Master's thesis - Safety verification of Deep Neural Networks for ADAS/AD Systems

The development team for ADAS/AD innovations at AUDI AG conducts cutting-edge multidisciplinary applied research on safety assurance of learning-enabled autonomous driving systems with a particular focus on providing methods and tools for the development of Audi's next generation autonomous driving systems. Your involvement in the pre-development of automated driving at AUDI AG will allow you to gain experience in this fast-growing field and participate in the development of conditional and highly automated driving functions in an environment that is both scientifically oriented and practical.

The Audi Verifiable AI (AVAI) project aims at bridging the gap between AI and formal verification areas to support homologation of Audi's next generations autonomous cars. It primarily focuses on developing methods and tools to ensure safety and security of learning-enabled ADAS/AD systems

Deep neural networks have been frequently applied in various domains such as image classification; however, they can surprisingly be unstable with respect to adversarial attacks or perturbations. This has led to safety concerns of deploying deep neural networks to safety-critical systems such as autonomous driving. To account for the latter and as a part of your thesis, you will develop new concepts and algorithms to ensure robustness and safety of neural networks.