Making the Relationship between Uncertainty Estimation and Safety Less Uncertain
Vincent Aravantinos1 and Peter Schlicht2
1Autonomous Intelligent Driving GmbH Munich, Germany
vincent.aravantinos@aid-driving.eu
2Volkswagen Group Research Wolfsburg, Germany
peter.schlicht@volkswagen.de
ABSTRACT
Uncertainty estimation is an intense subject of research in object detection. A commonly mentioned motivation is the usage of object detection in safety-critical applications. However, the precise connection between uncertainty estimation and safety engineering is seldom done: how shall uncertainty estimation really support safety engineering? How does uncertainty relate to classical safety engineering activities? Which problems does uncertainty estimation solve? Which new challenges does it pose? This paper sketches those connections more precisely and provides perspectives for future work in this direction.
Keywords: Deep Learning, Uncertainty, Safety.
