If Nobody Is Driving, What Exactly Are We Watching?

The Abu Dhabi Autonomous Racing League, known as A2RL, brought five fully autonomous racecars to the Autodromo Internazionale Enzo e Dino Ferrari as part of the ACI Racing Weekend. In a release issued Sept. 7, the league described the event as Europe’s first five-car autonomous race and its own first competition outside Abu Dhabi. The race covered 12 laps, with the UAE-based Kinetiz team taking the victory ahead of Constructor Racing of Germany, PoliMOVE of Italy, Unimore of Italy and Germany’s TUM. The finishing order looks conventional enough to fit onto any motorsport results sheet, but the machinery underneath the result changes the meaning of the competition.

A2RL says all five teams competed with identical EAV-25 hardware, an important constraint because it shifts the performance difference away from the familiar contest over engines, aerodynamics or mechanical packages. Instead, the organizers describe the differentiators as software: perception, planning, control and real-time decision-making. In other words, every car entered the same physical argument with a different way of seeing the track, predicting what might happen next and choosing what to do about it. The race was not simply a demonstration of autonomous vehicles moving quickly in the same place. It was a contest between competing systems of judgment.

That distinction became visible even though the software itself did not. Unimore started from pole and set the race’s fastest lap at one minute, 40 seconds before a technical issue brought its car to a sudden halt. PoliMOVE, which A2RL says reached a top speed of 252.3 kilometers per hour while running close behind, could not avoid the slowing Unimore car and struck it from the rear. The incident ended both teams’ races and elevated Kinetiz into the lead, which it held to the checkered flag. Two-time champion TUM had already encountered trouble on the formation lap and returned to pit lane before the rolling start. Autonomous racing, it turns out, can still produce mechanical frustration, traffic, contact, opportunity and the strange luck that has always inhabited racing.

What it does not produce in the conventional sense is a driver making those decisions in real time. There is no helmet camera capturing a glance toward an apex, no radio exchange revealing hesitation, no late-braking move that can be attributed to courage or impatience, and no post-race interview in which a competitor explains why the car was placed half a meter to the left at exactly the right moment. The decisions still happen, but they occur inside software that has been developed long before the green flag. Human agency has not disappeared from the sport; it has moved backward in the chain of causation.

That relocation may be the most culturally significant part of autonomous racing. In traditional motorsport, the engineer and driver collaborate across time. Engineers build the conditions for performance, and the driver interprets those conditions under pressure. In autonomous racing, the engineering team must attempt to encode part of that interpretation itself. The car has to identify another competitor, estimate its motion, determine whether an opening is real, choose a trajectory, manage grip and speed, and keep revising those decisions as the environment changes. The skill being tested is no longer located entirely in the cockpit because the cockpit has become an execution point for decisions designed by a distributed group of people and expressed through code.

That does not make the result less human. It may actually reveal how much human work is usually concealed inside the word 'autonomous.' A2RL says the teams prepared for Imola through a compressed nine-day period of physical testing in conditions that included rain and hail, while its broader development program included more than 5,000 hours of simulation testing and racing in 2025. Engineers built perception systems, refined control models, validated behavior in digital twins and tried to anticipate the situations the cars might encounter before they encountered them. When Kinetiz crossed the line first, nobody had been holding the steering wheel, but plenty of people had been building the decisions that eventually moved it.

This is where autonomous racing becomes more interesting than the increasingly common spectacle of attaching the letters 'AI' to an existing activity. A racing circuit is a deliberately hostile place for a decision system. Speed compresses time, competitors create moving obstacles, grip changes, passing lines disappear, technical failures occur without warning and small mistakes produce immediate consequences. A2RL calls the series a public testbed for autonomous technology, and the description is useful so long as it is not mistaken for proof that a racecar and a road car face identical problems. Public roads contain pedestrians, cyclists, intersections, construction, ambiguous human behavior and legal obligations that a closed circuit does not reproduce. Racing offers a specialized stress test, not a complete model of ordinary transportation.

Competition still has value precisely because it is unforgiving. A system that performs beautifully while circulating alone may reveal different weaknesses when another car occupies the line it expected to use. An algorithm that is stable in simulation may behave differently when sensors, weather, vibration and physical hardware introduce uncertainty. Racing compresses those problems into a format designed to produce failure quickly and publicly, which is why motorsport has long served as a laboratory for mechanical innovation. Autonomous racing extends that tradition into software by making the quality of a decision system measurable not only by whether it completes a lap, but by how it behaves when other decision systems are trying to beat it.

The harder problem may belong to the audience. Sport is not only competition; it is legibility. Spectators understand what it means when a goalkeeper guesses correctly, a pitcher misses a location, a driver locks a brake or a runner attacks too early because the human action can be connected to a visible person. Software complicates that relationship. A spectator can watch an autonomous car make an overtake, but the meaningful action may have occurred in a planning model, a confidence threshold or a control strategy that remains invisible unless the broadcast finds a way to explain it. For autonomous racing to become more than an engineering demonstration, it will eventually need a language that allows ordinary viewers to understand the differences between the competitors without requiring them to become robotics researchers.

That challenge is not entirely new. Modern motorsport already asks fans to care about invisible systems. Formula 1 audiences discuss tire degradation, energy deployment, brake balance, aerodynamic wake, simulation and strategy models despite seeing only fragments of those processes directly. Teams routinely transform abstract engineering choices into personalities and narratives that spectators can follow. Autonomous racing pushes the abstraction one level further because the decision-maker itself is no longer sitting inside the car. The series will have to make software behavior as narratively legible as a driver’s style: aggressive or conservative, opportunistic or patient, resilient in traffic or fragile under uncertainty.

There is also a philosophical wrinkle hiding inside the word 'athlete.' Calling the software the athlete is tempting because the software performs the racecraft, but it may be too simple. Code does not write itself, select its own objectives or decide what counts as success. The more accurate competitor may be the entire human-machine system: engineers, researchers, simulators, sensors, control architecture and vehicle acting together. Traditional motorsport already works this way, of course, but the driver has historically provided the face that allows the audience to compress a large technical organization into a single recognizable protagonist. Autonomous racing removes that shortcut and forces the team itself to become the visible competitor.

That could ultimately change what victory means. A Kinetiz win at Imola does not tell us that an artificial intelligence experienced pressure, bravery or satisfaction. It tells us that one team’s integrated system navigated the particular problems of that race more successfully than the others. The achievement belongs to design, preparation, reliability and encoded judgment. If that sounds less romantic than the image of a driver wrestling a car through Tamburello, it is because motorsport has spent more than a century teaching us where to look for the human drama. Autonomous racing is asking us to look somewhere else.

Imola therefore matters less because five driverless cars completed a race than because the event made a new kind of competition visible. The cars still accelerated, defended, failed, collided and crossed a finish line, but the decisive intelligence had been distributed across people and machines before the race even began. Nobody was driving in the old sense. Somebody was still competing. The fascinating part is that we are only beginning to decide what to call them.

Event: A2RL autonomous race during the ACI Racing Weekend at Imola.

Race notes