Pitching for Commercial Level 4 Could be Better AV Ambition

The achievement of full Level 5 vehicle autonomy is, by many accounts, the “North Star goal” of all those involved with designing and building AV technology.

The literal interpretation of SAE International’s Level 5 definition of “everywhere in all conditions”, however, would require autonomous vehicles to be able to react to the untold combination of fringe cases and driving conditions that could possibly arise. With the timeline for Level 4 capable vehicles up to a decade out from widespread use, the road to Level 5 will require the baking in of a host of real-world driving experiences into a fully integrated ecosystem of sensors, infrastructure, mapping and artificial intelligence and machine learning technology.

“Technically, if we were to be really literal about it, Level 5 may not be a realistic goal but the spirit of what it means is you’re not limited in taking the vehicle anywhere you want to go, within reasonable conditions,” explained Jack Weast, senior principal engineer, Intel vice-president, Mobileye. “It’s more about being able to take the vehicle out with no time of day or geographical conditions, no normal weather restrictions, all of these kind of things. That’s definitely what the industry is striving for.”

Weast says the evolution to advanced Level 4 and, perhaps, Level 5 means the removal of barriers to where and when the vehicle can operate in autonomous mode, which would make the feature more attractive to consumers and businesses. He said the information being gathered right now from ADAS would also be an important piece to evolving autonomous vehicle capabilities and points to the development of common industry standards for safe self-driving standards as critical.

“This is the big one. We don’t have a common definition for what it means for an AV to drive safely,” he said. “Our view is there are lots of ways to differentiate from an IP standpoint, but safety is something that should be common ground, and we should be collaborating on that instead of competing. AI’s a useful tool but it has to be used in a way that’s safe by design, and you have to make sure you’ve got a safety model that can detect against potential mistakes,” Weast said. “Most algorithms that make the driving decisions are AI based and, just like you’re gonna get a recommendation every once in awhile from Netflix that makes no sense, there’s a non-zero chance it will make a bad recommendation inside a car.”

Andree Hohm, Continental’s head of autonomous driving program, shares Weast’s skepticism about the probability of attaining Level 5 autonomy if the definition is taken literally. “If we take a look at you as a human are capable in driving, it is enormous. You can go from Alaska to Cape Horn, and you will experience a series of adventures but there is a high possibility you will arrive at your destination,” he said. “Level 5 would have to do exactly that, and by that definition, it’s almost unachievable.”

From Hohm’s perspective, the big focus should be on integrating technological solutions with real-world experience through two “attack points”: collecting data from closed loop systems, like campus shuttle busses, and collecting data from ADAS systems out in the real world. “Taking both things together in a sort of pincer movement, we have two big opportunities to collect real experience and that’s what will take us to the next level if we are to achieve higher levels of autonomy,” he said. “The secret here is all these things will not come overnight.”

Integration

He also called for tighter integration between automakers and suppliers, noting vehicle architecture is becoming vastly more complex, and pointed out that because the complexity is so high, the traditional connection between suppliers and automakers will not suffice, requiring “closer, more complimentary” partnerships. “The suppliers are not just producing devices that you put inside the vehicle with a few connections. These are high performance computing platforms that are constantly updated, and all this has to be clearly defined in regard to how it can work safely, what the interplay is between of all these components, because only through that will you reach Level 4 for certain use cases.”

Gartner analyst Pedro Pacheco, said achieving Level 5 in a broad geographical area, with the ability to drive with full safety and reliability, will take more than 20 years, unless there are “dramatic quantum leaps” in AI and other areas that somehow unlock potential to fast track this technology. Pacheco said the business case for fully autonomous vehicles needs to make financial sense, from the amount of investment required, to the amount of time it will take to recoup that investment, to say nothing of turning a profit.

“For a ride-hailing company, Level 4 would be fine, for example autonomous trucks bringing delivery from depot to depot, and even that is something that would still take a considerable amount of time to achieve,” he said. “We hear a lot of companies doing Level 4 trails across the world but doing trials and having a fully finished product available to the public are two different things.”

He noted the industry has been spending a lot of time on reaching Level 4 capabilities and it’s still a long haul project, so why aim higher? “When you achieve Level 4, which will take the better part of the next decade, once we get to that point, the industry will have to ask itself, do we go for Level 5?” Pacheco said. “Like many technologies that are too far away, society as a whole cannot grasp what it can do. It’s still a little bit Knight Rider, and society needs to see what it can be before it starts believing this technology.”