SAE summary of autonomy levels.
Society of Automotive Engineers
The Society of Automotive Engineers last week published an update to its widely used six-level automated driving taxonomy. SAE worked together with the International Standards Organization to improve taxonomy and turn it into a standard: “Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles, J3016.”
The full text of the standard runs 41 pages, and is available for free.
SAE “deprecates” a lot of terms in the standard, including “self-driving”, “autonomous”, and “robotic.” They don’t suggest that everyone should stop using those terms, but they refer to the terms as “vernacular” that adds confusion rather than clarity to a standards document.
Instead, the SAE makes heavy use of the acronym DDT: dynamic driving task. Each level of the taxonomy is defined by how much of the DDT it covers.
Historically, the most confusing of the six SAE levels has been Level 3. Indeed, in the SAE blog summarizing the new standard, the first bullet point is: “Further clarity on the differences between SAE Level 3 and SAE Level 4.”
Unfortunately, Level 3 may continue to raise concerns, even after this recent update.
The SAE’s summary infographic colors Levels 0, 1, and 2 as blue. Those levels specify “you are driving,” albeit with increasing support features.
Levels 4 and 5 are green. Level 5 is a complete self-driving (to use the vernacular) system that can operate everywhere a human driver can operate. Level 4 is a geo-fenced self-driving system that operates entirely autonomously, but in a limited area. Technically, the limitation could be something other than geography – weather, for example – but in most practical cases the binding constraint will be geography.
Level 3 remains amorphous. In the infographic, this is the only level that has both blue and green colors.
The new standard defines Level 3 as, “The sustained and ODD-specific performance by an ADS of the entire DDT under routine/normal operation with the expectation that the DDT fallback-ready user is receptive to ADS-issued requests to intervene, as well as to DDT performance-relevant system failures in other vehicle systems, and will respond appropriately.”
That is quite a long sentence, but it’s basically consistent with how the SAE previously defined Level 3: the human passenger isn’t “driving”, but has to be ready to take control of the vehicle when the system asks.
That leaves a fair bit of ambiguity, especially about how quickly the human must “receive” requests to intervene. That, in turn raises the question of how broad the range of other tasks is in which the human can engage, while still allowing enough time to take control if the system requires.
The standard clarifies this somewhat:
“Although automated DDT fallback performance is not expected of Level 3 ADS features, a Level 3 feature may be capable of performing the DDT fallback and achieving a minimal risk condition under certain, limited conditions.”
That caveat specifies that Level 3 systems do not need to bring the vehicle to a safe stop. That responsibility can be passed to the driver, who must respond within a “sufficient”, but undefined, time.
This definition may prove unworkable in practice. Some drivers would likely go to sleep or engage in other activities that would not permit them to take control of a vehicle within an effective amount of time. That lack of responsiveness would lead to collisions.
Perhaps carefully calibrated driver monitoring systems could resolve this issue. But more research would be necessary into response times that both allow the user to “not drive”, but also respond in a timely manner.
My guess is that in practice, most vehicle manufacturers will decide to effectively redefine Level 3 and require safe-stop functionality. Even if that means coming to a complee stop in the middle of a highway, it’s better to come to a collision-free complete stop than wait for a potentially sleeping user and instead suffer a collision. Manufacturers may in fact go even further, and define a “safe stop” as pulling over to the shoulder.
That seems like the only way to implement functionality that both drives the vehicle and requires the human driver to take control on-demand.
