Evaluation of a Fail-Over Mechanism for 1oo2D Architectures in Highly-Automated Driving

While self-driving cars show remarkable progress in their autonomy capabilities, they are still lacking a necessary and sufficient level of dependability. As a consequence, todays self-driving cars require a human operator to monitor the car for potential safety violations and to take over control in critical situations. The challenges on the way to a truly dependable and trustworthy self-driving car are manifold. One of the challenges is the design of an appropriate fault-tolerant architecture. In this paper we investigate the 1-out-of-2 with Diagnostics (1oo2D) architecture paradigm for highly-automated driving and study the fail-over mechanism in detail. We present an implementation based on industrial techniques and technologies and evaluation results of fault-injection studies.