Evaluating Self-Adaptive Architectures for Automated Driving Systems

Self-adaptive methods have been advocated for addressing challenges related to managing unknowns and uncertainties in autonomous driving, which in turn are caused by, e.g., machine-learning uncertainty, operation in an open context, and cybersecurity. Many works proposed specific vehicle architectures featuring self-adaptation mechanisms. However, each work tackles specific problems often using different levels of abstraction making the approaches hard to compare and making it even harder to compile complementary features into a single architecture. This paper systematizes the body of knowledge in self-adaptive vehicle architectures by proposing evaluation criteria based on the available literature (standards and research paper) and on identified desiderata for self-adaptive vehicles. We proceed to carry out a detailed analysis of selected papers based on the evaluation criteria in the context of the Level 3 Highway Pilot vehicle feature. The paper concludes by pointing out research gaps which we believe will foment future work on self-adaptive vehicle architectures.