with you Thank you OK Welcome back So it looks like we have three questions here So let’s get going with the questions So first question is, in terms of Cloud IoT Provisioning, I don’t have secure element right now, but plan to use it in future products Can I still use this service? Yes So clearly, we are trying to solve for the secure element use cases, but that doesn’t mean that’s the only use cases we wanted to solve for So as we go through this process, we will be solving for the non-secure element use cases, because there are ways for you to create public, private key pairs today And we want to be able to solve for the part where you’re confident about the private key, but the public key still needs to be connected, and put that into the cloud repository And we want to make that process easy So yes That’s the short answer So let’s see The next question is, do I have to use Edge TPU to take advantage of Google’s Edge processing? OK So yeah, no I mean, clearly, Edge TPU gives you that advantage for the use cases where you need heavy processing But there are a lot of Edge use cases where maybe general purpose CPUs is perfectly fine Because even on the ML and AI, there’s a huge range of computational requirements So no We wouldn’t support GPUs or support general purpose CPUs And whatever your model is which you have created, it can be used through our products to send back those models from Cloud IoT to Cloud IoT Edge And then you can use these models for inference in the Edge And then, if you’re going to use TPU, then we will support that too So Edge TPU’s not the only thing we’re supporting here Next question we have is, can we configure Cloud IoT Edge, plus Android Things, plus Edge TPU on the currently available Raspberry Pi? Yes So as I mentioned, it is a SOM, which is essentially going to have all these things in one SOM And yes, you could use the Raspberry Pi with that SOM But when that SOM is going to be available? Sometime in the upcoming months it will be available So you want to sign up for that early access So let us know We want to work with you And so make sure that you are one of the first customers

But yes You could use any general purpose Raspberry Pi And that’s the whole idea of that SOM It just goes on top of it And you can start off with your projects right now without the Edge TPU And then when that’s available, you can make that Edge TPU capable OK Well, thanks for all these questions And thanks for your time Stay tuned for the next session It’s CE Chat, Google Cloud Networking 102 It’s cloud routing and VPC pairing Thank you again for your time And feel free to go to the Cloud IoT core I mentioned there was this– let me see– go to Cloud IoT core website You can sign up for the Cloud IoT Provisioning early access And also, you can sign up for the Edge TPU and Cloud IoT Edge early access, as well, on our website OK Well, thanks again And have a good day Bye [MUSIC PLAYING]