Self-drive to survive: motorsport as a testbed for AI driving tools
What went down on track?
If you were one of the circa 600,000 viewers who tuned in to watch the race live, you might rightly feel sceptical about the future of AI’s role in motorsport. Although a landmark event in the advancement of AI technologies, there's no getting away from the fact that the event itself was a bit of a damp squib.
Most of the cars had been coded to race conservatively, rarely pushing hard and, in one embarrassing instance, forming an orderly queue behind a car that had stalled on track rather than performing an overtake. While the few brave souls who had coded their race cars to, well, race got to witness their cars’ AI brains misjudging braking zones and coming into contact with the competition.
An advert for modern self-driving systems it was not, particularly when comparing the fastest AI-driven lap of the weekend (Team Polimove’s lap time of 1 minute 57.8 seconds) with the benchmark time of 1 minute 47 seconds set by former Formula 1 driver Daniil Kvyat.
But some managed to shave off 20 seconds per lap in the week preceding the race, and top teams estimate that they are improving their pace by 60% per month. So, despite a disappointing headline result for the weekend, the inaugural A2RL race showcased the promise of self-driving technologies, and the potential for AI systems to surpass the natural intelligence and responsiveness of humans in the near future.
That's not to say that Max Verstappen and Lewis Hamilton are about to be made redundant. A recurrent criticism of the race weekend was the distinct lack of personality, jeopardy and flair – the human element that makes sport compelling. And despite their rapid development, I suspect we are still some way away from imbuing AI systems with personality (or a sufficiently realistic approximation thereof). So the human athlete looks here to stay for the foreseeable future.
Motorsport drives innovation
But motorsport has long been at the forefront of innovation in the automotive sector, with many of the sport’s bleeding-edge innovations and developments trickling down and making their way into tomorrow’s road cars.
Whether it be the active suspension pioneered in Nigel Mansell’s championship wining 1992 Williams FW14B, paddle-shift gearboxes, keyless ignition, or modern hybrid powertrain engines, there are plenty of features of the modern road-car that can be traced to the research and development facilities of motorsport teams looking to gain a competitive advantage over their rivals.
With a view to the introduction of 2026’s engine regulations, F1 teams’ current focus is on maximising the efficiency and power output of the V6 turbo-hybrid engine, which will be run on close to 50:50 split of electrical power and sustainable fuels. Formula E meanwhile, as its name suggests, continues to progress the development of all-electric powertrain technologies. While Toyota has eschewed a formal return to either Formula (at least as an engine manufacturer), instead committing to a 5-year development focus on the zero-emission hydrogen engine technologies showcased at 2023’s 24 hours of Le Mans. The drive for sustainable combustion is undoubtedly (and quite rightly) top of the agenda, and the technologies being pioneered trackside today will undoubtedly inform the design of next generation road car engines.
But speak to anyone working in the sector today and they will tell you that the push to realise the potential of self-driving technologies is not far behind. And, if the dramatic rate of improvement witnessed throughout the A2RL race week is anything to go by, it isn’t too great a stretch to imagine Team Polimove’s code being licensed and integrated into a future update of Tesla’s “Autopilot” systems or other proprietary autonomous vehicle technologies. That, rather than the admittedly lacklustre on-track action, is what proved so exciting about the launch of the A2LR project; the technologies being advanced and their broader practical and commercial applications.
How Motorsport can solve one of AI’s teething problems
While there are still knotty legal questions to be answered over AI technologies (such as where liability lies in the context of road accidents involving self-driving vehicles), the fundamental concern holding back the wholesale adoption of self-driving technologies today is safety. Do owners feel comfortable stuck in the passenger seat, and do manufacturers feel comfortable placing them there?
A lot of the discourse around self-driving technologies has revolved around how AI brains would respond when faced with complex moral questions like the Trolley Problem. But a less-talked about source of concern has been whether or not AI systems have the processing power and responsiveness to take evasive action at all.
In March 2023, WhatCar? conducted a comparative test of the AI driver assist tools available in 10 commercially available road cars. When placed in an emergency stop situation at 30mph, 9 of the 10 cars tested were able to take steps to avoid a collision. However, when this test was replicated at 42mph the results were reversed, with only 1 out of 10 cars successfully avoiding a collision, and 2 out of 10 failing to react at all to the impending collision. The test was not replicated at speeds approximating motorway conditions, but it is unlikely that participants would have fared better at higher speed.
By contrast, while the A2RL race did not proceed without incident, the cars on track were running at speeds far greater than tested by WhatCar? and, for the most part, took evasive action where required.
The systems operating the A2RL cars were processing, and acting on, environmental data with greater speed and accuracy than commercially available systems, and were improving at startling pace. With a prize pool of $2,250,000 up for grabs, the A2RL’s inaugural race provided a very real incentive for entrants to maximise the responsiveness and reactivity of their AI systems while working to an economically viable cost restriction. And looking ahead, the data and knowledge amassed by these teams will undoubtedly have downstream commercial applications in terms of both the performance and efficiency of AI systems.
A2RL’s next initiative will be an autonomous drone racing championship due to conclude in April 2025, with further prize money up for grabs to incentivise development. And although it is unlikely to supplant Formula 1 in terms of spectator interest and commercial viability, it is a welcome addition to the broader stable of motorsport and the automotive development landscape.
If you have an automotive, sporting, or technology query that you would like to discuss, please contact Bradley Howe.