#j3016 — Public Fediverse posts

What is a Level 3-capable vehicle?

Put simply, per the #SAE #J3016 standard (which is not safety standard), this is a vehicle that is capable (by definition, not necessarily in practice) of detecting a certain amount of failures to the Dynamic Driving Task (DDT) within a certain Operational Design Domain (ODD).

With some exceptions, the human driver can (theoretically) remain inattentive to the performance of the DDT until the automated system demands that they regain operational control.

Naturally, the entirely worthless #NHTSA (the US's theoretical roadway and vehicle safety regulator), in its infinite wisdom, has stupidly produced its own "version" of the #J3016 standard - adding to consumer confusion unnecessarily and dangerously.

But even the NHTSA's version does not include "self driving" in their own level descriptions.

**Do not use the NHTSA's version!**

The NHTSA is stupid.

https://www.nhtsa.gov/technology-innovation/automated-vehicles-safety#the-topic-road-to-full-automation

Ok. Back to the chart above.

Level 2 is labeled as "Occasional Self-Driving".

This is wrong.

All of the levels labeled above are wrong, but let's focus on Level 2.

Officially, a Level 2 is described as a "partial driving automation" in the #J3016 standard.

But the thing to focus on is the relationship between the human driver and the automated system in a Level 2-capable vehicle.

The human driver is the fallback for the automated system **at all times and under all circumstances**!

The second issue is that the #J3016 has large capability "gaps" between each level which invites the construction of certain "marketing language" that is not included in the standard.

You can see but one example in the Tweet above ("Level 2+").

These designations are invented because, for example, while a #FSDBeta-equipped vehicle **is** a Level 2-capable vehicle, the ADS has some Object and Event Detection and Response (OEDR) capabilities - which is a capability associated with Level 3.

Let's talk about some of the core issues with the #J3016 standard, though.

The first, most important problem is that it is not a safety standard.

It is just a basic taxonomy of capabilities for #AutomatedDriving systems (ADS).

But what good is a capability in a #SafetyCritical system if there cannot be a foundational safety case built around it?

There is none.

That leaves the door open for various laypeople to see safety, explicitly or implicitly, as a secondary concern.

While one could go to the official #SAE website and download the #J3016 for free... I would highly recommend Professor Phil Koopman's "User Guide" which draws only from the standard and attempts to explain the dry technical language within in a more approachable way.

Professor Koopman does not at all pervert the standard language and has taken great pains to not do so, unlike many individuals and organizations scattered about on the Internet.

https://users.ece.cmu.edu/~koopman/j3016/

Ok.

What are these "levels" everyone is talking about?

The original source of the levels is a standard published by the #SAE with the designation of #J3016.

The standard is entitled "Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles".

Note that the title does not include the term "autonomy" which is found no where in the actual standard.

If you want to utilize "the levels" in conversation, this is the only place you should be drawing from.

@takashioomoto Oof. Yeah.

But, truth be told, the #SAE #J3016 standard has been a mistake from the start, IMO.

What was *supposed* to be a broad taxonomy of automated driving features has been basically warped (understandably so) by marketing and the media to fit certain narratives.

Worse still, J3016 is not a safety standard and laypeople (again, understandably) attempt to cram it into some sort of safety argument.

We really just needed a safety standard - and not some arbitrary taxonomy.

Yup. You guessed it.

Level 3-capable vehicles are a complete mess from a Human Factors safety perspective... which is one of the major problems with the #J3016 standard existing without being a safety standard.

If a system definition does not incorporate safety, does the definition have any value?

It does not in my book.

But if the terminology in J3016 is to be used, which it is here and pretty much universally elsewhere, it is crucial that the terminology is used very carefully.

A Level 3-capable vehicle does not strictly allow the driver to "lose focus" (however "focus" is defined) in all situations.

The #J3016 standard specifies that a human driver in a Level 3-capable vehicle should be afforded a warning of "at least several seconds" before they should regain operational control of the vehicle ***BUT*** the human driver ***MUST*** ***ALSO*** be "receptive" to "evident" or "kinesthetically apparent" failures even if the system does not issue a warning.

"The levels" (i.e. Level 2, Level 3) are used in throughout this article, and while it is left unstated, "the levels" are part of the SAE #J3016 standard which is a basic taxonomy of automated driving systems.

The J3016 standard is not a safety document.

As such, the terms "semi-autonomous", "#autonomous" or "autonomous-adjacent" are not defined in the J3016 standard and, without a well-defined safety case, these terms lack meaning.

I think that these terms should be avoided as a result.

The term "driver assist" (or similar) is also not defined in the #J3016 standard - which is not a safety standard, but it is the closest thing we currently have to a generally-accepted #AutomatedDriving terminology and taxonomy document.

Do not know what the J3016 standard is?

Have you seen J3016 or "the levels" referenced in various articles over the years?

Check out this wonderful User Guide by an expert in the field, Professor Phil Koopman: https://users.ece.cmu.edu/~koopman/j3016/

I am not sure how I feel about this language highlighted here...

A conditional #AutomatedDriving system (likely described by "hands-free, eyes-off" which is not part of the official SAE #J3016 terminology by the way) seems entirely incompatible with the concept of "driver assist".

I really hope to see a detailed safety case here - and one should be demanded by regulators anyway, you know, hypothetically in the US.