 from Las Vegas. It's theCUBE, covering AWS re-invent 2018. Brought to you by Amazon Web Services, Intel, and their ecosystem partners. Now welcome back to Las Vegas. We continue our coverage here in the Cube of AWS re-invent. I'm John Walls, along with Justin Warren. We're now joined by Mike Miller, who is the senior manager at AWS AI. I'm Mike, good to see you this morning. Hi, gentlemen. And we love toys. We love props. And Mike's going to show us that, I guess, development is, it can't be child's play, but it's not. You've got DeepRacer here, one-eighteenth to a monster truck scale, but what's the real deal? What's going on here? Absolutely. Well, first of all, thanks for having me on. You bet. It's really my pleasure to join and have a chat with you guys and tell you all about AWS DeepRacer, which we launched today. So DeepRacer is a reinforcement learning platform designed to help developers along a fun and easy to follow journey to learn about reinforcement learning. AWS DeepRacer comprises two components. There's a console and there's the car. The console is designed to teach developers about how to build, train, and optimize reinforcement learning models in the cloud by leveraging Amazon SageMaker and AWS RoboMaker. It shows them all the tools and how they fit together to train a virtual model to drive a simulated car around a virtual race track in a 3D simulator online. Developers can then take that trained model, deploy it onto the car, and test how their model actually works in the real world by driving the car around the same type of track. So into this car, or I assume this is not the only car. That's right. So the cars are available today for pre-order on Amazon.com and they'll ship in early 2019. And our console will be available in a preview mode and interested developers can put their name on a wait list online today. Yeah, this is actually the front of the car here. That's right. So there's got a camera here. That's right. As we mentioned, the car is a 118 scale RC car chassis built on a four by four monster truck chassis. And we've enhanced it with other capabilities to allow it to autonomously drive. So for instance, we've got a camera in the front. We've got a main logic board with a Intel CPU running on it. And when that model is deployed onto the car, it's basically taking the input from the camera. It's looking at the road in front of it and then determining what kind of instructions to give to the car. Throttle up down, steering left, right in order to stay on the track and potentially maximize its time driving around. So what we'll make, well, let's first of all, I believe you, I think you told us ahead of time that you've actually set up a fun little competition that's going to kick off fairly soon. Tomorrow morning, I believe. So we're going to have a chance to see this put into practice here at the show. Absolutely. It's actually launching at 11.30 today, which I think is right now, over at the MGM Grand Garden Arena. We've essentially taken over the floor of that arena and installed six tracks. And we've got a system there and a series of workshops that we're going to be teaching over 2,200 developers over the next few days how to use AWS DeepRacer and get their first taste of building a reinforcement learning model, deploying it to the car, and then over at the MGM Grand Garden Arena, they can put their model on a fleet of cars and actually see them driving around the tracks. And now, what's a car without a little competition? So we also launched the AWS DeepRacer League today, and we've kicked off the inaugural season of the DeepRacer League here at that MGM Grand Garden Arena. So the fastest lap times for developers who are able to get there today before 10.30 PM and race a car will be eligible to race in a final that's hosted on stage before Werner Vogels keynote tomorrow morning. Very cool, wow. And then in 2019, this league will take place both in person at over a dozen AWS summits worldwide, as well as online, where monthly we'll be releasing new tracks that developers can then train new models and optimize new models to drive on those tracks. The winners of each of those competitions and the top point scores will be eligible for an all expense paid trip to reinvent 2019 to participate in our grand champions cup there. And hopefully they'll be able to like hoist the big champions cup that we've got for the winners. Very cool. So for people who want to get involved with this sort of thing, but don't know anything about machine learning or reinforcement learning, so maybe tell us about that. What is reinforced learning? How would you get started with this? Yeah, thanks, we get that question a lot. So in machine learning circles, reinforcement learning is getting a lot of buzz because it's a very exciting technology that's broadly applicable. What differentiates it from other machine learning techniques is that it doesn't require large data sets in order to train a model to make a prediction. So for instance, supervised learning or unsupervised learning where maybe you would train a model on how to recognize a flower based on labeled pictures of flowers. Reinforcement learning has a strength where it can make a series of short term decisions to optimize for a long term goal. So it's really applicable to challenging highly varied environments like healthcare treatments, autonomous driving, obviously. Optimizing manufacturing supply chains. And so what's interesting though about that technology is it's a very steep learning curve and it requires a lot of in-depth technical knowledge which effectively puts it out of reach of all but the most well-funded enterprises today. So we believe AWS DeepRacer is just another tool for us to help get this kind of innovative technology into the hands of everyday developers and data scientists and help them get up that learning curve and really see how to get some intuition for how reinforcement learning works for their world. So if you're providing firstly the same tools to everybody, right, the same capabilities, whatever what will make some cars faster than others? I mean, ultimately, if Justin and I are equipped with the same abilities or capabilities and we have your help, we're going to get to the same end. So how do you get a winner, if you will, and then how does that translate to my business? Great question. So part of the core of reinforcement learning is building what's called a reward function as well as a set of hyperparameters or the parameters for the training episodes. So remember I mentioned we train a virtual agent in a simulated environment. And the reason we do that is because these agents, they don't really know how to get the optimal goal initially, instead they wind up exploring their environment. So you've got an agent that takes an action and the action has some effect on the environment that's fed back into the algorithm so that it can then use that knowledge down the road to refine its choices. And so that's defined by both a reward function which determines what kind of rewards or lack of rewards you give your model during the training process. So for instance, when you're racing around a track, maybe you want to provide rewards for less, like fewer steering adjustments because you know it's going to go faster if it's very smooth or you want to balance the amount of time that you focus the learning on driving along the dotted centerline of the track versus finding sort of optimal angles as it goes around corners. So that's where the art meets the science of building and training and testing and optimizing these reinforcement learning models. I know people have done this kind of thing to get computers to play Super Mario Brothers and find all of the weird ways that you can cheat and do things. So I wonder, is there a demolition derby mode for this? Am I allowed to sabotage my competitors? Certainly not initially. Because if we both want the centerline, is the first one there wins? Or I mean, how does it work when you have 10 cars on the track at the same time and they've all been programmed to kind of do the same thing? Yeah, on the tracks today, we're operating a time trial format. So there's a single car on the track. So we avoid some of those situations. However, you know, this is an amazing platform that we plan to keep adding on and building over time. And so in the fullness of time, you might see us add some capabilities that allow developers to train different kind of behaviors, maybe passing behaviors, or maybe behaviors that recognize checkered flags or yellow flags or, you know, look ahead to curves. And so, you know, we can't count any of that out. And I think those are some exciting directions we can take this. Putting big spikes on the wheels in a bin-hook on a franchise. All right, so kicks off, and it has kicked off, tomorrow morning we're going to have a kind of a showdown of sorts before the keynote. And then you're, this is a year long opportunity of learning that we're going to, I guess, bring to a culmination a year from now, right back here, right? Absolutely, and for those, you know, you can have two opportunities to participate in that league. We've got an online version, of course, where you don't race a physical car. So you can just train and optimize your model in the virtual environment to race around the tracks. Or if you purchase a car, or just come to one of our summits, you can use one of our fleet cars to see how your model performs in the real world. And if you do it online, how would we find that? Oh, that's right, off of our detail page, aws.amazon.com, slash deep racer. Slash deep racer. All right, well, Mike, thank you for being with us. And I know a little guy, Max, back in Australia, who would love that. Christmas is coming up. Just think about it. All right, Mike Miller. Thank you. Thank you guys for having me. No, it was very cool. Thank you. Back with more, we're going to put the pedal down with this afternoon's coverage here on theCUBE. We'll take a little bit of a break and we'll be back in just a bit with more live coverage here from AWS RANDVET.