After a great deal of careful work, the California Department of Motor Vehicles (DMV) today released its final rule for the testing of "autonomous vehicles" on public roads in the state. Accompanying this rule is a Final Statement of Reasons that, on page 9, contains a striking exchange:
In response to Robert W. Peterson's comment that "[t]he regulation does not quantify how 'safe' a vehicle must be to be operated on public streets," the DMV explains that "[t]he federal scheme for the regulation of vehicle safety relies on self-certification by manufacturers. The proposed regulation is consistent with the federal regulatory scheme for vehicle safety certification."
As I've noted, the issue of "how safe is safe enough" is as important as it is difficult. What level is appropriate? What metric is meaningful? What proof is sufficient? In each case, who decides? These questions matter not only to research-and-development testing (the subject of the rule released today) but also to general consumer operation (the subject of the rule that must be released by the beginning of 2015).
The logic of the California DMV's response would seem to reach each of these domains. In other words, the California DMV may be moving, perhaps by necessity, toward a manufacturer-centered approach for operation as well as for testing. Three notes on this:
First, recall that the Nevada DMV was the first agency to specifically regulate automated driving. Its approach to R&D testing relied in part on entry barriers (posting a bond and driving 10,000 prior miles) on the implicit assumption that developers that could overcome these barriers would have internal mechanisms and external incentives to ensure the safety of their automated systems. But when the California DMV was looking ahead to its own rulemaking, the Nevada model faced criticism for being too deferential to these developers.
Second, the federal self-certification requirement is actually not as nebulous as the California DMV's statement suggests. In contrast, "safety" can be ambiguous in the context of state product liability law and federal automotive recalls. (I cite this last page reluctantly, because its introductory paragraph misleadingly suggests that motor vehicle defects as they are conventionally understood largely explain today's crashes and injuries.)
Third, it's understandable that any DMV would want to avoid these safety questions in the hope that the National Highway Traffic Safety Administration (NHTSA) will eventually answer them. Indeed, many state bills, including California's, invite NHTSA to develop performance standards.
Until that point, I'm interested in, though by no means committed to, an inchoate approach that would require developers of automated driving systems to make public safety arguments regarding their systems--arguments that could be publicly scrutinized and independently reviewed as proxies for more conclusive definitions or findings of safety.
This post also gives me an opportunity to highlight the guide to automated driving legislative and regulatory action that I started and that Gabriel Weiner expertly maintains. Given the proposed amendment of the 1968 Vienna Convention on Road Traffic (but not the 1949 Geneva Convention on Road Traffic, to which the United States is a party), we're likely to see increased activity in other countries as well. Please note, however, that the jurisdictions that happen to have a bill or a law are not necessarily the same as the jurisdictions that are thoughtfully and meaningfully preparing for increasingly automated motor vehicles.