Safety Concerns and Mitigation Approaches Regarding the Use of Deep Learning in Safety-Critical Perception Tasks

Deep learning methods are widely regarded as indispensable when it comes to\ndesigning perception pipelines for autonomous agents such as robots, drones or\nautomated vehicles. The main reasons, however, for deep learning not being used\nfor autonomous agents at large scale already are safety concerns. Deep learning\napproaches typically exhibit a black-box behavior which makes it hard for them\nto be evaluated with respect to safety-critical aspects. While there have been\nsome work on safety in deep learning, most papers typically focus on high-level\nsafety concerns. In this work, we seek to dive into the safety concerns of deep\nlearning methods and present a concise enumeration on a deeply technical level.\nAdditionally, we present extensive discussions on possible mitigation methods\nand give an outlook regarding what mitigation methods are still missing in\norder to facilitate an argumentation for the safety of a deep learning method.\n