Inside The Cockpit-Free Car: How A2RL’s EAV-25 Racecar Has Redefined High-Performance Autonomy

On November 15, 2025, the grand finale of the second season of the Abu Dhabi Autonomous Racing League (A2RL) saw six fully autonomous racecars take to the Yas Marina Circuit in the UAE capital, with the event—organized by UAE-based ASPIRE—setting a new record as the world’s largest autonomous racing event to date.

Central to this edition of A2RL’s fully autonomous racing series was the EAV-25, the racecar used by all of its contestants. A2RL provides all competing teams with the same hardware and basic software, which allows them to focus solely on developing artificial intelligence (AI) that’d get the racecars do everything from setting fast lap times to outsmarting opponents on track.

A2RL called the EAV-25 a major leap forward from the EAV-24, the race car that was used in A2RL’s inaugural season, which, by the way, was built by converting a Super Formula SF23 into a fully autonomous racecar. While the base vehicle for the EAV-25 remains the Dallara SF23, the autonomous system inside has been completely overhauled. “The EAV-25 was designed as a pressure test for autonomy, where milliseconds matter and the margin for error is zero,” Alexander Winkler, Head of Sporting at A2RL, told Inc. Arabia. “Every feature, from its upgraded sensor fusion architecture to its sensor redundancy safety systems, was developed to operate under extreme load and environmental unpredictability.”

Alexander Winkler, Head of Sporting at A2RL.

One of the most significant upgrades to the EAV-25 lies in the vehicle’s autonomous stack, which was developed in partnership with Steer AI, a UAE-based tech venture that converts standard industrial vehicles into autonomous units using AI mobility systems. “Steer AI has been instrumental in helping us push the boundaries of perception and system integration,” Winkler said. “Together, we’ve developed a fully modular autonomous stack that not only processes environmental data, but actively informs motion planning and control. Their technology allows for ultra-low-latency decision-making; blending Light Detection and Ranging (LiDAR), radar, and camera data in microseconds.”

Winkler also noted that while A2RL had already collaborated with Steer AI in its inaugural edition last year, Season 2 marked a clear step forward in their partnership. “What’s new this season is that we’ve expanded the collaboration beyond perception into behavior modeling and control logic,” he said. “That means Steer AI’s systems now help translate perception into intent, planning optimal trajectories, executing overtakes, and managing high-G manoeuvres in ways that sometimes diverge from human intuition, revealing new insights into control and precision.” 

So, what does all this mean for the future of racing? From Winkler’s perspective, A2RL is proving that both on track and in simulations, race engineers in the future will need to master both pit strategy and Python. “The future race engineer is a hybrid profile,” Winkler said. “Mechanical setup, racecraft, and strategy still matter, but performance now also depends on how well teams understand perception, planning, and decision-making algorithms. You don’t need every engineer writing code full-time, but you do need people who can read Python, understand Robotic Operating System (ROS2) behavior, and diagnose how an autonomous stack thinks at high, medium, and low speeds. In our environment, software decisions are performance decisions.” 

“We’re already seeing the best results from teams where motorsport engineers, roboticists, and AI researchers work as one unit,” Winkler continued. “That model will only become more common because autonomy at the limit is no longer a niche research problem—it’s a competitive edge. The collaboration between coders and race engineers is expanding traditional roles, and creating a new category of talent that understands both the physics of the car and the logic of the agent driving it.”

A scene from A2RL Season 2.

As for what A2RL’s achievements translate into for vehicles off the racing track, Winkler tells us that the league provides a contained and compressed testbed for autonomous mobility. “The breakthroughs developed on the A2RL platform translate directly beyond racing,” he said. “High-speed perception, multi-agent decision-making, and safety-critical autonomy are exactly the capabilities needed in autonomous logistics, industrial robotics, advanced driver-assistance systems, and smart-city digital twins. Racing compresses years of edge-case exposure into a single season, making it a powerful accelerator for real-world autonomous mobility.”

According to Winkler, the biggest takeaway from A2RL is that building for extremes helps autonomous solutions perform better in everyday life. “We’re finding that systems trained to handle complex, high-speed environments tend to demonstrate greater resilience when adapted for lower-risk, real-world scenarios, like managing traffic intersections or autonomous deliveries in mixed environments,” he pointed out. “Moreover, A2RL’s ‘race-to-road’ approach is redefining the way industry validates reinforced learning, prediction, and artificial intelligence. Instead of waiting years for lab simulations, we compress hundreds of real-world edge cases into every lap, turning each race into a live stress test for safety, reliability, and performance.” 

But there is more at stake than technical hains alone. “On the racetrack, you’re not just building speed, you're building trust,” Winkler pointed out. “The same algorithms that enable an autonomous car to overtake at 260 km/h can help us understand how  AI can better predict, react, and self-correct—insights that could later inform systems for urban mobility, logistics, and smart infrastructure.” 

Pictured in the lead image is the EAV-25 racecar used by A2RL Season 2's winning team, TUM. All images courtesy A2RL.