It’s Better if you Forget about L3-L5 Autonomy

Article By : Colin Barnden

Let's apply the SAE automation levels to the mass-market cars. Nobody should be surprised how the future of autonomous vehicles looks a lot different in Detroit, Stuttgart and Tokyo

Mass-market car OEMs are heading for extinction, right? We all know that they will be killed by Tesla and robo-taxi companies like Cruise, Uber and Waymo, who are just quarters away from deploying perfect “self-driving” technology — aren’t they?

Let’s take a look at the 2018 production volumes for an indication of where the power really lies here:

  • Tesla: about 250,000
  • Robo-taxis: 0
  • Traditional mass-market car OEMs: about 95 million

Does anyone seriously believe the traditional OEMs will supply robo-taxi companies in sufficient quantity to bring about their own demise? No, me neither.

The threat from Tesla and mobility startups has been massively overstated these last five years — media hype and ambitious timescales cannot disguise the fact that “self-driving” technology remains firmly in the R&D phase and is nowhere close to commercial deployment.

With that in mind, let’s do what no one ever does and have a run through the SAE automation levels — but specifically from the point of view of the boring old OEMs — to see just how different the world looks in Detroit, Stuttgart and Tokyo, compared with Silicon Valley.

Level 0 (No Automation): In order to understand where we are headed, we first need to understand where we are. Today, more than 90% of all light vehicles in use on our roads and highways have no automated driving features at all — which is Level 0.

There’s plenty of scope for OEMs to make human drivers into safer drivers with the adoption of ADAS at Levels 1 and 2, all done in parallel with the long, expensive and torturous journey to the successful development of autonomous driving at Levels 4 and 5.

Let’s also debunk the myth that 94% of crashes are the result of human error. In a frequently mis-quoted NHTSA report — which you may have never actually read — in Table 1 it states “94% driver error.” Many thanks to human factors specialist @AnnikaFLarsson for highlighting this important point.

The oft-cited argument from Silicon Valley that “94% of traffic deaths are the result of human error; humans totally suck at driving; ergo, replace human drivers with machine drivers and traffic deaths drop to zero,” is garbage.

Level 1 (Driver Assistance, or “Hands-on, eyes-on, mind-on”): Level 1 is typically speed assist, such as autonomous emergency braking (AEB). AEB is used to correct the longitudinal control errors of distracted and fatigued drivers, so expect to also see IR vision driver monitoring systems (DMS) adopted rapidly in Level 1 vehicles.

Level 2 (Partial Automation, also “Hands-on, eyes-on, mind-on”): Level 2 is both speed assist and lane support, such as AEB and lane-keep assistance (LKAS). AEB and LKAS are used to correct the longitudinal and lateral control errors of distracted and fatigued drivers, so expect to also see IR vision DMS adopted rapidly in Level 2 vehicles.

According to the World Health Organization (WHO), 1.35 million people died in road crashes last year. Mass adoption of AEB, LKAS and IR vision DMS might reduce this death toll by over 90% — equating to a ten-fold improvement in human driving. How this would influence the safety debate for machine-driven robo-taxis if they must be ten times safer than human drivers, is unknown.

For mass-market vehicles, Level 2 is forecast to be by far the largest segment over the next ten years. Euro NCAP’s 2025 roadmap — which specifies the use of AEB, LKAS and IR vision DMS to achieve a 5-star rating — will drive this growth, along with mandatory installation of all three systems, as required by the European Parliament, starting for all new models sold in Europe from May 2022. NHTSA, with no equivalent plans for improving road safety in the U.S., is currently asleep at the wheel.

Level 2+/3- (best described as “Hands-off, eyes-on, mind-on”): This is not an SAE automation level, but it approximates to a system with operational design domain (ODD) limits maintained via over-the-air updates and an advanced IR vision DMS to permanently monitor driver state. Level 2+/3- will be distinguishable from Level 2 by long duration lane centering for a “highway assist” function — à la Autopilot.

GM’s updated Super Cruise will meet the definition for Level 2+/3- in 2020 and you can expect similar systems to be launched by every OEM by 2023. Tesla’s Autopilot/Full self-driving will be classed as Level 2+/3- with the addition of an ODD and IR vision DMS.

In a Level 2+/3- system the human driver will be liable at all times — including when the highway assist function is operational. The legal position for the driver would therefore be: “You engage it, you are liable for it.”

Level 3 (Conditional Automation, or “Hands-off, eyes-off, mind-on”): Activating conditional automation at Level 3 is not the issue — the issue is navigating the dreaded “machine-to-human handover,” when the machine driver requests to hand back control to the human.

Since it is irresponsible for the machine driver to simply signal to the human “Here, you take over,” it is evident there must be a period of time following the handover request for the human driver to regain proper situational awareness.

The OEM will be liable for accidents and crashes when the machine driver is in control and all throughout this handover period — which may be as long as 45 seconds. The legal position for the OEM would therefore be: “You offer it, you are liable for it.”

We know from videos of Autopilot use (misuse?) that crazy things will happen in the handover period. Under the definition presented here, no safety conscious OEM will ever offer Level 3, electing instead for a Level 2+/3- system with no liability issues.

Level 4 (High Automation, or “Hands-off, eyes-off, mind-off”): How many times in the last couple of years have we heard a Silicon Valley mobility startup boast: “We skipped Level 3 and went straight to Level 4.” Well done! So where is it?

Let’s just forget about Level 4 bullishness a l'outrance and look at cold, hard numbers: OEMs are currently agonizing over adding ADAS and DMS features costing a couple of hundred bucks to mass-market models — are they seriously going to consider $5,000+ for an unproven sensor suite and neural net engine? Not any time soon.

Level 4 currently looks like a great way for robo-taxi companies to incinerate cash. For an OEM developing an automation strategy for the mass-market, Level 4 is not so much a path to the second chapter of autonomy, as the way to Chapter 11 bankruptcy.

Level 5 (Full Automation, or “Human-off”): The machine driver handles all situations, all extremes, all unknowns, all of the time. If you think this is realistic in any mass-market vehicle in the next ten years, you probably own a Tesla, or are an early stage investor in a Level 5-related company. I accept your opinion and will just agree to disagree.

Winners: Anyone involved in ADAS — particularly suppliers of camera and radar modules — and Mobileye (Intel). Xilinx also looks particularly well positioned for success, with design-wins for both ADAS and DMS.

With global shipments of automotive IR vision DMS forecast to rise to almost 70 million in 2025, from about 1 million this year, it is the suppliers of DMS software that look likely to be the major winners. Seeing Machines — which provides the DMS for Super Cruise in the Cadillac CT6 — is the clear technology leader and well positioned to take the largest market share for DMS in Level 2+/3- vehicles.

For lower end vehicles operating at Levels 1 and 2, the DMS design-wins look set to be mostly shared between Seeing Machines, Smart Eye, Eyesight Technologies and Jungo Connectivity, among others.

Not Winners: With a very limited opportunity in the mass-market for Levels 3 through 5, Nvidia is likely to find its Xavier GPU is too expensive and consumes too much power (30W, what were they thinking?) compared with Mobileye’s well-known and well-proven EyeQ family.

Similarly, the demand for high performance lidar looks set to be limited — which is not great news for the 50+ lidar companies targeting the automotive market.

In Summary: Forget about Levels 4 and 5 in mass-market vehicles — and Level 3 too, unless any OEM is brave enough to take on the liability issues of the machine-to-human handover. Levels 1 and 2 and Level 2+/3- look set to see the highest volumes, at least throughout the 2020s. As for the 2030s, I’ll get back to you in ten years!

SAE levels of automation

(Source: SAE)

Colin Barnden is principal analyst at Semicast Research

Leave a comment