What SAE Automation Levels Actually Mean for Safety

Article By : Colin Barnden

SAE Levels offer driving automation, but say nothing meaningful about safety. It doesn’t work in practice, because the safety question is completely different for L3-5 from L0-2.

To paraphrase Jane Austen: “It is a truth universally acknowledged, that a tech company in possession of a good fortune, must be in want of disrupting the auto industry.” Apologies to Austen fans, but this is what I thought whenever I heard an autonomous vehicle (AV) executive preaching about saving lives.

Those sermons were predicated on faith that upper levels of SAE automation were easily achievable. Which just goes to show you that faith and technology and two completely separate things.

Being a tech analyst for over 25 years and covering automotive for more than 20 gives me some perspective. If I were to reduce all my knowledge about the auto industry into a single sentence and offer it as advice to the deep-pocketed tech industry, it would be this: “The best way to finish up with a small fortune in automotive is to start with a large one.” Thomas Tusser, an English poet who died in 1580, put it much more succinctly: “A fool and his money are soon parted.”

The automotive opportunity looks enticing, with production volumes totaling tens of millions a year, about twenty significant automakers to focus on, and vehicle design-cycle times typically lasting four to seven years. Easy and once you are in in automotive, you are in forever, right?

But beware, the auto industry kills suppliers, ends careers and annihilates unsuspecting investors like no other. There is a reason why “OEM automotive grade” is often described as “military-spec products at consumer prices” and why automotive development timescales are forever. Automotive is a tough industry and always will be.

Battle-hardened auto industry veterans — particularly those with experience working for electronics, semiconductor and software suppliers – will be nodding sagely and chuckling to themselves. Naïve, optimistic VCs currently watching their “sure-thing” investments in AV or lidar suppliers turning to dust may soon come to realize what Thomas Tusser was talking about. Welcome to automotive. Welcome to hell.

What happened to the self-driving dream?
The future looked so much more appealing five years ago, when self-driving was a press release accompanied by a cute graphic of a human sitting in the driver’s seat reading a book. Remember those halcyon days? A couple of sensors and some software and all cars would be self-driving by the early-2020s.

Then came the dawning realization that the cost of the sensor suite and associated processors would be prohibitive for privately-owned vehicles, so the self-driving industry pivoted to a vision of robotaxis and “mobility-as-a-service.” The future became somewhere that no sensible person would ever own a vehicle because ride-hailing would be so cheap. Then reality interjected, Elaine Herzberg was killed by a test-level Uber in Tempe, Arizona and the dream died.

Today’s self-driving development efforts can be summed up in one word: More. More money, more testing, validation, sensors, processing, delays. More of everything. Except tangible results, profits and an unambiguous answer to the question “how safe is safe enough?”

I don’t dispute there is incredible technology and innovation here and that AV is a longer-term answer for something. Is there a commercially viable business model in operating robotaxi fleets? I don’t know. Is there a large untapped pool of funding available to find out? I doubt it.

How did we end up here? How did the self-driving industry lose its way? Maybe it’s because the entire map is wrong.

The levels don’t work
Officially known as J3016, the levels provide detailed definitions of driving automation guiding the industry from no automation (Level 0) to full automation (Level 5). At least that was the plan, but as Mike Tyson famously said “Everybody has a plan until they get punched in the mouth” and real life looks to have dealt J3016 a knock-out blow.

J3016 provides detailed definitions of driving automation, but says nothing meaningful about safety; it doesn’t work in practice because the safety question is completely different for Level 3-5 machine-driven vehicles (how safe is safe enough?) than for Level 0-2 human-driven vehicles (how to make human drivers into safer drivers?). J3016 conflated these two entirely unrelated development tracks.

When the rubber meets the road, Level 5 is nowhere close to reality, Level 4 is a work in progress and Level 3 is unsafe. A more realistic description of these levels might be:

Level 5: Futuristic Fairytale. The home of AI magical thinking and populated entirely by unicorns.

Level 4: Investor Incinerator. Burns everything put in to it, especially predictions, careers and reputations. Does Level 4 work? There’s still a #Waytogomo.

Level 3: Handover Hell. Up to 45 seconds? At highway speed? Are you serious? Great for lawyers, death for automakers. Missy Cummings has been warning about mode confusion forever. Level 3 is dangerous, so let’s just admit it and move on.

In comparison, for Level 0-2 human-driven vehicles the primary role of technology is to enhance safety by warning, assisting or intervening in the form of driver-assistance (ADAS) systems. However, as J3016 states:

Active safety systems, such as electronic stability control and automated emergency braking, and certain types of driver assistance systems, such as lane keeping assistance, are excluded from the scope of this driving automation taxonomy because they do not perform part or all of the DDT [Dynamic Driving Task] on a sustained basis and, rather, merely provide momentary intervention during potentially hazardous situations…and thus are not considered to be driving automation.

So, for human-driven vehicles J3016 actually discounts safety and instead promotes convenience. Which is how we ended up with Autopilot. Surely it is now time for J3016 to be dumped in the trash and replaced with two completely new standards, one for machine-driven vehicles and one for human-driven vehicles.

Euro NCAP, working in conjunction with Thatcham Research, has completed much of the work for a human-driven vehicle safety standard, encompassing automatic emergency braking, lane-keep assistance and vision-based driver monitoring systems to mitigate against human distraction, drowsiness and impairment. The updated European General Safety Regulation specifies installation of these systems in vehicles sold in Europe, coming into effect from June 2022 to June 2026. SAE will have to start over on a standard for machine-driven vehicles.

I’m not expecting much to happen in the short-term, at least that is until the money runs out. At that point some technology executives will understand that traditional automakers are slow and cautious technology-following businesses for good reason: These are the hallmarks of a safety-first culture.

Cash rich, Silicon Valley tech startups on a mission to save lives probably have little time to read literature. They would do well to remember that Jane Austen wrote “It is very difficult for the prosperous to be humble.”

Leave a comment