A Methodology to Use AI for Automotive Safety Using Model Explainability

EasyChair Preprint 6847

Abstract

Workload consolidation, use of powerful processors and the advancement in Artificial Intelligence (AI) has enabled deployment of complex algorithms on high compute processors. Over time, the ability to derive effective algorithms using AI has increased dramatically paving their use in safety critical application. However, their adequacy from the perspective of safety certification remains controversial. In this paper we explain a methodology on how to use AI for safety critical applications using the techniques of explainable AI (XAI). The approach is explained taking a use case of traffic light detection as a part of active safety implementation targeting ASIL B. Recommendations from recent standards like SOTIF and UL 4600 are incorporated to augment the safety case.

Keyphrases: AI, DL, FuSa, ISO26262, SOTIF, UL4600, XAI, hara analysis

@booklet{EasyChair:6847,
  author    = {Srikanth Kaniyanoor Srinivasan and V Krishna and Harsha Vardhan Sahoo},
  title     = {A Methodology to Use AI for Automotive Safety Using Model Explainability},
  howpublished = {EasyChair Preprint 6847},
  year      = {EasyChair, 2021}}