Would you watch Formula One if the cars were driven by computers instead of humans? That is the fundamental question asked by the Abu Dhabi Autonomous Racing League.
Also known as A2RL, the league comes from Aspire, a technology research and innovation hub that Abu Dhabi hopes will become the next Silicon Valley. With autonomous racing, Aspire also hopes to create not only a spectacle that replaces racing drivers with computer developers, but which could accelerate the development of roadgoing autonomous vehicles and their safety systems.
It’s easy to dismiss a form of motorsport that lacks even a hint of human jeopardy, or bravery. But AI racing is a proposition worth unpacking, since a skill mastered by humans still sits at the heart of what A2RL wants to achieve. “The competition will be in the brains of the engineers who programme the AI and the cars,” says Stephane Timpano, chief executive of Aspire.
The AI-driven car (right) on the grid at Suzuka with a human-driven car (left)
The league is made up of 12 teams from nine countries, including France, Germany, Italy, Japan and the USA, but not the UK, at least for now. Rather than compromising familiar names from the world of human-driven motorsport, A2RL teams mostly come from leading educational and technology institutions. The league says its ambition is to “attract some of the world’s brightest minds through the thrill of competition and a technological frontier, helping to develop autonomous technology for the wider mobility industry.”
Every team is given the same car and computerised driver. Made by Italian company Dallara, the car began life as an SF23, as used by Japan’s Super Formula race series, which sits just below Formula One in terms of performance. The SF23 is modified by A2RL to become a car called the EAV24. It features a body made partly from a sustainable flax-based composite and is powered by a turbocharged 2.0-litre, four-cylinder engine producing 500 horsepower.
The cockpit is home to a powerful computer that uses lidar, radar and seven cameras to see, a set of actuators to control the steering, accelerator, gearbox and brakes, and Nvidia RTX A6000 graphics cards for the artificial intelligence. There are sensors for tyre surface temperature and pressure, plus all of the sensors top-flight racing cars are normally equipped with, measuring everything from engine performance to the position of the suspension pushrods. All told, the system is producing 1 TB (1,000 GB) of data per minute.
The computer is housed in the cockpit of the modified race car
This vast stream of data is processed onboard by a computer that weighs 90 kg. It uses live data, GPS positioning and a pre-existing circuit data to work out where it is, where other vehicles are, and whether they can be overtaken; the latter is calculated between 20 and 40 times per second.
This data is cross-referenced with preexisting digital maps of the circuit. Given the accuracy of today’s racing video games, you might think this would be millimetre-accurate, but A2RL says the data it received of the Suzuka circuit in Japan was out by metres. Not letting that interfere, a team of engineers spent six weeks training the AI to tackle the Japanese race track.
Still very much a work-in-progress, A2RL isn’t yet a fully-fledged race series. The cars were last seen on-track during a demonstration at the Yas Marina circuit in Abu Dhabi in April, but struggled when a crash caused a yellow flag. This asks for no overtaking, but was complied with too literally, and the robo racers queued up behind the stricken car instead of overtaking it.
Fast-forward seven months and A2RL came to Suzuka to complete two public demonstrations during the final double-header weekend of this year’s Super Formula championship.
Attended by T3, the demonstrations aimed to firstly show how two AI-driven cars can share a circuit together, before a man-vs-machine duel where an AI car would take on a car driven by former Red Bull F1 driver, and current World Endurance Championship competitor, Daniil Kyvat.
The autonomous car successfully completed several laps of Suzuka Circuit.
The first demo went well, pictured above, with both cars successfully completing several laps of Suzuka, albeit at a pace some way off that of a human driver. This is partly because the AI-driven cars don’t yet know how to warm their tyres by spinning up the rear wheels or weaving left and right, as a professional would.
They also cannot drive over the curbs, as the vibrations would affect the car’s vision sensors and damage the computer. “We are working on designing a control technique for weaving,” said Dr Giovanni Pau, the team principal of TII Racing, a UAE-based team. “It looks simple for a human but for an AI driver it’s really crazy,” he added.
The cars lap the complex circuit confidently, pulling away from the start line and turning through corners just as a human driver would on a parade lap. During this first demo the cars appear fairly slow, but the team assures us they are saving their vehicles for a higher-speed demo in the man-vs-machine battle the following day.
To get to this stage is still impressive. Just a year earlier, pro driver Kyvat was lapping over three minutes quicker than the AI’s best effort; by the first race that gap had fallen to around 10 seconds, and today Pau says the car can operate at 92 to 95 percent of human performance. He also said how in December 2022 the car couldn’t go beyond about 20 mph – compared to over 100 mph today – and that the last few seconds of performance are the toughest to get. A2RL development driver Kyvat said: “The speed has increased massively from day one to now…The technology will contribute to the future of road safety.”
Away from the cars themselves, A2RL has plans to reinvent how fans engage with live motorsport from home. The league has developed a virtual reality app for the Meta Quest VR headset and an augmented reality system for the Apple Pro Vision. T3 found the latter was particularly impressive, with an AR environment showing a 360-degree onboard view from a car, plus live data, a broadcast TV feed, and a digitised top-down view of the circuit. Showing the live position of each car, it felt like a Scalextric track on your living room floor made real.
A2RL is developing an augmented reality app for the Apple Vision Pro headset
Although the cars are slow today, there is confidence that A2RL could genuinely pitch computer-driven cars against professional racers. “We will get very, very close to human performance, and at some point we may beat it,” Pau said, adding that it is his dream to create a race series where a grid of 10 humans and 10 AI-driven cars compete against each other, or even act as teammates. This is something Pau confidently says could be possible in the next two or three years, and after that he hopes to offer the car’s intelligence as a commercial product. In essence, it would be an AI for a road car that recognises when the driver has lost control, then steps in and takes over. Today’s stability control, but with AI baked in. “That is really where I think we will make a long-term impact,” he added.
Aspire CEO Stephane Timpano echoed this prediction, saying: “We want to use autonomous and robotics technology and AI to force new breakthroughs. We use these technologies in competitive environments like motorsport because it is the best way to prove to the world that they will have a great impact on everyday life.”
Unfortunately, during a warm-up lap ahead of the man-vs-machine headline act, the AI car spun at the exit of turn 11 and crashed into the barriers. An A2RL spokesperson said wheelspin was caused by “a loss of rear tyre pressure, possibly from a valve malfunction or debris on track.”
Lower track temperatures than the previous day, along with the car sitting on track for “an extended period” before the lap began, may also have contributed, they said. Due to a busy Super Formula schedule taking up the rest of the day, there wasn’t time to fix the car or try again using a backup.
Khurram Hassan, director of A2RL, told T3 as the crashed car was being recovered: “The simulation-to-reality gap is still a lot. You can do things on a computer screen, but this is why [real-world experience] is so important…We’re still a few months away from the car being able to warm the tyres up itself…We’ll learn from it and go on.”
Speaking to T3 before the crash, Pau emphasised, despite the rapid improvements, how primitive the self-driving technology still is: “Today I believe we are still in infancy. The child is still learning to walk. I have a dream of teaching the AI car how to control itself in very hard situations…at that point, an AI car could take over from a normal driver, such as when losing grip.”
There’s no doubt that the artificial intelligence used by A2RL will improve quickly, and that it will soon be able to warm its own tyres. But for now, the idea of a true human-vs-machine race feels a very long way away.
