Today's Edition
MAYS IBRAHIM (ABU DHABI)
Reigning Abu Dhabi Autonomous Racing League (A2RL) champion TUM heads into the series' first international race backed by software advances that have made its autonomous car faster, safer and better prepared for wheel-to-wheel competition.
A2RL will make its overseas debut at Italy's Imola circuit on September 5, with up to five fully autonomous Dallara Super Formula SF23 racecars taking on one of motorsport's most demanding tracks.
The event marks a milestone in the league's expansion beyond Yas Marina Circuit after two seasons in Abu Dhabi.
TUM's biggest breakthrough has been cutting software latency from about 500 milliseconds to roughly 100 milliseconds, allowing the car to process information and react far more quickly at racing speeds.
"When it comes to speed, latency is the key," TUM team lead Prof. Dr.- Ing. Markus Lienkamp told Aletihad.
At speeds exceeding 250km/h, he said, delayed decision-making makes competitive racing impossible.
Rather than one major innovation, the gains came through a series of software refinements and a redesign of the car's planning system.
TUM replaced graph-based routing with sampling-based trajectory planning, allowing the software to assess about 1,000 possible trajectories during each decision cycle. A PhD-led optimisation project then reduced computation time by a factor of three.
Safety Before Speed
Beyond outright speed, TUM has increasingly prioritised safety in multi-car racing. Lienkamp said the software predicts how rival cars are likely to behave and will voluntarily surrender track position if uncertainty rises and collision risk becomes too high.
"The point is not to be fast. The point is to make no crash," he said. "If you want to win a race, you have to finish first. But first you have to finish."
The system models whether competitors are likely to remain on the racing line, attempt an overtake or deviate under pressure. When confidence falls, the car automatically reduces risk by lifting or moving off-line.
TUM has also developed software that gradually pushes performance while remaining within defined safe limits.
The car raises its pace lap by lap, continuously checking stability and track limits. When risk increases, it automatically backs off.
To prevent spins at the limit, the software incorporates automatic countersteering – similar to a human driver's instinctive correction – and a race-adapted electronic stability control system that briefly brakes the outer front wheel to regain control.
According to Lienkamp, the approach paid off during the previous season.
"Last year, we didn't have any scratch on the car. No spin, no scratch, nothing," he said.
Race Prep Relies on A2RL's Track Simulations
Teams are preparing for Imola through A2RL's simulation ecosystem, including its Sim Sprint programme across digital versions of circuits such as Yas Marina, Suzuka and Imola.
Lienkamp explained that simulation was valuable for refining race strategy and testing multi-car interactions, although tyre behaviour and aerodynamics remained difficult to reproduce with complete accuracy.
A2RL, organised by ASPIRE under the Advanced Technology Research Council, has expanded its virtual racing programme to more than 5,000 hours of combined testing in previous seasons.
Across two seasons in Abu Dhabi, A2RL has also tracked rapid progress in autonomous performance relative to professional drivers.
The gap between AI systems and humans has narrowed from around 10 seconds per lap in 2024 to just 1.58 seconds in 2025.
Last year, TUM also participated in a controlled Human vs AI demonstration against former F1 driver Daniil Kvyat, with both sides operating under non-interference rules for safety.
Lienkamp said the team is now aiming to close the remaining performance gap and potentially move toward full competitive overtaking scenarios in future head-to-head formats.
For Lienkamp, A2RL's value extends beyond competition - shared cars, simulation tools and operational support also make the series accessible to university teams with limited budgets, he said.
