well, many people had discussed about something inautonomous driving safety in the past, and for Korean release first version for Level 3 legislation, for U.S releasing second version for Level 4 white paper,and China establishing aggressively for autonomous driving.

And the author had develops quite deep inside theautonomous driving safety, here lists some opinions for your reference

For EE, and AI, cloud computing, following is thefoundation:

1. RAMS and SOTIF, multi-agent safety holds for thebackground of traditional autonomous driving safety in EE side; and present inthe 2020 Jan, those for multi-agent safety and SOTIF not enough present for Level 3+, even in UL4600, the safety cases had to be refined more and more evenin the known limits, but we have one safety argument to clearly explain what wehad done in the detail autonomous driving development. And present the SOTIFPAS version in 2019.01, combined with SAE J3016, for those level3 and level4, even level 5, (1) fallback from machine to human didnot assure the human machinerelationship problems, which higlighted in UL;

(2) how to deal with AI related algorithms, we detailfind the maching learning as only AI, that is not true, and author had ever metsomething like this, even for the underestimate for the AI related algorithms,due to high complexity, and uncertainty, too much layer of neural networks, andwe retrospeck back to maths, it is truly no methods to detail hierachically to describle.

2. AI algorithms and control theory shall be into asfoundation for strategic and operation level, including path planning,behaviour decision, and perception, as well as we will finding the trainingsets, no matter what leading to this one, following aspects shall behighlighted in the real project and investigations:

(1)Law of uncertainty, back to probability theory andallocation;

(2)Law of measurement, even UL standard didnothighlight how to deal with, but which reflect as at least REQUIREMENT level inthe safety case;

(3)Law of ODD or scenarios establishment, we had seenmany scenarios built in many standards, however, those who

3. Mapping between different safety standards

4. UL4600 related arguments

Note: fromhttps://edge-case-research.com/wp-content/uploads/2019/10/191010_UL4600_TECH_Webinar.pdf

UL4600 present some guidelines to safety case, how toderive some safety cases measurement and to have better description for SPIs,but not in extra theory in safety, in some meaningful, which can still haveguideline better for real project members who not safety relevant to understandand for safety audits/assessment external of companies.

Still, the most safety state of art is to derive outin AI phase, which is the most fresh issues, we cannot solve now.

Thanks for your reading and hope to be discussed inthe near future.