Racing at the Speed of AI
In the Indy Autonomous Challenge, university teams replace human drivers with cutting-edge AI to pilot real Indy race cars at over 150 mph. Caltech’s team pushes their algorithms to the limit—fine-tuning code that must react in milliseconds to keep the car on track. Amid wet conditions and high-stakes laps at Laguna Seca, they achieve record speeds before a dramatic crash ends their run. Beyond the race, their work fuels breakthroughs in autonomy—from spacecraft to robotics—proving that innovation accelerates fastest at the edge of control.
Transcript
00:00:00 [Droning Music] Indy Autonomous Challenge – Indy means the Indy Car but they replace the human drivers with AI software. We use a vehicle with the horsepower that professional race car drivers use, with sensors and actuators that allow us to control it via drive by wire. And now university teams program the car to compete for the fastest lap time with the AI software. It pushes us to the limits of what our algorithms can do. [Music and Car Noise] Once you have that high speed, it's going to amplify
00:00:55 whatever challenges you had – that's very unpredictable. It's a good motivation for us to think about what kind of gap we have between what we do in the laboratory and what we need in the real world setting. Once you have an autonomous vehicle navigating very complex environments, you really have to push the envelope of what you can do in terms of autonomy software. Because we are moving at 150, 170 miles per hour, you cannot afford to have your code operate at any speed other than lightning fast. We're talking about differences in hundredths of a second that can move the car tens of meters. And that could be the difference between a beautiful corner or a nasty spin. ...We're coming into Laguna Seca.
00:01:56 It is kind of wet out there, so I'm kind of concerned about that, but, um, it should be cool. About half an hour before our race slot, we head over to race control. We set up, make sure that our telemetry is showing that everything is working the way that we expect. You push a button on your laptop and then the car slowly rolls out of pit lane. We watched our car do its out lap. It looked beautiful. The first lap that we took on also looked great. The second lap was much much faster. But when we took on the corkscrew, we experienced severe wheel lock. It's a very, very complex mechanical and dynamical system. So when you actually increase the speed on the corners,
00:02:46 it's going to put the car on the edge of the traction limit or friction limit. We really pushed the envelope. We got the fastest lap time in the first lap and second lap. However, on its third lap, I notice in our telemetry that our tires are locking. For a fraction of a second, I see a car sliding past the corner. I hear in the broadcast, "He's not come around yet, Rob. I think there's been an issue..." You know, they're all asking, "Where's the car?" Meanwhile, I feel this pit in my stomach. I know exactly where the car is. "So, Caltech Racer's two lap times, they were the fastest, but they exist no more..." Other teams have been doing this for five years.
00:03:41 Because we've been only working on this project for nine months, I think we have a very respectable performance so far. But still, we have to improve our method. It is exceedingly rare in research that you get to work on something that is so directly applied with such concrete real-world consequences. Autonomous racing is one application, but once you develop the full autonomy stack that can drive, and avoid an obstacle, map the unknown environment... This kind of research and method can be applied to spacecraft, to planetary robots, and so on. Autonomy is everywhere in our society and the types of algorithms that we develop, for example, for fault detection, for fault recovery, are things that could apply to oil rigs, to spacecraft.
00:04:36 There's no limit, truly, to where rapid response to failure and changing conditions will apply. I can't say that I'm satisfied with our performance. I think we left some performance on the table. But what we also did is prove to everyone that Caltech means business. And that we're coming for them next time. [Music and car noise]