How General Motors Is Using AI to Win on the Racetrack

Tucked away inside the General Motors Charlotte Technical Center in North Carolina is a room that would have made the Apollo mission controllers jealous. Multiple tiers of semicircular desks sit flanked with monitors, a collection dwarfed by a wall of displays streaming everything from live race footage to reams of telemetry and GPS data—all beamed straight from the track.

It's called the Motorsports Command Center. Considering GM campaigns cars in top-tier series across the country and around the world, it's not entirely surprising that such a room exists. What is unexpected, however, is that this room is being used to coordinate a fast-growing but rarely discussed aspect of motorsports: artificial intelligence.

"We've got 10 to 12 people in a room like this on Sunday, supporting 15 cars," Jonathan Bolenbaugh said, contrasting that with the dozens of engineers typically required to monitor the telemetry coming from a single car in a top-tier racing series. "Because we have this technology, we're actually able to support at a bigger level with less people."

Bolenbaugh is the analytics manager of GM Motorsports' software engineering arm, leading a small and agile team tasked with finding ways to process, analyze, and optimize ever more data as it’s being streamed. He founded the team two years ago, staffing it with Ph.D.s and data scientists, many of whom had never previously worked in motorsports.

The team's size and capabilities have grown through internal development and acquisitions, with its digital tendrils now reaching into everything from NASCAR to the FIA's World Endurance Championship.

See All 3 Photos3

AI, Trackside

AI is an increasingly overused and often misunderstood term, and that's partly because it means a lot of different things in different applications. The general idea is some sort of software that learns based on a set of data and can present insights, spotting trends or calling out key information in that data that would be difficult for a human being to spot.

When it comes to GM's AI efforts in motorsports, a fundamental piece is something that Bolenbaugh calls "multimedia intelligence."

He said that trackside photographers shoot "tens of thousands" of photos during an average race weekend, a volume of visual information that would quickly result in information overload in an average human being. AI, however, can Hoover all that data up and instantly highlight things like vehicle damage, not only in GM's own cars but also in the competition’s.

And it's not just visual information. The system monitors and transcribes radio communications from every team in the race, automatically highlighting relevant information being fed between the pit wall and the drivers.

"We can capture it,” Bolenbaugh said. “We do speaker tagging, we do some analytics around sentiment analysis, categorization, and start trying to infer what they might be doing based on what they're saying."

The systems can use all this data to make more accurate predictions about what other teams will do and when. Usually, a given team struggles to stay on top of what they should do with their own cars and when. But with feedback from GM's AI systems, teams can have digital eyes watching the competition, as well.

Part of this technology came to GM through a 2022 acquisition of a startup called Pit Rho, which used aspects of machine learning to analyze real-time race data.

"It provides predictions of what our competitors' performance capabilities will do, in terms of tire fall-off, their potential fuel strategies, tire strategies," Bolenbaugh said. "And then we can use AI to optimize what we should do."

GM's suite of AI tools can even provide feedback on a per-driver basis. The systems can identify when a given driver is losing time in a certain corner or section of track, information that can (hopefully tactfully) be relayed back to the driver.

"It can basically do some digital driving coaching so they can see where they're gaining or losing time, and maybe how they can change their inputs," Bolenbaugh said. "Or even maybe they might need to change the setup on the car to allow them to execute a corner."

Pit Now

Snooping on the competition and watching driver performance are key features, but pit strategy might just be the killer app for AI in motorsports. We've all watched races where a perfectly timed stop has made someone's race, but in endurance racing, it's more often about finding a perfect cadence of visits to the pit lane.

Regardless of where you're competing or for how long, maximizing available resources wins races, something that again requires monitoring endless amounts of data, from tire wear to fuel consumption, all while also keeping an eye on the weather and predicting the odds of on-track incidents leading to safety cars.

Juggling all that is complicated enough, a task that got even more difficult at this year's Cook Out 400 at Richmond Raceway, when NASCAR added yet another variable to the mix: multicompound tires for the first time in a points-paying race.

Bolenbaugh said that his team was able to pull data and predictions from other racing series where multicompound racing is commonplace and quickly port that over to its NASCAR systems. "We were able to, the first weekend, have a competitive advantage and put our teams in the best position to be successful because of that," he said.

Austin Dillon, driver of the No. 3 Chevrolet Camaro ZL1, took the victory at Richmond.

See All 3 Photos3

The Human Element

While bringing AI and machine learning technologies like this to bear at different series is obviously a challenge, another is presenting that data to team bosses and drivers. Humans are inherently biased, trusting what's worked for them in the past, and they may or may not care when a little piece of software thinks they should pit.

"Our team works incredibly hard to provide the best capabilities that we can to our teams, but at the end of the day, the person making the decision is going to be held to account for the outcome,” Bolenbaugh said. “And so, while we usually have pretty strong alignment, there's definitely been some times where it's like, you know, I would say maybe frustration."

He said that it's ultimately another weapon in the team's arsenal, a tool that needs to prove its worth. And the good thing is that, with systems like this, there's plenty of opportunities to analyze how it's performing.

Bolenbaugh cited a truck race at Pocono, where a truck brushed a wall. The team was discussing whether they should pit to address any damage. One of the software engineers was able to use the AI tools to retrieve an image of the truck within seconds.

"By the time they came around again and had to make that decision, there was a picture in the crew chief space saying, 'There's no damage to the car. Don't pit,'" Bolenbaugh said. 

The Competition

While there may have been some initial reluctance, AI is now showing up in more and more pit-boxes as everyone wants to have every trick and tool the competition is deploying. Patrick Canupp, GM's director of motorsports competition, calls it "racer's paranoia."

"You'll develop some technology, and you'll think you've done a great job developing it in secret and that you're unique in it,” he said. “And you'll show up at the garage, and you'll be amazed to see that another team has done the same kind of thing."

"If we were doing all this and no one else was, I don't think we would have the level of competition and parity that we've seen like NASCAR and IndyCar and sports car," Bolenbaugh said about GM's AI efforts. "So, it's a big, big enabler, but it's also kind of a cost of doing business these days.”

Few teams want to talk about exactly how comprehensive their AI pit-wall assistants are, and indeed both Bolenbaugh and Canupp were cagey about some aspects of their own operations. But they were clear about one thing: This technology is designed to augment, not replace, providing more and better insight into the action for the human beings who actually run the cars.

"There's so much information flying at you that I think it is very difficult to process all of it optimally as just a human being. And so, you use these AI technologies, tools to help you process that data a little better than the next guy," Canupp said.

So, while fully AI-driven race cars are coming, for now at least, GM will keep the AI in the back seat.