Now showing 1 - 1 of 1
  • Publication
    Assured Resilience in Autonomous Systems - Machine Learning Methods for Reliable Perception
    ( 2024) ; ; ;
    Schwaiger, Maximilian
    Machine learning in the form of deep neural networks provides a powerful tool for enhanced perception of autonomous systems. However, the results of such networks are still not reliable enough for safety-critical tasks, like autonomous driving. We provide an overview of common challenges when applying these methods and introduce our approach for making the perception more robust. It includes utilizing uncertainty quantification based on ensemble distribution distillation and an out-of-distribution approach for detecting unknown inputs. We evaluate the approaches for object detection tasks in different autonomous driving scenarios with varying environmental conditions. The results show that the additional methods can support making the perception task of object detection more robust and reliable for future usage in autonomous systems.