The development of technologies enabling self-driving cars has received widespread attention in recent years. The anticipated benefits of this technology are tremendous, as it could not only increase road safety and productivity of commuters, but also, for example, enable door-to-door mobility for disabled people. While the number of self-driving cars on the road is rapidly increasing as more and more research projects enter real world pilot testing phases, the question of how to systematically test and eventually certify self-driving vehicles in the future remains an open problem and an unsolved challenge. It is non-trivial to proof that a self-driving vehicle drives more safely than a human in general. Even harder it is to demonstrate that it is capable of handling all the difficult corner cases that it may encounter in long-term deployment. In addition to that, self-driving vehicles employ more and more artificial intelligence not only for tasks such as the detection and recognition of potential obstacles, traffic signs, and other traffic participants, but also for interpreting and reasoning about the traffic situation as a whole. These functional modules based on artificial intelligence pose their very own additional challenges with respect to testing and certification, as they are no longer programmed explicitly, but rather deduce their behaviour implicitly from large amounts of training data instead.
Therefore, novel and innovative ways are needed to address the problem of certifying self-driving cars. They need to be agnostic to the various vehicle platforms, as well the different soft- and hardware configurations, need to be efficient, effective, reliable, and, last but not least, practical. Take on the challenge and come up with a smart way to send self-driving cars through the driving test of the future!
Challenger | Mathias Bürki, PhD Candidate Autonomous Systems Lab,
Mathias has studied Computer Science and Robotics at ETH Zurich and currently conducts his doctoral studies at the Autonomous Systems Lab lead by Prof. Roland Siegwart. His fields of interest are mobile robotic perception, computer vision, and visual self-localization of autonomous ground vehicles. As a researcher, he has participated in the European FP7/H2020 research projects V-Charge and UP-Drive, which both address autonomous driving in urban environments.