How Artificial Intelligence Changes Racing as we know it — a Visit at Roborace
Claudia Buck, Manager Chassis Innovation and Strategy at Porsche AG and Daniel Bareiss, Innovation Manager at Porsche AG explain what the development process of AI driven racing cars teaches us about AI. They contemplate whether or not software developers will be the new race drivers of tomorrow and about the change that happens right in front of our eyes right now.
Ever since we first heard about this company called Roborace that builds driverless AI racing cars, we knew we had to see it in action at some point — change was in the air. A few weeks ago, we finally had the opportunity to do so and joined the final track test for Roborace’s Season Alpha.
But first, the basics. Roborace is an autonomous-vehicle racing series that combines fully-electric race cars with artificial intelligence, the first series of its kind in the world. Roborace was established to “accelerate the development of autonomous software by pushing the technology to its limits in a range of controlled environments.”
In 2018, one of Roborace’s driverless vehicles, the RoboCar, completed the first-ever autonomous hill climb at Goodwood Festival of Speed. Using only computer vision, sensors and artificial intelligence algorithms, the vehicle successfully navigated the famous 1.86-km hill climb track in West Sussex, England. Besides the stunning maturity of the prototype, it also looks as pioneering as Roborace’s vision: RoboCar was designed by the German designer Daniel Simon who is probably most famous for creating iconic automotive vehicles for Hollywood blockbusters such as Tron: Legacy and Captain America: The First Avenger.
A racing league for human and AI drivers
In the winter of last year, Roborace unveiled its latest prototype, the DevBot 2.0. Roborace’s racing cars are powered by Nvidia Drive, an autonomous vehicle development platform, and four electric motors that generate a combined 500-plus horsepower. In contrast to its unmanned counterpart Robocar, DevBot 2.0 is equipped with a cockpit for a human pilot. This now allows human pilots to race together with AI drivers. The AI “driver” is, in this case, the intelligent software that gathers all data from the sensors and other touchpoints to “drive” the car. What is more, digital maps of the environment can be created by manually driving the car, meaning that the human drivers can push the car to edge and teach the AI driver where the limits are.
The second-generation all-electric vehicle made its first public appearance at the start of Season Alpha earlier this year at Monteblanco Circuit in Spain. Season Alpha is Roborace’s inaugural racing competition, which premiered just this year and takes place at various locations in Europe and North America. It involves three challenges: wheel-to-wheel, object avoidance and localization. As its name suggests, it is currently in its alpha stage, with a beta planned for next year. Season Beta will feature more teams and increasingly ground-breaking formats and challenges being trialed across the globe.
The localization challenge, or: the battle of algorithms
To understand, explore and manage the complexities inherent in writing software for autonomous vehicles, Roborace has partnered up with our colleagues at Data:Lab Munich and Italdesign. For a period of four months, Roborace, Italdesign and Data:Lab worked together to develop new Machine Learning functionalities regarding perception, trajectory planning and torque vectoring. Why? Machine Learning is extremely important for autonomous vehicles, of course. Satellite navigation is a core technology for automated driving systems in this respect. Yet, sometimes GPS signals are blocked by tall buildings, tree tops or other obstacles. This poses a significant challenge to automated driving systems, and this is where the Roborace localization challenge comes in.
The challenge marked the end of Data:Lab’s four-month collaboration with Roborace — and we were more than excited when we received the invitation to join this milestone!
How to race autonomously without satellite navigation
In the beginning of August, three teams — Arrival, Technical University Graz and University of Pisa — competed against each other in the localization challenge at Zala Zone, a proving ground in Hungary. In this competition, all teams use the same vehicle, Roborace’s DevBot 2.0, but develop their own software and artificial intelligence algorithms to drive the car.
Each team had 30 minutes to complete the tack, with three timed lap attempts. The cars had to avoid obstacles and stay on the race track, completing the lap as fast as possible. They weren’t allowed to use satellite navigation, though. Instead, they had to rely on cameras and sensors to execute software and navigate the race track. The teams used LIDAR (light detection and ranging) and IMU (inertial measurement unit) sensors to measure distances, identify lane markings, and map the road ahead as well as optical speed sensors to detect the rotational wheel speed of the vehicle.
A moment of change right in front of our eyes
The first time we encountered DevBot 2.0 was rather impressive: Usually you have the human driver sitting behind the steering wheel in case something unfortunate will happen. But for the final test, the car stopped and the driver got out of the car, giving the car a little ‘tap on the shoulder’. Once the track was cleared, DevBot 2.0 continued racing, without a human behind the wheel. This moment was an eye-opener to us, we have just realized that there’s actual change happening right in front of our eyes — racing, or to be exact, driving is changing fundamentally. Right in front of our eyes.
For several decades, motorsport enthusiasts have been entertained by the rivalry between racing drivers. When it comes to Roborace, though, the real competition is carried out between algorithms and software developers. This leads to the question: Are software developers the racing drivers, our heroes, of tomorrow?
This transformation in racing us somehow comparable with e-sports. And yet, its relevance extends far beyond motorsport and e-sport!
Spurring innovation: the transfer from autonomous race cars to road vehicles
For carmakers, racing has always been a way to advance their own research and development. Competitions like 24 Hours of Le Mans have helped to generate innovative breakthroughs. And Roborace is no different. It has the potential to change the world outside of motorsports — and has already inspired a multitude of innovative approaches. Moreover, on-road testing involving automated driving systems has been confined mainly to highly controlled environments. Roborace isn’t only about localization or how the AI software perceives its environment, it’s also about how the vehicle behaves at its maneuvering limits. Roborace takes autonomous driving to an extreme level and provides valuable real-world experiences and training data, which help developers to improve AI driving systems.
We are happy that we were able to experience a small part of Roborace’s history in Hungary — thanks for letting us be part of this!
Now, let’s see where this is leading us. What do you think: Are developers the race drivers of tomorrow? Do we need a concept and racing series like this?
Claudia Buck is Manager Chassis Innovation and Strategy at Porsche AG and Daniel Bareiss is Innovation Manager at Porsche AG. For more details, get in touch with us on Twitter (Porsche Digital Lab Berlin, Porsche Digital), Instagram (Porsche Digital Lab Berlin, Porsche Digital) and LinkedIn (Porsche Digital Lab Berlin, Porsche Digital).
