In 2014, SAE International proposed levels for vehicle automation which was later adopted by National Highway Traffic Safety Administration (NHTSA) and updated in 2018.
It can be downloaded for free from here. The official title is a catchy “Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles” or “J3016”.
This 35-page document is fundamentally excellent and fills a legitimate engineering need. But these concepts may have been too successful and are now permeating the wider culture. They have been picked up, published, re-published, misquoted, misread, not read and oversimplified.
You find yourself getting caught up in endless details, acronyms, rules and definitions. (e.g. “driving automation system” needs to be written in lower case but “Automated Driving System” needs capitalisation).
The document is in dire need of an executive summary, a table of contents, rudimentary formatting and a glossary of terms. It’s not an easy read especially when one part of the document references obscure acronyms that are not defined until later in the document. The juicy part does not start until an obscure heading at the bottom of page 17.
In part, this may be a reason for so many misconceptions.
It defines 6 levels of automation - but it starts counting at “level 0” where there is no autonomy up to “level 5” where the vehicle does all of the driving and can get itself out of trouble in all circumstances - hence the confusion between the total number of levels.
The levels are discrete and mutually exclusive. They are well defined and are in no way a continuum or spectrum (despite what the popular press may say). A vehicle cannot be a low functioning level 4 or partial level 4. You cannot have a level 2.5.
But you can have different levels based on features; a vehicle may have a level 4 valet feature, level 3 freeway feature and level 2 arterial feature.
This could get incredibly confusing in the future when you pick up a hire car (or a new car) and you assume the car does one thing in a certain situation – and it doesn’t (assuming that it will stop at a red light when it doesn't could be problematic).
Compressing the 35 page document into little more than iconography propagates misunderstandings. The following is from Google.
The document in no way specifies, and nor should it, any requirement for the car to be “connected”; i.e. communicate with other vehicles, the road infrastructure, the car manufacturer or the internet (despite what the popular press may say).
The document stresses (albeit on page 30) that the level assignments are nominal, rather than ordinal. This means that level 4 is not “better” than 3 as such; just different. Much of the industry seems to have ignored this advice and “achieving level 5” is described like achieving a goal such as going to the moon.
This is probably my greatest concern;
This whole taxonomy is entirely counterproductive. It is inadvertently causing the market to attempt to solve problems to achieve a “4” or even “5” when the energies would be best invested elsewhere. Some automotive companies are being run by marketing departments and need to pay more attention to their engineers.
Next week, we will explore this last point more deeply and investigate the generally overlooked “ODD” (Operational Design Domain), how it is key to understanding the levels of autonomy and how it has been abused.