Amazon re:MARS 2022 - Robots and assistive technologies (ROB317)

Amazon re:MARS 2022 - Robots and assistive technologies (ROB317)

Show Video

- Well, thank you for joining us here today. This is fantastic to have this great audience. Ha, ha, we're gonna have a great session. I'm honored to be here today with Sebastian Arias from Cosmos Robotics. We'll get to talk about robots in assisted living environments.

In skilled nursing facilities, empowering humans with disabilities. I had the honor to visit a skilled nursing facility with Sebastian a few weeks back in Atlanta. At Presbyterian Village. That was a fantastic experience. It gave me a very good hands on and first hand account for how humans are using robots in day to day operations that you normally wouldn't think of.

But as our population ages, I think these robots are gonna become more common and more useful over time. So before we get started, I would like to show you a video that kind of gets things going and exciting about the great uses of robots in the world today. Let's take a look. (light music) - [Armstrong] That's one small step for man, one giant leap for mankind. (dramatic music) - Well, hopefully that video got you all pumped up and excited about the future of robotics. But let's take a look at what traditionally building robots has been for most robotics developers.

You know, most robots in assembly lines are systematically programed to do very repetitive tasks. They're usually caged. There is no real interaction or human-machine interfaces. But with the power of the cloud we're now able to create interfaces that are much more dynamic.

We're able to create advanced human-robot collaborations. We're able to create natural language interfaces for the future of robots. And we're able to validate and test autonomy at scales that has never been possible in the past. So simulation brings a capability to robotics development that hasn't been traditionally leveraged.

So we look at some of the large trends that have made this possible in the industry. Most certainly the advent of open source software, the robot operating system has created an ecosystem of tools, software developers and companies that have sparked the current robotics revolution. There's about 3,000 robotics startups worldwide. This is just analogous as the advent of the PC, the internet in the 90s, mobile phones, and the cloud. So we're gonna see in the next 10 to 15 years a very exciting future of an industry that is just ripe to explode and at the beginning of it. So, being able to do robots as a service where you just lease the robot and get it working on your operation.

Also has increased the adoption and gets people testing robots sooner than they would make a capex investment, for example. With IoT and with the capabilities that the cloud provides, you're able to connect all your machines and tools across and have a network of connected products sharing data, communicating. And once you have that data, once you have those pipelines in place you can use artificial intelligence and machine learning to gather insight from the data, perform better operations and improve over time to online learning on your data and on your tools and machines. So the cloud has become a very powerful enabler for the world of robotics, that's indisputable.

And we're gonna show you some examples on what that looks like. Before we jump right in there, let's take a look at how AWS empowers the next generation of robot developers by providing a full, seamless stack. Where before robots had a very compute constrained environment, now you can leverage powerful compute in the cloud with GPUs, FPGAs. You can do hybrid architectures where you can split your computational graph of your robot and run things that you don't need to run at high speed.

You can split your graph, run things in the cloud, run things on your robot, and you have a distributed computational graph of your robot. So at the bottom part of the layer you have the robots, the fleets of machines. And now you're able to, whereas before everything was pre-programmed, local operations, you couldn't make many changes. Now you can increase your autonomy and integrate our out of the box AI/ML artificial intelligence and machine learning APIs and capabilities so you can integrate those.

You can also do offline. And perform things, perform computation at the edge. Because you know, you have the data and the platform and you're able to ingest that data into the analytics services.

And you're able to push that. As where before, everything was very tightly coupled and monolithic, so. This gives you the capabilities to have operational dashboards and do live monitoring. And then that brings the capabilities to having a digital twin and a real time digital analytics of your environment that also allows you to have large scale simulations. And integrate into your development process with continuation integration and continuous delivery.

So, you know, after we sort of have these capabilities and this technology what we really see by learning from most of our customers is that, most robotic startups have the same set of challenges. Robotic startups can be capital intense. It requires a large investment in sensors, actuators, mechatronics, GPU hardware, etc. There's a large skill gap. There's not enough roboticists in the world. So if you're looking to make a career change and go from mobile to any other type of development, robotics is the right place for that.

So traditionally the first generation companies in robotics have not been able to, have had trouble specifying, you know, validating the business use case and the specific solutions and value they provide to their customers. So finding a problem that has the right unit economics in robotics has traditionally been challenging for most of these companies. And then once you actually figure out what type of robot you wanna build and you've actually found a business case for it, deploying and operating a large fleet of robots can be challenging and complicated. So that is where AWS brings a significant amount of capabilities that makes it possible, right? You can leverage the agility of the AWS cloud where your teams can experiment, develop, test quickly, and deploy. You get the benefit of our SaaS model where you only pay for what you use at any given time.

So you have significantly less upfront investment. You get the elasticity of the cloud once you have things locked down and you're ready to deploy. You can scale and deploy worldwide in any region that we operate. You can innovate faster and focus on your business value and get to market as soon as possible, so. Let's dive right in into some of the topics that we'll touch today in assistive technology in robotics. Something that we've done for our partners is we've created environments during the pandemic, during the COVID-19 pandemic, it was very hard for our customers to test their robots in the real world.

So we went ahead and created a simulated hospital world. And many customers got the benefit of it, and got to quickly test, develop, and deploy their robots in simulations. So, Sebastian will tell us more about how Cosmos actually used that to get to the field faster. So that is one of the benefits and value adds. We also provide very easy and reusable architecture for our customers.

So we have architectures and patterns that our solution architects develop and our customers can reuse and easily get started with. We have workshops and we have sample applications that are readily available. So here, what the cloud has made possible is to create a set of products that are connected and intelligent with cloud capabilities. We have a service called AWS IoT Greengrass which allows software updates, over the air software updates in a much, much speedier manner than has traditionally been possible. With Greengrass you're able to take your existing robot code, robot applications and deploy it very seamlessly.

Whether you're using ROS or you have a docking container, containerized application. It becomes a repeatable, scalable process to deploy and configure robots. And then once you have that ready, once you have that pipeline setup then you can get data and telemetry streaming back into the cloud to perform more of the analytics and insight that we discussed with AI/ML services. Let me, I'd love to show you a video of a customer who has been able to leverage all these capabilities and services that I just talked about. In a very friendly assistive robot, Lea. Which is, it's an assistant walker.

It was able to get to market much faster using AWS cloud and all of our services. And we're also able to make a much lower cost hardware because, you know, leveraging the cloud they were able to off put some of their hardware costs and just spread it over time with cloud usage. So this is actually a medical class device. It has passed regulation. It has some level of autonomy.

It has a great level of interactivity and a very friendly user centered design. So let's take a look at how Lea works in the real world. - Lea, let's dance! - [Lea] What would you like to dance? - I'd like to dance the samba. - [Lea] Okay, let's dance the samba. - Okay, here we go.

(funky music) (singing in Spanish) - There you go. (singing in Spanish) And I don't need a Lea yet, but the day I will I'm ready to dance the samba with Lea. So, let's take it. So, that was a very fun example. Let's take a look at a much more critical, chronic situation where it also becomes very helpful.

- [Sebastian] It doesn't have the audio. That's a Parkinson's patient. - Yeah, so this is a patient that's suffering from Parkinson's Disease.

And actually the intelligent capabilities in Lea are able to predicatively monitor and predicatively anticipate the gaits. It's able to learn over time and provide certain capabilities that have not been traditionally possible by other devices. So, the data that we're able to gather with the device provides a much better experience and it's able to tune and attune to the patient needs.

So this has changed the life of hundreds of hundreds and probably now thousands of patients suffering from this disease. So, we work very closely with the Lea team to make a very low cost device. Although it has a lot of sensors, they were able to save on CPU and compute and they were able to leverage open source software to get up to speed and up to market much faster, so. Another great example of a customer in this space is MultiplyLabs. MultiplyLabs was able to develop software for their lab automation in a much friendlier manner, a very modular and redundant architecture and they're able to do continuous validation of their manufacturing facilities.

And they were able to scale from tens of robots to hundreds of robots. Saving them literally years of development. So, MultiplyLabs is another customer in this space that has made a significant impact.

And we recently have had actually a very large pharmaceutical company that has, you know, they're somewhat very interested in how these startups are able to innovate so fast and get to market. So, you know, you see the enterprises looking at these technologies and the capabilities that makes it possible for the startups to innovate. At AWS we have a set of services that are specifically made for the robot software development life cycle. So for developing your application, testing, validating and simulating your application we have AWS RoboMaker. Which is a developer infrastructure, simulation infrastructure. And to do final testing and to deploy, manage, and update your applications we have a host of AWS IoT services like Greengrass, IoT Core, IoT TwinMaker, Kinesis Video Streams.

And we're also very excited about a new product that we're getting to market this fall. AWS IoT RoboRunner, which allows customers to have a heterogeneous fleet of robots, connect them, program them, and orchestrate them in a much easier manner. So most operators today, if you have different robots, from different vendors then you have to develop, use different applications, disparate applications. Where with RoboRunner you're able to run them in sync.

And this has become very powerful in hospitals that are running robots from different OEMs. There's examples in Singapore by the team at Open Robotics. And this is a significant use case for our customers in the healthcare world.

So, you know, as I mentioned, RoboRunner becomes your central data repository where all the fleets of robots stream their data to. We have a common API across OEMs. We partner with the major robot OEMs like MiR, Clearpath and others.

We have a number of sample applications to orchestrate and do task management of the robots. And then we have vendor integration for easier deployments. So we work very closely with those vendors.

And just to close all this, take a look at some simulations that leverage RoboRunner and the host of our services. (light music) So just to wrap up on RoboMaker. We essentially have a service layer, serverless contain infrastructure. You only run the simulations for as long as you need them and no longer. You can execute them without having to manage them. We do all the management of compute and storage for you.

We have a batch simulation infrastructure as well, if you wanna parameterize and do random, domain randomization across your simulations we have an API that makes it very easy. It makes it very easy for you to accomplish that. We're able to generate worlds and we're able to easily integrate with your existing continuous integration and delivery pipelines. The RoboMaker containerized architecture essentially looks like this.

You can create a container of your simulation environment and then you can create a container of your autonomy stack. We take those containers from ECR and we run them and operate them for you. We also have capabilities to be able to do multi-container architectures. Once you have tested and validated your simulations, then you can leverage all AWS services to do analytics. You know, you can use Cloud Watch and then you can deploy those applications to the Edge with IoT Greengrass 2.0.

So without any further ado, I'd like to hand it over to Sebastian Arias. I think he'll tell you more about Cosmos and how they are operating in the skilled nursing facilities. Thank you so much. - Thanks, Cam. - Thank you. - How's everybody doing? I'm gonna ask you a few questions initially.

So, we'll make it very simple. If you can raise your hand, do you think there's staffing issues at nursing homes? You can raise your hand. Probably every single industry has 'em, right? Do you think they're caused or amplified by COVID? Raise your hand caused. Caused? Raise your hand amplified.

Okay. So we can show this really cool video that I found just a few days ago, but I'll show you the link if you wanna for copyright issues, we couldn't show it, but I'll show you the link if you wanna check it out. But let's look at the stats. There's gonna be almost a million nurses that are required in the workspace by 2030 that are not gonna be there.

And it's one of the fastest growing jobs. Hospitals experiment, including some of our customers, they experiment. This is actually Atlanta based numbers.

We're based in Atlanta. 24% turnover. So imagine you hire somebody and have to replace that person.

So let's say you have a 100 nurses and you have to replace 24% of them continuously every year. That's a bit excessive. That's the video that I was telling you about. Now, another question. Do you believe that robots could solve this problem? If you wanna raise your hand, anybody who believes that.

Oh yeah? Okay, the majority. We believe that too, at Cosmos. We're 100% proponents that robotics is apart of the solution as a whole. Not necessarily the only solution, but an integral part of the whole solution.

This used to be only on cartoons, but this is a raw, uncut video from a facility. That actually Cam visited. - [Man] Yeah it's gonna go all the way there. - And that's actually a resident that communicates with our robots on a daily basis. Friends with our tele-operators.

- [Man] What is he saying? - So imagine if you're an elder and you live in a nursing home, and nobody visits you. So you're lonely. Who are you gonna talk to? You're gonna talk to your nurse, to the CNA, so the certified nursing assistants. But what if there's none? What if there's one nurse per 20 residents? And what if they require full time care? Meaning, you gotta feed 'em or at least deliver the food. You gotta change their diapers and you gotta bathe 'em. What do you do in that case? So we use a hybrid solution.

I'll explain it to you in a second. But this happens on a daily life at this facility. And it's actually, Cam saw it, it's actually pretty cool to see it, right? - [Cam] Yeah, it's pretty amazing, yeah.

- So meet Aby. This is our solution. It's a robot that assists nurses and CNAs with their admin tasks. So we let clinicians do the clinical work, which is what they love.

Which is why they became that person. They don't like inputting data when something can automate that. So Aby provides several services.

Primarily we provide administrative automation. For example, in hospitals we provide customer satisfaction rounds, billing rounds so that the charge nurse, or the leader nurse can do that. Can actually focus on providing better care. In nursing homes we provide companionship.

Like what you saw in that video, an example. And we've been operating in a facility and I'll tell you the stats. Where we can actually quantify now how much time per day each of our operators are spending.

That it's 100% time saved for the CNAs. We can also deliver small items within the units. And we can take about 10 to 20% of the nurse calls. So whenever they click the button they need something. Such as a remote or they wanna call somebody. They can, if they don't have a cell phone, or they wanna see what the menu is for the day, etc.

So we operate at Presbyterian home of Georgia since December 2021. And some stats. We operate daily for more than six months now. There's three robots there continuously and there's a new one, the one they used on the video, that's a new robot. We use ROS and we also use AWS. We have two terabytes of annotated data collated, are collected, back data from ROS.

Our operators are spending between 20 to 30 minutes per day per facility. Which is a lot of time that is usually spent by the nurses. And at this specific facility, the main tasks that are performed by Aby are companionship and sitting.

So people that need to be looked over because they might fall or they have other cognitive issues. So we haven't announced this, but we will be announcing soon. We're doing a partnership with a large health system in Atlanta. And one of the, so that specific unit where we're working at, the leader nurse for that unit, it's a 30 bed recovery, like long term recovery unit for after surgery and COVID. She needs to do two jobs. The leader nurse, she needs to do two jobs.

So a lot of times she doesn't have time for us to go there and map the data and get real data, like Cam was saying. So we use simulation for that. Especially when the data is considered privacy, or there's either privacy or security issues. We need to get some simulation initially. We, Cosmos uses AWS primarily because we need a pretty robust privacy and security solution. We're a startup so we don't have the resources to have a system mapping person on staff.

AWS is pretty much the best cloud solution in healthcare from all the conference that I go to. We use RoboMaker like where we showed you for a simulation. And soon, with the new robot, we're gonna deploy Greengrass. I think Greengrass version two. Because our robots are on a private cellular network.

So we need to manage data pretty good and we need to manage power consumption pretty good. And with that solution we can do that. And we're here for your questions, thank you so much. We look forward to continuing to working with AWS. So, that's it. Thank you.

Do you have any questions? We're gonna repeat the questions, so you can say it and then I'm gonna repeat it. Or Cam's gonna. Did she tell you? You have to repeat it because it's being recorded.

- [Cam] Got it. - [Audience Member] How do you handle disinfection on your mobile units in healthcare? - So if I hear the question correctly, how do we handle which part of the? - [Audience Member] How do you make sure that the unit isn't spreading germs? - Got it. - [Audience Member] Do you have a cycle where you? - Yeah, so the question is how do you ensure that the robots are not creating more challenges in the facility. - So, in hospitals we deploy one where it can be wiped down for that purpose.

And we do it manually. There is no automatic disinfection within the unit. There, again, we're a startup. So we try to focus resources where it matters.

But there could be a solution where you just put it through a disinfection system. But the issue with that is the hardware components have to be modified a little bit. So, either if it's water based it has to be fully water contained. Yeah, did I answer your question? - [Audience Member] Question and a suggestion. One, probably the biggest thing would be air that a person walks through. - Yeah.

- [Audience Member] So the robot actually has an advantage because it's not sucking in air and blowing out germs and viruses and you know. Which is the only place I can see that pushing through would be a fan. - Yeah.

- [Audience Member] And it would be a lot cleaner than a person walking back and forth. - Agreed. So for example, one of the units where we test in we're gonna have a designated COVID robot essentially. Where it won't go in between the med room. Usually most units have like a med room and then they separate the COVID rooms with non-COVID. But this is high risk for the nurses, or the doctors because even if they're not working in the COVID unit, there could be a possible contamination.

So in that case what we're doing is we're separating our robots specifically for that. - [Audience Member] One of the questions I had was like, do you still have the person that has to be on that iPad for the camera itself. So you're not really replacing a person, they're just, they can sit at home and do the job. - Exactly. So, the question was, we still have a person that is having communication with the resident or patients, but we're not replacing a person.

It's just replacing the way that they do it, right? That was the question? - [Audience Member] Yeah. - Yeah, so initially we did it automated with human-robot interaction. But since our census at that facility is about 75 years average, they didn't accept it pretty well. We haven't seen pretty good HRI that can replace a human. So for our companionship, that's actually a value where you become friends with a human behind the screen and then the adoption rate, it's easier.

The goal is that, to transfer over more to an automated solution when it becomes good. So in our pipeline we do have some NLP projects that we're working with, where you can maybe go deeper than two or three questions. And that would be pretty cool. - [Audience Member] And another suggestion, they're building avatars of like Tony Robbins and stuff like that to answer whatever question you put up.

And I wonder if your companionship, if you record each interaction, if at some point you'll be able to have it seamless where somebody wouldn't know that they're not really talking to somebody? - Yeah, it's a cool idea. So unfortunately we work in privacy constrained environment so we can't record because of HIPAA. But you could, yeah, with the avatar concept they're getting really good, so yeah. - [Audience Member] The chat bots are like, amazing. - Exactly. - [Audience Member] So yeah, I would think you're gonna have an enormous amount of data.

- Yeah, so we, there is an enormous amount of data. You're right, in hospitals and nursing homes. And they're pretty antiquated systems, so. Exactly. Did I answer your question? - [Audience Member] Yeah.

- Okay. Do you have more questions, guys? - [Audience Member] So regarding the simulation that you showed earlier. Like, in the past we have seen some challenges building such a simulated environment in that it can be significantly different from the reality of like lighting, and then different conditions. So, I guess my first question is how much manpower you put in building this kind of simulation? And second is, how accurate do you think the result is in delivering the solutions? - Yeah. You wanna take this? - Yeah, so the question is, there are some challenges with simulation and getting the right simulation environment that mimics the world. How do we go about getting the right level of fidelity and viability of the simulations? So, absolutely.

Simulation is only as good as how verifiable and close to the real world it is. So first of all, when we work with our customers we usually try to find what is the actual use case that you need to simulate? And getting the right level of fidelity. That, once we understand the right level of fidelity needed for the simulation problem, then it's a matter of getting the right tool for the right job. So today in the simulation world there is probably close to 100 simulation tools and they all serve a very specific use case or a very specific set of use cases. And it's really a matter of being able to pick the right tool for the right level of fidelity needed on the given use case.

And then obviously you have to create a, and we had a workshop earlier on simulation and there's this expression which is, all models are wrong, but some are useful. Right? So, you have to get the simulation models to a level of reliability and validation that matches the use case. So there is, depending on what the problem is, and what the use case is, there could be a significant level of involvement in creating a simulation model and environment that mimics the real world.

For some of the navigational use cases in hospitals, those are pretty common, you know, physics engine type of navigation simulations. So they didn't need the level of fidelity that perhaps let's say a surgical robot would need in terms of contact forces. So those are completely different tools that you would use for those particular use cases, so. - Yeah, for Cosmos one of the benefits for simulation is, like I said, data privacy. There's no, we haven't found any dataset that, since this is pretty much like a unique solution.

We haven't found a dataset that we can work with that's HIPAA compliant. So that's our main, with a specific robot, to use that and deploy faster. But we're big proponents of deployment first. But kind of like a hybrid model.

Did we answer your question? - [Audience Member] Yes. - Awesome. Y'all have anymore questions? Discussion. - Well, we'll be right outside after this talk so if you wanna have some more discussions or conversations, we're happy to do it over coffee. Thank you.

- Thank you. - Yup.

2022-07-03 14:29

Show Video

Other news