The Abu Dhabi Autonomous Racing League (A2RL) is an international championship created by ASPIRE, part of Abu Dhabi’s Advanced Technology Research Council (ATRC). A2RL is a professional driverless motorsport series that fields fully autonomous Super Formula race cars, capable of exceeding 280 km/h. Beyond the racing itself, the championship is designed both as a platform to accelerate research in AI, robotics and advanced vehicle-control technologies, and as a way to showcase the future of mobility and its possibilities to the widest possible audience.

In 2025, for A2RL’s second season, 11 racing teams from all around the world gathered at Yas Marina Circuit in Abu Dhabi. Rapson, built around a Hungarian engineering team, competed alongside entrants from Italy, Germany, France, the United Arab Emirates, Singapore, China, Japan and the United States.

During race preparation, Rapson installed its own in-house reinforcement-learning (RL) controller on its A2RL race car. Under this RL-based control, the car performed controlled, low-speed manoeuvres on the circuit. To the team’s knowledge, this was the first time globally that a reinforcement-learning controller had been applied to such a high-performance single-seater formula car in the physical world; until now, experiments of this kind had essentially been limited to simulation.

Reinforcement learning is a self-learning AI technique in which the system improves itself in simulation: it tries different actions in different situations and receives positive or negative feedback (rewards or penalties) based on the outcomes, gradually converging on better and better strategies. In Rapson’s case, the AI “driver” completed millions of training episodes in simulation before ever touching the real car, effectively learning how to control a race car on its own. The milestone lies in the fact that Rapson then took this trained AI driver out of the virtual world, put it into a real race car, and actually handed over control on a live circuit.

The significance of RL is underlined by the fact that, in recent years, it has been the technique behind several “superhuman” performances, mostly in virtual domains – from AlphaGo and AlphaStar to GT Sophy. Rapson believes that, in the physical world, after drones, high-performance racing – and, through it, road traffic – can be the next domain where this capability truly comes to life.

This first on-car demonstration is therefore a key step in Rapson’s long-term vision: to develop superhuman motion intelligence – AI with proven race-pace capabilities – that can gradually be transferred from the racetrack into everyday road vehicles.

The weeks of preparation and qualification before the race presented Rapson’s engineers with a wide range of challenges. They had to get to know the car as a race car, understand the behaviour of each subsystem, layer their own decision-making and control software on top of the baseline stack provided by A2RL, and then optimise it all for the unique layout and conditions of Yas Marina Circuit

To qualify for the main event, teams had to meet three criteria:1. respond correctly to race-control flag signals. 2. achieve a lap time within a strict window of a given reference lap. 3. complete a high-speed overtaking manoeuvre in a controlled way.

Of the 11 teams on the grid, only seven managed to pass all the tests and earn a place in Grand Final qualifying – Rapson among them, lining up alongside some of the most experienced outfits in the field for three qualifying runs.

During the second qualifying run, however, Rapson’s car suffered a crash. As a result, the team was unable to record a valid lap within 115% of the reference time (the fastest lap of the fastest team), which meant they did not advance to the multi-car Grand Final. Instead, Rapson competed in the Silver Race, where the team ultimately finished fourth.

Rapson’s A2RL programme is supported by a focused group of partners who help turn ideas from simulation into real laps on track – from deep engineering expertise, through the infrastructure needed to train its AI driver, all the way to cybersecurity.

Together, Bosch Group, GanzKK and Riptides form an ecosystem that enables Rapson to put its AI driver on the starting grid of the world’s largest autonomous racing series.