How Google is using AI to make faster Racing Drivers

0:00 A few weeks ago, Google Cloud sent me a message. Tom, we know you're a lover of AI and a big fan of motorsport. Do you think that AI could be used to help drivers go faster? To which I responded, probably. So, let's find out. Following this swift and efficient email communication, they invited me to the Google Cloud Summit event in London to put this hypothesis to the test using a simulator. And once I've done my driving, we'll also break down how this AI project works behind the scenes. So stay tuned for that. Thank you to Google Cloud for sponsoring this video. Okay. So Jack, before I jump in, can you tell me what I'm going to be doing? Yeah, sure. So it's going to be three laps around the circuit using a Formula E simulator. Uh we're going to take all of the data from all the sensors on the virtual car and we're going to send all of that data over to Google Cloud and

0:46 then we can do some driver analysis afterwards and compare your performance to a professional. So hopefully then afterwards I'll improve, right? Yes. The idea is the coaching will hopefully make you improve. But let's do it. Cool. Jump in. So, it's a bit of a squeeze all. It's a little bit tricky to get in and out of. It's a And there's a bar between your legs to move yourself forwards if you need to. So, left foot for braking, right foot for accelerating. Okay. Steering wheel point where you want to go, of course. The red lights will tick on when the red lights go out. Off you go from the line. Uh, three laps around and see how fast you can go. Oh, I went way too deep. Way too deep.

1:34 Okay, so that was lap one. Hopefully the next one's a little bit better. [Music] Oh dear, Tom. Oh dear. No, it's not in the bag. Happy. Awesome. Happy, but not how how do you feel about that? Amazing. You did a 110 728 which is I think literally puts you in second place

2:20 behind the professional drivers. Oh, sweet. So fantastic result. I feel really good. Yeah. Well done. So the idea now is that I'm going to look at my results. I'm going to get an analysis generated by AI and then that is going to hopefully give me some insight into how I can improve my lap time and essentially coach me to become a better driver. Uh so using Gemini to review all of the data um that the car collected and then we'll break it down in comparison with our professional that we've uh used before. The initial launch speed yep slightly slower was slightly slower compared to professional drive but not a huge difference is what I'm taking out. Yeah. Very good. So they carry more speed. So I'm braking really harshly. That that's what I'm getting from this is I'm just kind of going speed up and then stand the brake

3:06 off and they have more of a flow through the through the corners. Okay. I think it's a consistent theme throughout, right? It's just about the smoothness. I need to chill out. Okay. Should we give it another go? I'm sure you're all dying to find out how my second attempt on the simulator went. But before I show you that, we need to ask the question, how does this actually work? Conveniently, I bumped into Dan from Formula E, who knew the answer. So, you were you were racing on a piece of software called RFA 2. Okay. And what that did is it that that wrote the the data from your drive. We brought that in through uh an agent that we put on the on each of the sims and we send that up into a BigQuery database. Okay. From that BigQuery database, we can then compare that to a professional driver and we also compare it to some perfect

3:53 laps. Now the perfect lap is never is never driven, right? You can never see anyone do that. But we if you compare it to the professional driver, you can see what's achievable for you to do. And then we try to give you feedback. We look at overall what level were you? we can see some of the previous ones you've done if you'd logged in and then we can tailor that feedback that we've given you based upon your level of driving. You will, you know, you're a good sim driver but you're not a professional driver. And that's the the real thing that the large language model can do is it can tailor the feedback to the person that needs it in in the right way for them. What is the real uh the real use case of this technology here outside of Formula 8? So what what Google Cloud would like to do is see how this evolves into we call

4:39 it a race race to roads. So we develop something in the racetrack what we can put into the into the real world. So we we do this here. We can show it in in racing. This teaches me how to be safe and and teaches me how to how to use that most effectively. Now at the moment we're we're giving you written feedback from the simulator. What you'll see is over the next year is we're working with a deep mind team to make that real time. So you put in some earbuds and you can have constant real-time feedback about how you're doing and how you're you're performing and help you. It also helps then is as you get tired through the day or you get um bad habits start to be introduced, it can help you stop them. So re realistically this not only comes from education but also from a safety point of view. Correct. And this is going to help larger work forces

5:26 essentially be able to enforce their safety policies on site in a live and interactive manner instead of it just being walls and walls and walls of documents. The real thing is you're saving time now. We can crunch that that time into maybe an hour that you lose rather than a day that you lose. It's almost like when you're playing a game, you have that crash course, don't you, to this is how you accelerate, this is how you break. We're doing that in the real world. Exactly. Amazing. Now that we know how it works, here's the moment we've all been waiting for. Fully informed with all of my feedback from the AI and Dan's insight into how it works, it was now time for the moment we had all been waiting for. The opportunity for me to prove everyone wrong. Three laps, one goal to go faster than I did 5 minutes ago.

6:15 I think the time was better. But was it faster? Let's take a look. My fastest time from my first run was a respectable 1 minute and 10.728 seconds. I'm pretty happy with that. My fastest time from my second run was 1 minute and 8.201 seconds. That's over 2 seconds faster on my second try. So end result, I am much faster. But what are the real world applications of technology like this? I think if you look slightly outside of the world of motorsport and maybe operating heavy machinery that someone has never operated before or trying to keep them within guidelines of using that machinery, that's a great application. Another one could simply be helping people to learn to drive or even

7:00 just any kind of high precision performance task that someone needs to do to a high quality. To summarize, I think that the use cases of this kind of technology are completely endless and will hopefully transform the way that we interact with objects, machinery, and tools in the future. If you'd like to find out more about the AI agent in this video, you should check out this post on the Google Cloud blog. I'll leave a link in the description for you to find it. Thank you to Google Cloud for sponsoring this video and having me at this event. I'll see you next time.