How McLaren Racing benefits from DataRobot’s AI-powered predictions

At the Miami Grand Prix last month, McLaren Racing’s Edward Green was monitoring a data dashboard with a graph showing a predictive model of track temperature. This is a key metric for Formula 1 cars, as hotter asphalt wears tires faster, affecting strategic decisions about when to pit and which of the three tire compounds should be put on. car.

Suddenly the line on the chart began a rapid descent, so unexpectedly that Green, McLaren’s business technology manager, joked, “Oh, that doesn’t look good. We must have broken something.

A colleague, however, assured him otherwise. The algorithm predicted rain, which would cool the track. Although Green saw blue skies and sunshine, he shared the information with people in the hospitality area, who grabbed precautionary umbrellas.

“Of course, 20 minutes later it started to rain and the clouds came in,” says Green. “It just made me smile, thinking it’s pretty powerful that the racing team that’s out there is using that information.”

This graphic is the product of a new AI model built and maintained by DataRobot, McLaren’s new partner this season. DataRobot helped weed out the data and fine-tune the AI ​​models to make better predictions, even taking into account details such as how Monaco’s hillier train affects cloud movements compared to the flat course from Miami. The rain in Miami cooled the track enough for the F1 cars to do a few extra laps on their tyres.

DataRobot is integrated with the McLaren Racing infrastructure, providing AI-powered predictions and insights.

“It sounds a bit pedantic, but it’s one of the most extremely important things you can do for the race strategy team,” says DataRobot AI Evangelist Ari Kaplan. “So one of the successes has been being able to better predict the actual air temperature, as well as the track temperature, more accurately than before.”

An F1 car is equipped with 300 sensors, many of which sample data at different rates. The fuel flow sensor measures 2200 times per second. Collectively, they produce tens of thousands of data points per second – and that doesn’t include any of the myriad external sources – which will be reviewed by the on-site team, which is in Azerbaijan this weekend, and engineers from Mission Control, the nickname for McLaren’s headquarters in Woking, England.

“It’s a very data-rich sport, and we’re actually collecting and analyzing more data than we know what to do with,” says Green. While emphasizing how smart the engineers on his team are, he notes that a bit more thoroughness was needed in their analysis.

“We’ve been managing models and using AI for a while now, but like many organizations, as part of the IT team, we kind of let people handle it,” he adds. . “And we didn’t really ask too many questions about the model they were using, how they made that choice where they were deploying it. It really was done without much oversight or oversight.

This is where the DataRobot partnership, which was announced last November and includes McLaren’s F1 and IndyCar racing teams, plays a role. DataRobot maintains an AI cloud that serves a third of Fortune 50 companies.

DataRobot's Ari Kaplan (L) with McLaren Racing Director of Strategy and Sport Randy Singh.

DataRobot’s Ari Kaplan (L) with McLaren Racing Director of Strategy and Sport Randy Singh.

“Whenever you have an AI opportunity,” Kaplan says, “you look at all the different factors: What data is readily available? What kind of insights could be given? What is the real value? , is it actionable and practical? Or is it just Trivial Pursuit and you get some funny data? It might be good for the fans, but if it’s the racing team, [they need] the things they can do.

Every area of ​​race strategy can be optimized to support drivers Lando Norris and Daniel Ricciardo. McLaren’s pit times have been among the fastest on the circuit this season, with Norris making the quick pit stop during the Emilia-Romagna Grand Prix of Italy, needing just 2.27 seconds to restart.

Kaplan, who was a pioneer in baseball analytics, describes the role of AI in planning when these stops – based on car wear, fuel management and weather – as well as the complex “ballet sequence” of so many crew members attending to the car simultaneously.

“You can spend hours and weeks in the wind tunnel trying to gain hundredths of a second on the car, and you could lose that to a pit stop,” says Green.

The redesigned cars racing this season in F1 are optimized to create ground effect, an aerodynamic phenomenon that generates downforce. Green describes the change as going from “how much air we can push over the car to how much air we can pull under the car”.

To facilitate this transition, McLaren has built a digital twin of the car. Each new part begins life on a designer’s computer where it undergoes rigorous computational fluid dynamics (CFD) modeling. Once it meets a certain standard, Green says they use a 3D printer to build a 60 percent scale part for wind tunnel testing. Only after passing these tests is the part physically built for in-car testing. These simulations are one area where AI has helped “massively,” Green says.

But despite all the proliferation of data and technology, AI models remain complementary to the process.

“We have this ‘human in the loop’ concept,” says Green. “So it’s not about letting the AI ​​make the decision for us. It is not by letting the machine model dictate your strategy that you click with. It’s really about giving the human in the loop the best information, the best insight possible, so that they can go and execute that decision with all the other systems and information that comes out of the rest of the team, whether that either on the trail or be that back in Mission Control.