It’s long, isn’t it? But that’s exactly the quantum leap that AI-based startup Neural Concept and its co-founder and CEO Pierre Baquet have made in just six years.
In 2018, the company’s fledgling software helped create the world’s most aerodynamic bicycle. Four out of ten F1 teams now use an advanced version of the same technology.
Along the way, Baquet’s company won contracts with aerospace suppliers such as Airbus and Safran. Obtained $9.1 million in Series A funding in 2022. Swiss-based Neural Concept, which currently employs 50 people, is working towards the Series B round and its software will help historic F1 teams like Williams Racing return to the pinnacle of the world’s premier motorsport format. It helps me find my way.
But where F1 cars rely on 1,000-horsepower hybrid V6 engines, Baquet’s first practical application of the technology was human-powered.
pedal power
In 2018, Baquet studied at the Computer Vision Laboratory at the Ecole Polytechnic de Lausanne, where he worked on applying machine learning techniques to three-dimensional problems.
“I got in touch with the guy who was leading this team and was designing the sixth or seventh generation of bicycles, and their goal was to break the world record for bicycle speed,” Baquet said. The man was Guillaume DeFrance and the team was IUT Annecy from Savoie-Mont Blanc University. The cycling team had already gone through six iterations of his bicycle design.
“Two days later, I came back to him in almost the same shape as the current world record holder,” Baquet said. Impressed, the team asked for further iterations. The result, Baquet says, is “the most aerodynamic bike in the world to date.”
This is a strong claim, backed up by multiple world records achieved in 2019. We’re not talking about airfoil-shaped downtubes or dimpled rims to reduce drag. This bike is fully enclosed, allowing cyclists to sweat inside a composite cocoon completely protected from the wind.
The core technology is a product called Neural Concept Shape (NCS). It is 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 an advanced software suite to perform his three-dimensional aerodynamic simulations.
CFD is much faster than sculpting a physical model and putting it into a wind tunnel. Still, this puts a lot of strain on the system and relies heavily on humans making good decisions.
At its core, NCS helps engineers move in directions they never envisioned while avoiding potential aerodynamic pitfalls. In “co-pilot mode”, the engineer can upload an existing 3D shape of her to provide a starting point, for example.
The NCS then takes a closer look at the neural network to suggest improvements, fixes, and possible paths for your choose-your-own-adventure 3D game. Human engineers then select the most promising proposals, undergo further testing and refinement, and repeat the path to aerodynamic glory.
It’s not just about “cheating the wind”
NCS is useful not only in racing, but also in the automotive and aerospace industries. “The road to widespread adoption in these types of companies is slow,” Baquet said of working in the somewhat conservative aerospace industry. “In this way, we started collaborating more with the auto industry, where the needs are even more burning and will soon change.”
Neural Concept has signed agreements with several global suppliers, including Bosch and Mahle. Aerodynamics is becoming increasingly important in the automotive world, and manufacturers are seeking ever more aerodynamic vehicles that deliver the maximum range possible for a given size battery pack. Masu.
But it’s not all about cheating the wind. NCS is also being used to develop things like battery cooling plates, which can be made more efficient to keep batteries at optimal temperatures without consuming too much energy in the process. “You can make a lot of money,” Baquet said. This means a wider range.
The ultimate testing ground for these technologies will always be the road, but the ultimate laboratory is F1. F1, a global motorsport phenomenon since 1950, 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 Formula 1 to a whole new audience. While the series focuses on the politics and drama between the teams, success on the track has much more to do with aerodynamics. That’s where neural concepts come in.
Baquet started watching F1 before Netflix was even on Reed Hastings’ radar. “I’ve always watched it since the days of David Coulthard and Michael Schumacher.”
Today, parts developed with the help of his company’s software are in operation at the pinnacle of this global motorsport. “It’s a great sense of accomplishment,” Baquet said. “When I started the company, I thought of this as a landmark, not just for F1, but for making parts designed using software usable on the road. And yes, this is… Every time it happens, it feels great, great.”
F1 is also a very secretive sport. Of the four teams Neural Concept works with, he was the only one who wanted to be identified as a client, and even that team was pretty tight-lipped about the whole process.
Williams Racing is one of the oldest teams in Formula One. Founded in 1977 by racing legend Frank Williams, his team was very dominant in his 1990s, winning his five constructors’ championships from 1992 to 1994, including his third straight victory. I won the world championship.
However, like most sports, success for F1 teams is cyclical, and Williams are currently in the midst of a rebuilding phase. The team finished the 2022 season in last place and finished in seventh place last year.
The NCS is one of the tools that will help Williams regain competitiveness. “We are using this technology in a variety of ways, some of which improve simulations, and others we are working on to deliver better results for the first time with CFD. “It helps,” said Hari Roberts, head of aerodynamic technology at Williams.
Again, CFD simulations are time-consuming and costly, a situation made worse by F1 regulations that limit teams’ testing capabilities. The physical time spent in wind tunnel testing is severely limited, as is the budget for computing time each team can use to develop the car.
Any tool that helps teams quickly bring their aerodynamic designs to life is a potential benefit, and NCS is actually very quick. Baqué estimated that his complete CFD simulation, which normally takes an hour, can be completed in just 20 seconds using NCS.
NCS is also largely exempt from F1’s strict restrictions because it is not performing calculations based on actual physics, but rather AI-driven guesswork based on a network of aerodynamic learnings. “The more knowledge you can extract and the more performance you can get with every CFD or wind tunnel experiment, you have a competitive advantage,” Roberts said.
But teams still have to pay for it. Baquet said NCS costs vary depending on team size and type of access, but typically range from €100,000 to €1 million per year. Considering that F1 teams also operate under a $135 million annual cost cap, this is quite a commitment.
Williams’ Roberts wouldn’t point to specific parts or lap time improvements thanks to the NCS software, but said it had an impact on the car’s performance. “This technology is being used as part of a toolset to develop the car aerodynamically. Therefore, lap times cannot be directly attributed to it, but it is a correlation and investigates new aerodynamic conditions. I know it helps with the speed.”
Beyond aerodynamics
The continued advancements in AI don’t stop there. There is also talk of an artificial agent on the pit wall controlling race strategy and machine setup.
“The growth of the AI/ML industry is exponential and these are exciting times,” Roberts said. “But this is also a real challenge facing everyone working in technology today: Which new tools will you spend your time researching, developing, and deploying?”
It’s not the kind of plot that captivates the average Drive to Survive viewer, but for many F1 fans, the behind-the-scenes of racing is the ultimate source of drama.
As for Neural Concept, the company continues to push deeper into the non-motorsports side of the automotive industry, developing more efficient electric motors, optimizing interior heating and cooling, and even crash testing.
Baquet said the company’s software helps engineers optimize a vehicle’s crash safety while removing unnecessary weight. But for now, the company can only run crash simulations for individual components, not the entire car. “This is one of the few applications that is hitting its performance limits,” he said.
Probably another application Fastest growing AI supercomputing platform in EU?