WP.29-175-21
Artificial Intelligence and vehicle regulations

In 2015 and 2017, public figures and experts warned about potential risks related to Artificial Intelligence use. AI has found prominent applications in the automotive sector in infotainment, vehicle management, and self-driving capability development. Some AI implications fall within WP.29’s remit, including HMI distraction and automated vehicle performance. AI in automotive applications searches for optimized solutions, looks for patterns, and predicts future events. Machine Learning, a subset of AI, uses data-based computational techniques to improve performance without explicit programming. Deep Learning uses multiple layers of nonlinear processing units for feature extraction. AI applications in the automotive sector include HD Map building, image analysis for object detection, driving policies for automated driving, and human machine interface. AI raises questions regarding model transparency, traceability, validation for safety and environmental performance functions, and whether vehicles of the same type should have identical AI performance.

UNECE server
Excerpts from session reports
WP.29 | Session 175 | 18-22 Jun 2018

37. The secretariat presented the informal document WP.29-175-21 on artificial intelligence, which was welcomed by ITU.

38. WP.29 welcomed the presentation and referred the document to the Task Force on Automated Vehicle Testing.

GRVA | Session 1 | 25-28 Sep 2018

21. The expert from UK presented the developments in 2018 on the creation of the Task Force on Automated Vehicle Testing (“AutoVeh”) and its two subgroups called “SG-1” and “SG-2”, established under WP.29. He presented that the aim of these groups was to develop a novel innovative concept for the assessment of Automated Driving technologies. He recalled that Terms of Reference (ToR) for this group were not adopted and that “AutoVeh” was transferred to GRVA in June 2018 (including the corresponding draft ToR). The expert from Japan reported on the status of the discussions in the groups and presented slide 3 in GRVA-01-35. GRVA agreed that the parent group (“AutoVeh”) would no longer be needed as the management of SG-1 and SG-2 would take place at GRVA.

22. The expert from Japan Co-Chair of “AutoVeh” presented ToR of both Subgroups (GRVA-01-07 and GRVA-01-11). The expert from EC sought clarity on the scope of SG-1 and on possible overlap with IWG on ACSF. The expert from China mentioned the lack of clarity of the proposed structure and proposed to classify existing and new work items to create groups with sufficient compatibility for existing and new ones in the future. He also mentioned the challenges related to the work on simulation and on-road testing due to the differences in traffic conditions and traffic rules within the contracting parties. He recommended to not endorse the ToRs proposed at this session. The expert from Germany asked about potential work duplications and asked whether GRVA or WP.29 already agreed on the three pillars concept mentioned in the ToR. The expert from Japan answered the questions.

23. After discussion, GRVA did not endorse GRVA-01-07 and GRVA-01-11 but, noting arrangements were already in place for further meetings, agreed that the groups continue to work until the next GRVA session. GRVA suggested that the group should refine the ToR to develop the novel innovative concept to verify the compliance with technical requirements (such as those developed within the IWG on ACSF) and demonstrate the validity of the new approach. The group should also prepare a full discussion on the three pillars approach and start working.

24. The secretariat recalled the decision of WP.29 to defer the document on artificial intelligence and vehicle regulations (WP.29-175-21) to SG-1.