It’s a long way from pedal bikes to Formula 1. But that’s precisely the quantum leap that AI-based startup Neural Concept and its co-founder and CEO, Pierre Baqué, made in just six years.
In 2018, the company’s fledgling software helped develop the world’s most aerodynamic bicycle. Today, four out of 10 Formula 1 teams use an evolution of that same technology.
Along the way, Baqué’s company picked up contracts with aerospace suppliers like Airbus and Safran, earning a $9.1 million Series A raise in 2022. Now at 50 employees, Switzerland-based Neural Concept is working toward a Series B round while its software helps historic F1 teams like Williams Racing find their way back to the top of the world’s premiere form of motorsport.
However, where Formula 1 cars rely on 1,000-horsepower hybrid V6 engines, Baqué’s first practical application of the technology was human-powered.
Pedal power
In 2018, Baqué was studying at the École Polytechnique Fédérale de Lausanne’s Computer Vision Laboratory, working on applying machine learning techniques to three-dimensional problems.
“I was put in contact with this guy who was leading this team, designing the sixth or seventh generation of bike, and their goal was to break a world record of bicycle speed,” Baqué said. That guy was Guillaume DeFrance, and the team was IUT Annecy from the Université Savoie Mont Blanc. The cycling team had already gone through a half-dozen iterations of bike designs.
“Two days later, I came back to him with a shape that was almost looking like the current world record holder,” Baqué said. Impressed, the team asked for more iterations. The result was, per Baqué, “the most aerodynamic bike in the world at the moment.”
That’s a strong statement, but it’s backed up by multiple world records earned in 2019. We’re not talking about aerofoil-shaped downtubes or dimpled rims to reduce drag. This bike is fully shrouded, with the cyclist sweating away in a composite cocoon, completely sheltered from the wind.
The core technology is a product called Neural Concept Shape, or NCS. It’s a machine-learning-based system that makes aerodynamic suggestions and recommendations. It fits into the broad field of computational fluid dynamics (CFD), where highly trained engineers use advanced software suites to run three-dimensional aerodynamic simulations.
CFD is much faster than carving physical models and throwing them into wind tunnels. Still, it’s also hugely system-intensive and largely reliant on human beings making good decisions.
At its core, NCS helps engineers avoid potential aerodynamic pitfalls while pushing them into directions they might not have considered. In “co-pilot mode,” an engineer can upload an existing 3D shape, providing a starting point, for example.
NCS will then dig into its neural network to suggest improvements or modifications, possible paths in a 3D game of choose-your-own-adventure. The human engineer then picks the most promising suggestions and runs them through further testing and refinement, iterating their way to aerodynamic glory.
Not just “cheating the wind”
NCS is useful not just for racing but also in the automotive and aerospace industries. “The path to wide adoption in these kinds of companies is slow,” Baqué said of working within the somewhat conservative aerospace industry. “That’s how we started working more with the automotive industry, where the needs are a bit more burning, and they’ll be quick to change.”
Neural Concept secured contracts with several global suppliers, including Bosch and Mahle. Aerodynamics is increasingly key in the automotive world, with manufacturers searching for ever-more aerodynamic cars that deliver the greatest possible range from a given-sized battery pack.
But it’s not all about cheating the wind. NCS is also used in developing things like battery-cooling plates that, if made more efficient, can keep the battery at its optimal temperature without sapping too much energy in the process. “There are massive gains that can be made,” Baqué said, meaning yet more range.
While the ultimate proving ground for these technologies is always the road, the ultimate laboratory is Formula 1. A global motorsports phenomenon since 1950, F1 is currently experiencing an unprecedented wave of popularity.
The power of Netflix
The Netflix series “Formula 1: Drive to Survive” has brought the excitement of F1 to a whole new audience. While that series focuses on inter-team politics and drama, success on the track has much more to do with aerodynamics. That’s where Neural Concepts comes in.
Baqué started watching Formula 1 before Netflix was even a twinkle in Reed Hastings’ eye. “I always watched, since the time of David Coulthard and Michael Schumacher.”
Today, parts developed with assistance from his company’s software are running in this pinnacle of global motorsport. “It’s a great, great sense of accomplishment,” Baqué said. “When I started the company, I was seeing this as a landmark. Not only Formula 1, but just to have parts that were designed with the software on the road. And, yeah, every time that this happens, it’s a great, great feeling.”
Formula 1 is also an extremely secretive sport. Of the four teams that Neural Concept works with, only one was willing to be identified as a client, and even it was pretty tight-lipped about the whole process.
Williams Racing is one of the most storied teams in Formula 1. Founded back in 1977 by racing legend Frank Williams, his team was so dominant in the 1990s that it won five constructors’ world championships, including three in a row from 1992 to 1994.
But like in most sports, success is cyclical for Formula 1 teams, and right now, Williams is very much in a rebuilding phase. The team finished dead last in the 2022 season, rising only to seventh last year.
NCS is one of the tools helping Williams regain its competitive edge. “We use this technology in various ways, some of which improve our simulation, and other methods that we are working on will help deliver better results first-time in CFD,” said Williams Head of Aerodynamic Technology Hari Roberts.
Again, CFD simulations are time-intensive and costly, a situation compounded by Formula 1 regulations that limit a team’s ability to test. Physical time in the wind tunnel is heavily restricted, while each team also has a limited budget for computing time they can use to develop their cars.
Any tool that can help a team get its aerodynamic designs in shape quickly is a potential advantage, and NCS is very quick indeed. Baqué estimated that a full CFD simulation that typically takes an hour would take as little as 20 seconds through NCS.
And, since NCS isn’t running actual physics-based calculations but making AI-driven guesses based on its network of aerodynamic learnings, it’s largely exempt from F1’s draconian restrictions. “Anything we can do that allows us to extract more knowledge and therefore more performance from each CFD and wind tunnel run gives us a competitive advantage,” Roberts said.
But the teams still have to pay for it. Baqué said that NCS costs vary depending on the size of the team and type of access, but typically, it’s in the range of €100,000 to €1 million per year. Considering F1 teams also operate under a $135 million annual cost cap, that’s a substantial commitment.
Williams’ Roberts wasn’t willing to point to any specific parts or lap time improvements thanks to NCS software but said it has affected their car’s performance: “This technology is used as part of our toolset for developing the car aerodynamically. We, therefore, can’t attribute lap time directly to it, but we know that it helps our correlation and the speed at which we can investigate new aerodynamic conditions.”
Beyond aerodynamics
The ceaseless march of AI won’t stop there. There is talk of artificial agents on the pit wall calling the shots for race strategy and even car setups.
“It’s a fascinating time as the growth in the AI/ML industry is exponential,” Roberts said. “However, it’s also a real challenge that faces anyone involved in technology today. Which new tools do we devote time to exploring, developing, and adopting?”
That’s not the kind of intrigue that will captivate your average “Drive to Survive” viewer, but for many F1 fans, the race behind the race is the ultimate source of drama.
As for Neural Concept, the company is continuing to push deeper into the non-motorsport side of the automotive industry, working to develop more efficient electric motors, optimizing cabin heating and cooling, and even getting into crash testing.
Baqué said that the company’s software can help engineers optimize a car’s crashworthiness while stripping away unnecessary weight. But, for now, the company can only do crash simulations on individual components, not whole cars. “That is one of the few applications where we have been hitting the limits of performance,” he said.
Perhaps another application for the EU’s burgeoning AI supercomputing platforms?
