View Categories

AI – Generation of Specific Scenarios of Interest

As aforementioned, increased automation raised the need to complement ISO 26262 with the safety of the intended functionality standard ISO 21448 (SOTIF), dealing with hazards because of functional insufficiencies in the presence of system failures. However, due to the typical failure modes and performance limitations of ML, an absolute level of correctness is infeasible. 

This question wants to deal with the fact that data used to train an ML model rarely covers all possible scenarios the model will face when put into production. Here, we focus on the scenarios, and to validate the safety (even new and unsafe) of ADFs, based on a multi pillar approach including simulation, virtual testing, test track testing and real-world testing. In order to validate the ML system to be within acceptable error margins, some references papers are (Henriksson et al., 2023), (Burton et al., 2022) and (Yang et al., 2021). 

Main Question

Has the generation of specific scenarios of interest been considered?

Sub-Questions

  1. Can “edge-cases” scenarios be foreseen?
  2. Can unsafe scenarios be identified for the AI system?
  3. Does the model improve the performance for the specific corner cases?
  4. Have specific techniques been considered, in particular allowing for risk mitigation in ML components and aiming to reduce the false positives? (see references in the text)

References

  • Henriksson, J., Stig Ursing, Erdogan, M., Warg, F., Anders Thorsén, Johan Jaxing, Örsmark, O. and Mathias Örtenberg Toftås (2023) ‘Out-of-Distribution Detection as Support for Autonomous Driving Safety Lifecycle’. Lecture notes in computer science, pp.233–242. doi:https://doi.org/10.1007/978-3-031-29786-1_16.‌
  • Burton, S., Hellert, C., Hüger, F., Mock, M. and Rohatschek, A. (2022) ‘Safety Assurance of Machine Learning for Perception Functions’. Deep Neural Networks and Data for Automated Driving, pp.335–358. doi:https://doi.org/10.1007/978-3-031-01233-4_12.
  • Yang, J., Zhou, K., Li, Y. and Liu, Z. (2021) ‘Generalized Out-of-Distribution Detection: A Survey’. ArXiv (Cornell University). doi:https://doi.org/10.48550/arxiv.2110.113