Artificial intelligence definition and specifics of its application for automated road vehicles
Document GRVA-09-23
1 February 2021
Submitted by Russia
Download document
Previous Documents, Discussions, and Outcomes
4. (b) | Deliverables of the Informal Working Group on Validation Methods for Automated Driving
12. (b) | Artificial Intelligence

113. The expert from the Russian Federation introduced GRVA-09-23 suggesting definitions (taken from the International Standard Organization) for Artificial Intelligence (AI) with relevance for wheel vehicles, their subsystems and parts. He proposed to consider that AI is considered as a software i.e. a design feature of an automotive product and that therefore no design requirements should be developed. He mentioned the impossibility to verify the performance of AI systems. He also suggested that any risk mitigation may be part of the audit as defined in the NATM.

114. The expert from CLEPA explained that methods to validate the performance of AI agent were existing. He mentions methods such as blackbox testing, statistical approaches and therefore disputed the word “impossible” as GRVA could rely on methods developed by Institute of Electrical and Electronics Engineers (IEEE), ISO and International Electrotechnical Commission (IEC) for this purpose. He mentioned existing issues to be addressed at the regulatory level, such as the use of continuous learning features at “Vehicle Identification Number (VIN) level” (i.e. individual vehicle level, opposed to vehicle fleet level).

115. GRVA could not further discuss this proposal, due to time constraints, but agreed to prioritize this item at its May 2021 session.

12. (d) | Framework document on automated/autonomous vehicles (FDAV)
Relates to Artificial Intelligence | Automated Driving |