Everyone in the technology sector knows about the race to be the winner in artificial intelligence (AI). Less well-known outside the high-tech, high-performance world of Formula One (F1) is the challenge of using AI to win a race.
As its early adopters are learning, success with AI is all about data – and in F1, data is the way to deliver success. Formula One cars are packed with more than 300 sensors, recording every tiny variation that affects its speed – from aerodynamics to ride height to air temperature and pressure, vibrations, bodywork stresses, engine performance and the condition of the tyres.
Those sensors generate enormous amounts of real-time data – during a race, as much as 7MB (megabytes) every second; across a Grand Prix weekend, that adds up to about 1.5TB (terabytes) per car. And each team runs two cars.
It’s a sport that operates at incredibly fine margins – each team continuously developing their cars to cut a few more milliseconds off lap times. Every F1 team is focused on the same objective – going faster.
“What we try to do as a company is only one thing – to get the car faster. That’s the only thing this company does,” says Laurent Mekies, team principal at Visa Cash App Red Bull (VCARB), the F1 team formerly known as Alpha Tauri, or before that Toro Rosso – or, for the real F1 enthusiasts, Minardi.
“First, [that means] having a development rate as high as possible. Second, it’s about time to market, which here we call time to race. The first is known – if you speed up your car to go faster, you’re going to beat the other guys. The second one is a bit less known but has a mega impact in Formula One.”
Mekies says the gap in performance between the fastest and the slowest of the 10 teams on the F1 grid is smaller than it has ever been, and the need to find marginal gains is, as a result, greater than ever.
“Take an example – if you took the car we raced in Abu Dhabi, the last race of the [2023] season, and put that car back to Bahrain [the first race of the season] nine months [before], it would probably win the race. But we did not win Bahrain. So it’s very much about how fast can we develop and how fast can we bring that to the drivers.”
Advances in AI
Under the current regulations, all F1 teams operate under a cost cap of $135m for the 2024 season, with further limits on how that money is spent, such as the amount of time available for testing new aerodynamic designs in wind tunnels, or for drivers using race simulators.
Advances in AI have presented teams with a fresh opportunity to enhance their operations and performance at every stage of their business, from design through manufacturing to race day and competitive analysis
The advances in AI in recent years have presented teams with a fresh opportunity to enhance their operations and performance at every stage of their business, from design through manufacturing to race day and competitive analysis, as Mekies explained when Computer Weekly was invited to go behind the scenes at VCARB’s factory in Faenza, Italy, before a Grand Prix at the famous Imola track, the team’s local race.
“How do you structure a company to be the best at ‘time to race’? The backbone of that is ERP,” he says.
VCARB uses enterprise resource planning (ERP) software from Epicor – a strategic partner and sponsor of the team – to support every stage of the manufacturing and engineering process, from design through to the labour-intensive processes of building the car, laminating carbon fibre and fitting together more than 14,000 individual components to make one car. About 80% of those components are manufactured in-house – and are constantly being analysed and updated to deliver the slightest improvements that can contribute towards race speed.
ERP – a system common to every manufacturing company – may not seem the sexiest part of an F1 car, but here too, the data it holds is an AI goldmine. VCARB is one of the first Epicor customers to adopt the supplier’s new Prism generative AI (GenAI) tools.
Every manufacturing process that can be accelerated by using GenAI represents a further improvement in “time to race”. VCARB is initially targeting three use cases: to speed up coding for faster reporting; to automate the sending and receiving of requests for quotations from suppliers; and for natural language queries of the database (see box, How the VCARB F1 team is using GenAI to enhance its ERP platform).
A faster car
But how does that lead to a faster car? According to Guillaume Dezoteux, head of vehicle performance at VCARB, it’s all about reducing the time taken from identifying a favourable upgrade to the car to making it a reality.
“We look at the car data, we talk with the drivers, we see an opportunity for improving the car, we test it in a simulator, and then when we find a valuable upgrade, the rest [needs to be] very fast,” he says.
Getty Images/Red Bull Content Pool
“It would be interesting to use [AI] technology to detect patterns in your competitor behaviour. They have a race strategy, there is something going on. And you [could] have a means of predicting what your competitor may do. That’s one application that may be very relevant in the future”
Guillaume Dezoteux, VCARB
Dezoteux cites a recent example, where driver Daniel Ricciardo was having problems with the steering: “The steering feeling is a key parameter for the driver to gauge performance and the car balance. We’ve been working on that, trying different options to make the steering heavier, lighter, to change the parameters of the power steering we have on the car. And once we define a new target, then we drop it [to the factory team]. And then the time to market is incredibly fast.”
Bringing those improvements to the car one race sooner than would otherwise have been possible makes an enormous difference. “Everybody is developing, everybody’s improving their cars. And [this is] one way to make an additional opportunity. We’re talking about small differences. That’s why it’s so important for us that once we have defined our target, we have a very good time to market,” he adds.
With the budget cap, every decision and every possible upgrade to the car has to be assessed to optimise the combination of improvement potential, spend and “time to race”.
“You need to know which kind of development you would like to perform during the season,” says head of IT and innovation Raffaele Boschetti.
“It’s not a question of just saying, ‘I’d like this new floor [for the car]’. The point is you can do that if you have the budget, you have the resources, and the lead time is fine. So the benefit of this [GenAI] platform is that we’re really able to analyse this stuff and to understand if we can or not, depending on the way we are developing the car through the season, because the car is an R&D project – it’s never the same.”
Boschetti is already thinking of other ways GenAI can help the manufacturing process. For example, training an AI engine using images of parts to help identify potential defects in newly manufactured components.
Gaining an edge
Dezoteux is also excited about the potential for ERP-based AI to help gain an edge on track during a race.
“During a single race weekend, it’s difficult to find patterns in the behaviour of the car, or the tyres, or the interaction between the car and the driver [that take place] over a big amount of races,” he explains.
“We go to Imola [for example], we look at live Imola data, we analyse all that data, so we have a good understanding of what’s going on. But it’s difficult to find a pattern, if something that happens on the car may be linked to something that happened [in previous races], because the car was different then.
Monitoring the status of a car on the track is a challenge
“So monitoring the status of the car on the track is a challenge. [This is] where the ERP system is a fantastic tool because you have constant monitoring of what is the car configuration, you know how the car was at any time. Then [combining that] with the telemetry [from the race] to find a pattern is something that in the future will help us a lot.”
Of course, ERP is not the only area where AI can help a Formula One team like VCARB. As team principal Mekies explains, F1 has been a leader in automating data analysis for a long time, because of the vast volumes of data it generates.
“That enormous flow of data that is being analysed – how much of that is automated? Already a huge percentage – at least 70% to 80%,” he says.
“But there will be so many ways to better use this data if you’re able to create some intelligent analysis that can extract what you need to extract, or maybe can extract what you don’t know yet, but that you should be looking at. Have we been doing that for a long time? Yes. Is it exploding exponentially now? It is as well. We are discovering every day new ways to make good use of it.”
F1 regulations
Engineers are exploring ways that AI can help to alleviate the demands of F1 regulations that limit the amount of testing that can take place on car design and componentry during a season. For example, as well as limits on the use of wind tunnels, teams have restrictions on the number of hours of work they can complete using computational fluid dynamics (CFD) software, which helps to model the aerodynamic performance of the car.
Mekies describes CFD as a “virtual wind tunnel”, and with all the accumulated data across many hours of CFD use, AI algorithms offer an opportunity to provide the same answer without having further CFD runs.
“Because [the AI] already looked at 10,000 runs before, you can say – well, it’s already [analysed] that modification and can tell you what it’s going to do, so you don’t need to press a button to [complete a CFD run]. In terms of the regulations that’s interesting because we didn’t actually press the button,” he says.
AI already supports strategy during an F1 race. As data from the car – and information about rival cars – comes in, the team is analysing its options, such as when to change tyres or how to react to the introduction of a safety car, which slows down the race for a period of time.
“Strategic decisions are being taken on the pit wall, and that’s also AI-based software doing billions of calculations. The race is running and the machine is continuously analysing what happens,” says VCARB CEO Peter Bayer.
“Ultimately for these guys [making race decisions], they end up with one or two options which the human has to decide rather than 300 options. It’s quite fascinating to see that.”
For “these guys” too, AI is stimulating discussions about further ways of gaining an edge on the track that go beyond engine power and tyre performance.
“It would be interesting to use this kind of technology to detect patterns in your competitor behaviour,” says vehicle performance head Dezoteux. “They have a race strategy, there is something going on. And you [could] have a means of predicting what your competitor may do. That’s one application that may be very relevant in the future.”
Virtual sensors
Another potential application is virtual sensors. Those 300-plus sensors on the car that measure force, temperature, speed and so on, may be tiny, but they all add weight, and weight adds milliseconds to lap times. They’re also often expensive and can be damaged in a crash. With AI you can create a virtual sensor.
“So, while the [physical] sensor is fitted, the system is learning about its behaviour against the car parameters and [the AI] will find for itself what are the car parameters that are sufficient to create the behaviour of the sensor – then you remove the sensor and you just have a signal,” says Dezoteux.
“Currently it works, but the level of accuracy is not good enough. But in future, we expect that we could have a configuration of the car for practice that has more sensors and is more expensive. And then we make the car lighter, cheaper and simpler for the races.”
The VCARB team is not alone in exploring the use of AI – this is just one more area where F1 teams are constantly trying to outdo each other and find that extra little bit of speed that could make a difference on race day.
As a result, there’s also a new race going on, both between the teams and with the big tech companies – to capture the best AI talent.
“It’s important that F1 remains a place where these people want to come and that we don’t lose them to all the big tech companies that are also in a different race,” says Mekies. “So we need to make sure that we, as a sport, are attractive enough to get all these guys to want to come here to do the ground-breaking stuff that they want to do.”
In corporate IT, people talk about keeping a “human in the loop” where AI is introduced. But in Formula One, it’s still all about the human in the car. Might there ever be a day when AI can compete with the likes of Lewis Hamilton?
“The real answer is no. It’s not AI versus human, it’s AI to support the human. So the human layer both in motor racing and other applications still has that extra layer that you will not replace – you will just allow the human to concentrate on what they need,” says Mekies.
“Maybe it will help us in giving them the car they need, in the set of conditions they are in. So if it starts raining, do ‘ABC’ with your car settings. At the moment, some of the process is filtered manually by the engineers. Tomorrow, they will get more and more help from the live data of what’s happening on the racetrack to compute the changes they need to make to the car to support the driver better. But I don’t think it will tune our drivers’ characteristics.”