AV stuff to read
Preview alert: I will be getting to these two reports and I will also write something tying together all of the fabulous UK Law Commission AV reports. No promises on timing.Comments tally
Disclaimer/Explanation: Please note that anonymous comments and ones where the author's background is undeclared or not easily discovered are counted as "regular people" unless it becomes apparent that the commenter has professional expertise, which then shifts the comment into that category. Also note that the comments tally includes those comments summarized below in this blog post. All categorizations are my own and might seem random or wrong to someone else. I am not an engineer, an automotive geek, or a software expert.
Comments are due by Feb. 1, 2021.
As of Dec. 29, 2020, total comments to the Framework for Automated Driving System Safety Advance Notice of Proposed Rule Making (ANPRM) issued by NHTSA - 14
Anonymous - 2
Companies or Organizations - 2
Safety standards or framework - 2
Individuals with professional expertise - 6
Safety - 5
Cybersecurity -2
Regular people - 4
Regular people who want increased safety - 3
Regular people who do not believe AVs will be safer than human drivers - 1
Regular people who do not want regulation to constrain innovation - 1
Spam comments - 1
Instructions for submitting comments can be found in the ANPRM.
Please note that these comments are not necessarily representative of the opinions of the US population; rather they come from corporate or non-profit entities or professionals with an interest in AVs, or people with the time, energy, and space in their lives attend public meetings - in this case to devote attention to Federal Register notices. The regulation-making process is, if not hidden, then placed out of the way in the bowels of federal agencies, and is oft forgotten when we vote or choose where we participate in our democracy.
Two comment summaries - one short and one very long
(1) Comment of Craig McKee: This short comment is brief and expresses the opinion that we should remember that we will be shifting from a human-operated technology that has been with us for over 100 years, but that daily commute driving is "mundane" even for an avowed "driving enthusiast." Mr. McKee is hoping that AVs bring increased safety.
Juicy comment should be read slowly and carefully
(2) Comment of Mapless AI: This 10-page comment is submitted by a company that develops "safety critical artificial intelligence for automated driving." The firm is located in Boston and Pittsburgh, and it is seeking staff, as of today, to work on machine learning and robotics. Mapless AI is active on LinkedIn for anyone who wants to learn more. This comment is structured according to the NHTSA questions in the ANPRM and it was written by authors of AV safety publications.
Mapless AI favors the NHTSA development of a safety framework to enable AV safety and public acceptance. Because the Mapless AI comment uses engineering language and refers to systems engineering standards with which I am unfamiliar, I will do my best to do this comment justice and I welcome corrections, though this is a summary and not a play-by-play reiteration. The comment argues in favor of adopting specific systems engineering standards in order to remain technology agnostic and to reduce crashes. Numbers for those standards are mentioned in the comment.
NHTSA is in the singular institutional position to declare standards, according to the comment. "Others have tried and failed to agree on standards. The data could be expressed as the expected value of miles between accident types in specific ODDs." In its focus on ODDs - operational design domains - Mapless AI is wisely calling for a comparison of apples with apples and oranges with oranges so that rural winding roads with little traffic and no signals are not lumped in with 34th Street in the middle of Manhattan or even, I would hope, with the suburbs of Kansas City. But, as Mapless AI contends in the comment, this could be classified by speed rather than my own personal distinctions.
This would allow development to focus on safety of the design with clear data available to construct targets. The Agency would not restrict new development. The Agency would support innovation.
Time out: AV video from a student in Australia. The concept AV pod is somewhere between a Coney Island bumper car, a personal rapid transit vehicle, and a space-age, mobility-on-demand transit vehicle. I like the almost Ragtime musical accompaniment.
Comment's central theme presupposes adequate funding for NHTSA
To ensure accountability, the Mapless AI comment suggests that self-certification would be permitted, but with required information to be supplied to NHTSA about "core safety elements." Mapless AI recommends "accident metrics" for particular ODDs. [Note: This comment should not, in my opinion, be employing the word "accident" when it is not at all an unforeseen or rare occurrence that crashes happen on US roads. "Crash" is the preferred term.] The comment presupposes that NHTSA will have sufficient funding and expertise at the agency to adequately review the information submitted in the self-certifications and that performance requirements would aim for safer operation of vehicles with AVs than we currently have with human drivers. The comment comes back again and again to its theme, that we have technology agnostic performance standards that focus on safety.
While “see, think, act” is sometimes seen in the industry, it comprehends sensing, perception, planning, and control. Safety has been defined as the absence of unreasonable risk. Failures in any of these areas may lead to unreasonable risk in an ADS. However, while a set of core elements may enhance comprehension, safety does not require core elements. The result must be safe independent of the presence of a specific choice of core elements.
Even later in the comment, when responding to NHTSA's questions about additional research, the Mapless AI comment returns to this theme.
Mapless AI also suggests that NHTSA reframe its way of judging and categorizing in order to simplify standards. The comment posits that only two categories are needed: "the safety-preserving portion of the system and the non-safety-preserving portion of the system."
Time to regulate?
In response to whether regulation is necessary at this early stage, the Mapless AI comment advises that we will not be sacrificing innovation, and we may increase safety, if we require the submission of summary information to NHTSA. The comment offers a weighing of costs and benefits, that the cost to companies "may be partially offset by the data provided by NHTSA. Experience using this proposal may lead to improvements including simplifications or increased focus. More prescriptive regulation is not recommended. More prescription requires more experience." Indeed, instead of requiring that companies provide AVs to NHTSA, the comment advises that the above summary information and framework will ordinarily be sufficient.
For those truly educated in the engineering weeds, I recommend that you read this comment because it poses possible standards for systems engineering and safety analyses that I do not understand well, if at all. The bottom line for non-engineers does not cover the very important details in the comment. I would posit that the bottom line is contained in this response to one of NHTSA's specific questions: "The minimum evidence [that NHTSA should require] is a combination of vehicle testing, simulation, and analyses showing that ADS has exceeded the current expected value of miles between the types of accidents that the ADS has the authority to cause."
Disclosure recommended
The comment recommends (1) well-organized, particular information summary requirements, (2) minimum performance metrics, (3) flexibility in adjusting standards and information requirements, and (4) openness and sharing of information. "Such a summary could be requested even if it is an interim for a product in development. This enables improvement of the supplier’s safety case, uniformity of safety criteria, improvement of consumer confidence, and a basis for improvement of future administrative mechanisms." Very much implied in this comment is a NHTSA with the capacity to monitor AV information and research data.
Time out: Movie recommendation
(The movie clip features Joan Blondell as the wonderful aunt in A Tree Grows in Brooklyn.)
The comment seems to state - in my translation of engineering-oriented language - that the price of requiring information and establishing basic performance standards are (1) sufficient expertise and resources at NHTSA and (2) private companies adjusting to a uniform system, which could spur competition based on safety results. Again, comparing those apples with apples instead of oranges would be enabled.
I see the clouds clearing and the sun emerging with the logical arguments of this comment.
The three standards referenced in the framework are already well known in the automotive industry and are being included in the development process already by some ADS suppliers. The NHTSA data supports a uniform basis for competition and consistent consumer information. NHTSA fulfils [sic] its mission per the Safety Act. Technological advancement is not hindered. Compliance rewards results.
Finally, and in response to NHTSA questions in the ANPRM, Mapless AI argues in its comment that NHTSA clearly has statutory authority to follow the Mapless AI recommendations. I apologize for not adequately summarizing this comment for those of you who are engineers.
No comments:
Post a Comment