Man Beat Machine in the First Human/Autonomous Car Race—But for How Much Longer?
May 02 2024

Man Beat Machine in the First Human/Autonomous Car Race—But for How Much Longer?

Standing trackside as cars scream by you at over 150 mph is a thrill. Whether they're big cars with fenders or little ones with wings, the rush of proximate velocity is the same. This past weekend, I learned that thrill applies not only regardless of the type of car, but even if there's nobody on board.

It was the inaugural event for the Abu Dhabi Autonomous Racing League, or A2RL, a historic test that not only gridded four autonomous race cars against each other on a road course but also pitted a human-driven car against an AI-powered one.

I was, of course, eager to see who would win, but my bigger question was whether deleting the meat bag behind the wheel would similarly neuter the fun factor. I'm surprised to report that the answer is no. This was one of the most interesting races I've ever watched, even if things didn't quite go according to plan.


The event took place at the Yas Marina Circuit, a purpose-built shrine to Formula 1, the wealthiest of motorsports arenas built in one of the richest nations in the world.

It's thanks in large part to the oil beneath that circuit that the UAE has the amazing resources it does, but as it strives to shift its exports from bubbling crude to sentient code, the UAE is making big investments in some fascinating things.

Abu Dhabi's Advanced Technology Research Council, or ATRC, was established in 2020 to push development in areas like artificial intelligence and autonomy. Both are being put to the test in an interesting way with the A2RL, while a $2.25 million purse provides the incentive.

Making Autonomy Interesting

We've all seen the robotaxi concepts, self-driving shuttle pods so anonymously styled, you can't tell whether they're coming or going.

The A2RL cars are far more compelling. Borrowed from Japan's Super Formula Championship, they mix a delightfully sculpted Dallara open-wheel chassis with a 2.0-liter, turbocharged four-cylinder engine. Those engines have been stolen from unsuspecting Civic Type Rs, then tuned to within an inch of their lives and augmented with only a notional exhaust, creating a 550-horsepower machine that weighs under 1,500 pounds and sounds shockingly good.

Formula 1 it ain't, but with pushrod suspension, composite brakes, and Yokohama Advan racing slicks, it's not far off. At Suzuka Circuit, Super Formula pole position last year was a 1:35, about seven seconds slower than Verstappen's most recent F1 qualifying lap.

However, A2RL cars have a significant advantage over those that compete in the Japanese series: They're not saddled with somewhere around 150 pounds’ worth of extra baggage in the form of a human driver. Instead, they’re piloted by a funny little box filled with hydraulics and enough circuitry to control them, and the articulation necessary to manipulate the car's physical controls.

The real magic is the software that controls these little robo-jockeys, plus the myriad sensors they use to see. A package perched on top of the cockpit, plus additional units scattered around the car's fuselage, encompasses lidar, radar, cameras, and the obligatory GPS—all powered by an array of GPUs offering enough floating-point performance to make the most hardcore of PC gamers wildly jealous.

Crucially, cars and sensors are the same for all the teams. Not even setup changes are allowed. Thankfully, the robot drivers have no mouths and so cannot cry about understeer at the limit.

To be clear: The cars drive themselves. The teams just look on nervously.

Finding Personalities

Delicious-looking cars with world-class performance are a great start, but anyone who's watched the explosion of F1 popularity in the wake of Drive to Survive knows that personality brings mass-market appeal.

For His Excellency Faisal Al Bannai, Secretary General of the Abu Dhabi's Advanced Technology Research Council and the real driving force behind this series, that was front of mind from the beginning. "When we announced the league, a number of people said, 'Where is the fun? We don't see the driver.' The fun becomes when we start seeing these programmers and developers," he said. "They are the guys that are that virtual driver of this car."

It seems far-fetched, people cheering for one coder over another. But if we look at F1 again, historically, there hasn't been much intrigue in team principles, either. Only outrageously colorful characters like Flavio Briatore or legends like Frank Williams earned any amount of attention. It's only recently that excessively profane team bosses like Geunther Steiner and Toto Wolff have earned their own cadres of fans.

So why not talented software engineers? Especially when they're as refreshingly diverse as many members of the eight teams who lined up to compete. Those few I had a chance to speak with had far more compelling and relatable backgrounds than your average F1 pilot these days.

That's doubly true if you look at the A2RL's impressive civic outreach programs. The American team, Code 19, has partnered with the Boys & Girls Clubs of Southeastern Michigan, giving students a chance to develop and race their own 1:8-scale autonomous model cars. Programs like this will open a whole new path to motorsports for the next generation and, perhaps more important, engender interest and excitement around more tangible careers within the broad spectrum of sciences, technology, engineering, and math: STEM.

But again, none of that really matters if the racing isn't good. The inaugural weekend was hit-or-miss.

The Teams

Code 19: This Indianapolis-based team was founded by a pair of military veterans and is staffed by a mix of students and full-time engineers.

Constructor: This German team with the utilitarian name is based at Constructor University in Bremen, Germany.

Fly Eagle: Splitting allegiances between China and the UAE, this team is based out of Khalifa University.

Humda: Hungary-based Humda is a non-profit focused on research.

Kinetiz: Like Fly Eagle, Kinetiz is a multinational affair, this time bringing talent from Singapore and the UAE.

Polimove: A part of Politecnico Milano, this Italian team has fielded successful teams in Roborace and the Indy Autonomous Challenge. Given that experience, it was one of the favorites.

TUM: This German team spun out of the Technical University of Munich. Like Polimove, TUM has experience in Roborace and the Indy Autonomous Challenge, winning that series' most recent round in January.

Unimore: Italian Unimore is part of the University of Modena's High-Performance Real-Time Laboratory and has previously raced in the Indy Autonomous Challenge.

And then there's the ninth team, team TII; the Technology Innovation Institute within the sponsoring ATRC, something like a factory team. TII helped to develop the car and the sensor stack and started testing well ahead of any of the other eight teams.

In other words, TII had a massive advantage. Given that, TII didn't take part in any of the team competitions. Its main event was the exhibition of human vs. machine on Saturday evening.

Getting Up to Speed

Even with your eyes closed, you can tell there's something different about the A2RL cars as they take to the track. On a warm-up lap, your typical human driver will vary speed, run through the gears, maybe even do a little swerving to cure boredom if not cold tires.

The AI-piloted A2RL machines don't do any of that. They fire their engines, idle down the pit lane exit, and then maintain a precise speed at a precise rpm. There's no human imperfection, just precise adherence to their software-defined run plan.

It's only during hot laps on warm tires that things start to sound normal, like any other race car being pushed.

The virtual drivers blast up and down through their six-speed gearboxes with the same aggression as a human, occasionally blowing a braking point and locking up a wheel or two. And, yes, more than once, cars got something wrong. An early incident saw a car spinning and getting stuck on track, requiring a flat-bed extraction.

A later incident saw the TUM and Polimove cars coming together. "We predicted the other to go faster into the turn. That's why we tried to go in behind them on the race line," TUM's team principal, Simon Hofmann told me. "They braked faster than we anticipated, and we rear-ended."

The damage was relatively minor, including a shattered rear radar sensor. That was trashed, but its alloy mounting bracket was unceremoniously worked back into shape by an Italian mechanic wielding a hammer, a refreshingly vintage moment amidst all the modern marvels.

The Main Events

The full competition was spread over all three days, with numerous events featuring multiple winners throughout the weekend, including the so-called "Pass and Defend" session where the TUM and Polimove cars came together. But it was a series of qualifying runs on Saturday morning that set the grid for the evening race, teams each going out and setting fast laps, the top four earning berths into the main event that evening. The result was:

  1. Polimove - 2:00.653
  2. Unimore - 2:01.314
  1. TUM - 2:01.864
  2. Constructor - 2:23.256

These laps were actually slower than those the teams had achieved earlier in the week, with Polimove setting a 1:57.8. A baking, midday track was surely a factor in the slower times (temperatures were approaching 100 that day), but you could feel the teams getting increasingly conservative as the main event approached.

There was only one spare chassis left between them, and by then, there was no time left to prep it.

Man vs. Machine

Though the crowd was much smaller than an average F1 race, the amount of fanfare was surprisingly on par. A trackside DJ spun tunes, dancers and other performers were everywhere, and roving reporters added hype at every opportunity.

And instead of Porsche Supercup or Formula 2 as support races, attendees were treated to a series of exciting, if a bit chaotic, drone racing events. Thanks to massive interference in the pit lane, the pilots were forced to fly their craft through a haze of digital static.

When the TII car hit the track, the grandstands along the main straight were remarkably full, and the crowd was well primed. The autonomous car ran a few installation laps before driver Daniil Kvyat joined it on the track, his cherry-red machine easily identified thanks to the presence of a halo hoop above the cockpit—and, of course, a helmet within it.

Kvyat took a lap or two to get acquainted with things, tracking down the autonomous car. Then he darted to the inside on the back straight and was through. As far as passes go, it wasn't the most exciting, but the significance was massive. Humanity had proven its worth for at least another year.

There was a huge cheer from the crowd, the sort you’d hear in the grandstands at Le Mans or Indianapolis after a pass for the lead. The people were genuinely into it.

Kvyat pulled to the side of the track, and the TII car passed him without hesitation. Kvyat then shadowed his competitor through the rest of the lap, blasting by it again. This time the pass was on the front straight, in front of the bulk of the crowd, who again let him hear their approval.

The Main Event

The primary race was set as eight competition laps of the circuit for the top four cars. However, we learned later that two of those laps would be run under yellow flag conditions, with reduced speed and no passing allowed, presumably to give everyone a chance to make sure this would work.

I wasn't sure it would. The cars struggled to maintain the same, reduced pace even on those slow laps. There was nothing of the nose-to-tail (or, in the case of Lance Stroll, nose-under-tail) posturing you see in an F1 race. In the final corner, the lead car slowed to bunch things up.

The TUM car, in third place, seemed to get confused by the sudden reduction in speed. It came to a standstill, front wheels twitching with indecision, while Constructor in fourth place sat patiently behind. TUM had been one of the fastest cars all weekend, and it seemed like it might not even feature.

Finally, it moved, all four cars moving again, but again, the last two cars struggled to keep pace. On lap four, the leading Polimove car was finally allowed to go, screaming down the front straight with the Unimore car in hot pursuit.

The two ran close for much of the lap until the Polimove car came charging too aggressively into a tight left-hander. Its rear wheels locked, its tires probably too cold after all that idling around in the chilly desert evening. It spun wide, and the Unimore car ducked to the inside, taking the lead.

The race officials threw a yellow flag immediately before the third- and fourth-place cars arrived at the scene. Both came to a stop on the track, presumably because they're not allowed to pass under yellow. The Polimove car tried to turn itself around, to get going again, but it lacked space.

The action ground to an unceremonious halt.

Take Two

The Polimove car was trailered back to the pits, and the cars lined up to try again, but only the top three runners headed back out on track. Polimove failed to restart.

Unimore was in the lead, TUM now in second, and Constructor in third. It was to be a six-lap race now. Constructor struggled to keep up during the warmup and was quickly distanced when things went green again.

It was an exciting start. TUM was seemingly quicker but playing it safe behind the Unimore car. But, in Turn 5 (the same place Verstappen got Hamilton in 2021), the Unimore car misjudged the corner and ran wide. The TUM car dove to the inside for the lead. It was a clean pass, one that genuinely elicited a cheer out of me.

Sadly, I was one of the few. Much of the crowd had headed home by this point amidst the confusion between the first race attempts. Shame, as they missed a truly exciting pass plus a little bit of history, not to mention a hell of a fireworks show afterward.

Next Steps

One thing that was abundantly clear throughout the weekend was how vital preparation was. Polimove qualified first with a time of 2:00.653. The new team Flying Eagle's fastest lap was over eight minutes.

The rookie American team, Code 19, was closer to the sharp end, with a fastest lap of 3:05. That may seem disappointing, but members of the other teams I spoke of were amazed they were that quick.

"It's pretty astonishing how new teams that have never done this before can perform," TUM's Simon Hofmann told me.

Why so impressed? Because while Code 19 only got started earlier this year, other teams like TUM and Polimove had years of experience running, and winning, in Roborace and the Indy Autonomous Challenge.

The deck was stacked against the new teams, but they made some amazing strides. Back in April, when Code 19 started testing at the Abu Dhabi Circuit, it ran lap times of well over six minutes. Two months later, it was a minute off the best.

At this point, nobody knows exactly what the next step is for the A2RL, but organizers have committed to more events, meaning we'll see these cars running again. Kvyat clearly wasn't challenged, and the fully autonomous race featured more technical faults than fine technique, but the pace from nothing to here for many of the teams is remarkable, and it’s only going to improve.

I'll put my neck on the line here: Assuming this series continues to sponsor this level of autonomous competition, within five years, Kvyat, or his future replacement, will be looking a little less smug in the pit lane.

The bigger test will be whether the series can survive that long. Can it drive enough fan interest to bring more sponsors and more teams to the table? Only time will tell on that front, but the packed grandstands full of cheering fans on the day of the event gave me reason for optimism there—despite all the hiccups.

Source :