IAV and TU Munich Bring AI to the Racetrack

Science fiction in real-time

On 26 September 1982, the first episode of the now cult television series “Knight Rider” flickered onto the screens of American televisions. Since then, countless fans across the world have rooted for the heroes of the science fiction action series: Michael Knight and his talking car K.I.T.T. The latter was a black Pontiac Firebird Trans Am, which was equipped with artificial intelligence and could think, speak, and drive on its own. Today, nearly 40 years later, vehicles networked with artificial intelligence are no longer science fiction. It’s actually quite the opposite, as a cooperation project from IAV and TU Munich called “rAIcing” demonstrates.

My house, my car, my algorithm

Those involved in “rAIcing” have ushered in the next phase of automated driving. “As part of the research project, we have developed two novel software modules for automatic decision making in racing situations”, reports Joao Graciano, Function Developer for Autonomous Driving. “With the Decision Manager (DM), a rule-based concept was pursued, one which is modularly expandable by criteria.“

These criteria are rules, which are necessary for driving on a racetrack – like “do not collide“, “stay on the track” or “drive as fast as possible”. A racetrack places high demands on vehicles, requires the quickest reaction time, and therefore offers the perfect conditions for putting limits to the test – even those of artificial intelligence. But is AI-based software in a position to learn and apply extreme handling? Yes, the innovation project shows. After all, according to Joao Graciano, the development of the second software module revolved around just this question. “This module makes decisions based on a supervised-learning algorithm, which can learn a specific racing style based on training data.”

Research continues

The rAIcing project ended in April – but not the research into its subject matter. Thorsten Scheibe, Head of Department AD Validation & Automation, draws a definitive conclusion: “Being able to autonomously perform maneuvering decisions is among the most important functions in autonomous vehicles. At the same time, AI algorithms offer enormous potential. Combining and researching these two aspects was simply exciting. It gave us deep insight into an essential technical field and generated important know-how, which can now be applied to other projects.” And that’s exactly what’s happening. With the “Directive-Based Decision Manager” (DDM), the Decision Manager has already been enhanced for autonomous driving on the highway. AI-based approaches will be further investigated and demonstrated in other projects.