A field of fully autonomous racing cars will take part in an inaugural championship race at Yas Marina in Abu Dhabi in April 2024.
On the morning of the 2023 Abu Dhabi Grand Prix, at a technology campus not far away from the Yas Marina Circuit, the covers were taken off a Dallara Super Formula SF23 development car ahead of its racing debut in April 2024.
But no driver can climb behind the wheel of this machine – there’s no steering wheel, for a start. Nor is there an actual cockpit. Instead of where a driver would normally squeeze themselves in, the car is filled with sophisticated sensors, computers and technology that will allow it to drive itself. And with the car able to learn from its on-track endeavours, it will also be able to turn itself into a highly-tuned racing driver…
How does A2RL’s autonomous racing car work?
Named the Abu Dhabi Autonomous Racing League, A2RL has been put together by Abu Dhabi-based company ASPIRE and combines artificial intelligence and mechanical autonomy in order to develop a racing series that will see driverless machines go to battle against each other.
Ten teams, put together from university entrants and research institutions, have already signed up for the championship, and will fight for a prize purse of $2.25 million.
All will be given a fully autonomous Dallara Super Formula SF23, with each team then being able to adapt their software in order to go racing – the nature of how to approach the racing and the level of risk-taking is entirely up to how each team wants to participate.
Speaking to PlanetF1.com in an exclusive interview, Enrico Natalizio – chief researcher of ARRC (Autonomous Robotics Research Centre) at the Technology Innovation Institute (TII) – explained how the car interprets the racetrack.
“[There are] seven cameras, with three LIDARS for radar,” he explained, showing a picture on his phone of the complex electronics encased within the ‘cockpit’.
“The idea is that the vehicle has to be able to perceive itself inside an environment with all the objects and the things that are around. So each of these elements positioned in strategic places in the car gives a 360-degree view of what’s happening around and can fuse this data all together.
“What happens is that all this information is mapped and fused into an engine of perception, to extract information and features that are relevant for the rest of the control stack.
“After that, there is a planner and controller that receive this information, the planner decides which action to perform and passes the command to the controller – that is the one that is in charge of making the motion of the car, steering the wheels, to accelerate, to decelerate, etc.
“The competition itself is on all these elements together as a system, but on each of them. Having better-performing LIDAR or better-performing radar or whatever, it gives you a better perception.
“So all these elements are given the same way to the competitors. The thing on which they will compete on is the software that they will put on, extracting the information from each of these sensors altogether but also how to feed this to the perception stack in order to fuse them and extract the relevant information. How this is done, it depends on the team.
“What data they consider more relevant or less relevant, and what weighting of that information?
“What an F1 driver does – an experienced driver anyway – is not totally ‘OK, I’m gonna over[take] the car [in front] now.
“It’s what is after that. Is there a curve? Is it the best moment to do it, or maybe should I wait to stay behind and overtake when I can?
“The global planner decides this kind of thing, whereas the local planner decides at the moment the behaviour to do, according to the global plan, to move out from the trace of the guy that is in front.
“Then there is the strategy of the team because it’s very similar to the personality of the driver. You can give the car the strategy of being very aggressive, trying to always use the trajectory that is the most highly rewarding at the risk of losing control.
“Or you can give a more conservative one and say: ‘No, just stay on the easiest thing, don’t risk anything’.
“It’s kind of the personality of the car, this can also be programmed. The car can also learn from the mistake of being too aggressive, for example, or being too conservative and then trying to have a higher reward in making a quicker laptime.”
Autonomous racing cars take away fear, but also intuition
Given that a real-life driver spends their time feeling the car through their body, constantly providing inputs and micro-inputs to control and correct, how is it possible that an autonomous machine can respond with the same level of ‘feel’, given there’s so much more to car control than mere visual input or awareness of objects.
For instance, how does the car respond to a slide on a slippery surface that isn’t expected?
“There are several features that, from the perception, they transform these features that become relevant for the planner,” he said.
“So, for example, slide a little bit on the kerb, it’s perceived by the car because it’s not only the camera and the sensors and the radar – there are sensors in the wheels and all over the body of the car – to understand the distance from the ground and the temperature and the pressure on the tyres and everything.
“So all this data is collected. From all the sensors, we have understood that this is the pattern that happens.
“The other thing is what the car is able to do, directly with the data that has been considered already relevant and that’s been taken out for deciding what the planner will do.”
Part of the competition will be helping the car to learn as quickly as possible, which is the human element of the competition.
Natalizio explained that TII, handling the automatisation of the Abu Dhabi water taxis, the boat had never encountered waves on the water as the AI had only been trained on flat water. The boat considered the small waves as objects, with the boat then trying to avoid them – a futile exercise.
As a result, similar to a driver needing to train in the rain, it’s imperative the car is exposed to as many variables as possible in order to allow the AI to learn.
From there, it’s down to the weighting given to the different neurons based on the situation, meaning that some cars can be better at different skillsets.
Put simply, a team could develop a ‘Rainmaster’ car that might not be as outright fast in dry conditions as another, while another team may have a more aggressive AI car that is less adept in the rain.
“What we have done is to let it drive on the Autodrome at 50kph just to start learning, to start understanding what is around, and to start being able to plan the next section,” he continued.
“Of course, we had to run all the tests for the safety part, for the flags, the Race Control, and all the things that are like Formula 1.
“But the thing is, once ASPIRE hands the car to the competitors to push and say, ‘OK, now it’s not 50kph anymore’, it will be able to do that. Now it’s up to you, whether you want to push more, you want to be more aggressive with the kerbs, or want to take it easy – it’s up to the competitors and the car learns.”
But the intriguing part of the technology is the removal of the human weaknesses that take out considerations like distractions, emotion, and fear.
After all, no matter how good a driver is, just how close are they really to the outright maximum potential of the car? In reality, no driver can do that – but machines could. How far away is that day?
“The idea is to get to a point that we can do a superhuman competition, in the sense that we can race against the human drivers and try to beat it,” he said.
“This happened already for drones, for example, racing drones. There is an international competition, where the AI beat the best pilot of the drones.
“The researchers are working to be in the same exact condition of the driver, for beating it. So this car, the human-driven car has been tested for the setup and so on by Daniil Kvyat.
“He has been working with Aspire to test the human-driven car to push it to the limit. We are not yet at the level of competing with an F1 driver, but the idea is that the car will learn.
“The car has all the choices and there is a good and a bad. The bad is you strip off the emotions from the car in the sense that there’s a spectrum of choices to do every millisecond or even less, and it has to decide what is the best in that moment.
“The more rational you can be in those moments, what we always highlight in drivers is that they are cold as ice, right?
“So the car can be very cold, and it doesn’t have this emotion of choosing the thing just because of fear, or any other elements that can disturb. You’re taking the fear out of the equation.
“On the other side, what you’re doing is stripping off the intuition and the creativity of the driver. You have predetermined the behaviour that can be learned and can be improved constantly.
“That’s true. But at the same time, you don’t have that creativity of doing something that is completely out of the box. We will reach there too. Generative AI is pushing towards that boundary as well but, it’s, let’s say, more regulated.”
Another element to consider is that real-life drivers frequently have to compensate for things happening with their cars that may take it out of its prime operating window. For instance, damaging a front wing means a fundamental handling change.
But this is not a problem, Natalizio explains, saying the sensors can learn from mistakes on the trajectories imposed, and will thus adapt quite quickly to the issues encountered on track.
For bigger problems, like losing a gear or even losing a wheel, Natalizio laughs – there’s no need to consider the safety of the ‘occupant’.
“This now is aimed more at pushing the performance rather than to be safe,” he said.
“Formula 1, it’s much more important to be safe, and performance is the second objective. For us, it is reversed, we want to push the car as much as possible. Of course, if it gets destroyed, it’s just a car. It doesn’t involve human lives and the safety comes after that.”
Are there concerns AI racing won’t capture public imagination like ‘gladiatorial’ F1?
But while the autonomous racing league promises some thrilling racing and mishaps (after all, there is the chance every team programmes their car for hyper-aggressiveness and the inaugural race just has all the cars barrelling into each other with no concessions made), the fact of the matter is that A2RL lacks the gladiator aspect of racing.
After all, for the vast majority of F1 fans, the thrill is the human aspect – the Verstappen vs. Hamilton battles, the Senna vs. Prost intrigue, the spiciness of drivers coming to terms with their own inadequacies when lined up against stronger team-mates etc.
When you’ve got computers all capable of 100 per cent accuracy and erring on the side of logical decision-making, does it risk being a damp squib for viewers?
“I was on the same wavelength,” Natalizio admits.
“But then, getting closer to the team that was preparing the car, the human effort that is behind it, all that is done by the car is human in the end. It is a human that has programmed.
“So it’s true that the element of decision in the corners to do that instead of that will be on the pilot, but all the rest of the work is on the team.
“The tension and the level of conflict that this can generate and so on is super emotional. So I think it will switch a little bit more toward the team. But still, there’s a human component who will be there.
“Not in a race, in that moment, but in the preparation of the race, in the outcomes, assessment, and so on. But it will be a more 360-degree experience for someone who wants to approach this world.”
For the first race and the foreseeable future, ASPIRE will serve as the governing body and commercial entity overseeing the series.
But, like Formula E did under Alejandro Agag over a decade ago as the FIA helped with the organisation of the first all-electric series, might A2RL eventually look to partner up with F1’s governing body?
“We’ve talked about building ecosystems and building collaborations, so we don’t rule things out,” said Tom McCarthy, executive director at ASPIRE.
“But I think there is a sense in which we need to spend a bit of time seeing what that kind of ecosystem that we’re developing looks like rather than jump under the umbrella of another ecosystem.
“It’s important for us that we understand what the driving logic is because, if we jump under someone else’s driving logic, then we may miss the opportunity. There will be a certain sense of independence where we grow and develop.
“I think it’s important that we mature and clarify, in practice, our driving logic. That is a logic that is about an evolving ecosystem, how OEMs get involved, and so forth like that – that’s really important for us. So, down the road, you might see that there would be an advantage for dual organisations doing things jointly.
“But I think it’s important for anything like this, given its broad and innovative nature, not to jump in too early, and go in a way that might cut off the driving logic to the kind of industry it might become.”
With McCarthy getting ready to head to the Abu Dhabi GP himself as we reach the end of our interviews, the Irishman explains how he sees A2RL as being sufficiently different a spectacle for viewers that it doesn’t need the gladiatorial aspect.
“It’s completely different from the type of media strategy that you have for F1 today. It’s gladiatorial,” he said.
“What percentage of people attending a race today will have a huge insight into the Constructors’ Championship, or the technology of the cars? They care about the press conferences and what they’re saying, and that’s something autonomous racing won’t have.
“Until we have primadonna coders that end up getting beaten! I could tell you, ‘I’m going to take the coders and focus on that’ – we will do some of that, but that won’t be for creating a psychodrama between them.
“That will be more for giving heroes to the kids to see it: ‘Hey, wow, this is what they’re doing with their career’.
“It’s a gamified experience that has a real activity happening. We have thoughts down the road where, let’s say, we try to have participant activity where online people can do things that have to be dealt with by the car on the track so that you have interaction with the car on track, but that’s off in the future.
“It’s a very different type of experience, but you become the gladiator yourself ultimately. That’s a different participation. Is the future of F1?
“Absolutely not. I mean, who would want not to have cars with humans in them going at speeds like that?
“It’s great for science development.”