The T1TENTH Grand Prix is not won by flooring the gas pedal, but with sophisticated computer software: the cars have to complete their laps completely autonomously, without human intervention during the race. The objectives are simple: "Don't crash and minimise laptime!" The artificial intelligence, which has been developed by the various teams for their respective vehicles, must therefore understand the track, avoid collisions and complete the laps as fast as possible. After a total of three rounds of competition, the "Scuderia Segfault" of the TU Wien came out on top as the overall winner at the IROS conference in Prague (27 September – 1 October).
Virtual last year, real this time
Last year's Grand Prix was already won by a team from TU Wien - however, due to the Corona pandemic, the Grand Prix 2020 took place purely virtually: Instead of driving physical model cars, the artificial intelligences competed directly on the computer.
This year was different: at the 2021 Grand Prix in Prague, it was possible to have real model cars compete against each other. They are modelled on racing cars, on a scale of 1:10. Cameras and sensors provide the necessary data, which then has to be interpreted in real time by the autonomous software and converted into control commands for the model car.
“The trajectory and velocity profile is optimized offline and is then used for racing as fast as possible”, says Luigi Berducci from the TU Wien team. “Using machine learning, a set of promising parameters are obtained, they are optimized using computer simulations and eventually validated on the real car.”
The improvisational talent of "Scuderia Segfault”
The TU Wien team, consisting of Daniel Scheuchenstuhl, Dennis Erdogan, Fabian Kresse, Felix Resch, Luigi Berducci, Moritz Christamentl and Stefan Ulmer, with the support of Andreas Brandstätter and under the supervision of Prof. Radu Grosu, spent weeks preparing for the competition. During the first round of the competition, there was only one vehicle on the track at a time. "Some problems appeared at first, but the team utilized their full problem solving capabilities and managed to improve the racing algorithm within very little time," reports Radu Grosu.
In the second round, the lap times were extremely close and the TU Wien team managed to come out on top: With a time of 18.70 seconds, they were just ahead of the runners-up (18.95 seconds) and were also able to complete the largest number of rounds overall in the allotted time.
Final Round: Head to Head
The final round of the competition was even more challenging: the fastest possible lap times had to be achieved while a competitor was on the track. In the process, a nasty crash occurred: the "Scuderia Segfault" vehicle was rammed violently by an opposing vehicle, the team had to repair the hardware in the shortest possible time. Fortunately, it succeeded, and the "Scuderia Segfault" was able to prevail against all other teams in a convincing manner and clearly won first place overall.
The acquisition of the F1Tenth cars for the underlying course on Autonomous Racing Cars was made possible through the BMWFW CPS/IoT-Ecosystem infrastructure award. Related research was also supported by the BMBWF DC-RES and the FFG ADEX project awards.
Prof. Radu Grosu,
Institute for Computer Engineering
+43 1 58801 18210