A Blog by Jonathan Low


Jun 20, 2019

Self-Driving Cars Have a Problem: Tech Is Making Human-Driven Autos Safer

As in other areas of automation, using machines to enhance human performance may be superior to total replacement of humans by machines. JL

Christopher Mims reports in the Wall Street Journal:

If you buy new makes and models of car today, you might be surprised to find that, as a standard feature, it will take over when it thinks you’re making a mistake. technology developed to give us driverless vehicles could delay their adoption. When car makers put incremental tech advances in human-driven cars, they pre-empt one of the fully self-driving car’s supposed advantages: safety. New systems marry the best machines capabilities with the best of the human brain. For decades ahead, fusion of human minds and machine reflexes will be the norm.
If you buy one of many new makes and models of car today, you might be surprised to find that, as a standard feature, it can do something your previous car couldn’t: It will take over when it thinks you’re making a mistake.
In the coming years, many cars will do more than that, even driving mostly by themselves, at least on highways. And not just luxury models such as the latest Audi A8 or Cadillac CT6, but something as mainstream as a Nissan Rogue.
Some of this technology has been in development for years, but the newest versions of it—with advanced object recognition, radar-and-laser detection and lightning-fast artificial intelligence—were created for autonomous cars. Many tech entrepreneurs have argued that fleets of robo-taxis would convince us to abandon personal car ownership in favor of “transportation as a service.” Some of them have predicted these robot cars will start populating U.S. roads within the next two years.
But the paradox of how this evolution is playing out is that technology developed to give us driverless vehicles from the likes of Tesla Inc. and Alphabet Inc.’s Waymo could actually delay their adoption.
When car makers put these incremental tech advances in human-driven cars, they pre-empt one of the fully self-driving car’s supposed advantages: safety. These new systems marry the best machines capabilities—360-degree sensing and millisecond reflexes—with the best of the human brain, such as our ability to come up with novel solutions to unique problems.
U.S. regulators designate six levels of driving automation, from none (0) to full (5). To get so-called Level 3 autonomous driving—where humans must be in the driver seat but can take their hands off the wheel—self-driving systems “have to exceed the human level of skill at driving,” says Danny Shapiro, senior director of automotive at Nvidia Corp. which offers self-driving automation tech to partners including Audi and Mercedes-Benz.
“We are sentient beings, and we have the ability to reason from first principles, from scratch if you will, while AI on the other hand is not conscious, and doesn’t even understand what it means that there’s a physical world out there,” says Raj Rajkumar, a professor of engineering at Carnegie Mellon University who collaborates with General Motors Co.
Machines are already a huge help to drivers. Take automatic emergency braking, or AEB. That’s when your car stops itself if it detects that you’re about to hit another vehicle or other obstacle. According to new data from the Insurance Institute for Highway Safety, AEB reduces rear-end crashes by 50%, and reduces crashes with injuries by 56%. In the U.S., there were 1.7 million such rear-end crashes in 2012, resulting in 1,700 deaths and 500,000 injuries, according to the National Transportation Safety Board. Mass deployment will take many years, but the NTSB has estimated that this technology could eventually reduce fatalities and injuries from rear-end crashes by 80%.
By 2022, nearly all new vehicles in the U.S. will have at least automatic emergency braking, thanks to a voluntary commitment made by 20 auto makers, from a small but growing percentage of new cars today.
AEB and related Level 1 technologies such as lane-departure warning, blind-spot detection and reverse automatic braking—collectively known as ADAS, or advanced driver assistance systems—are just the beginning.
Cars are also beginning to incorporate technology developed for fully self-driving cars, such as ultra-detailed, centimeter-accurate maps of much of the U.S. highway system. Cars navigate on through these maps and the real world using a combination of GPS and other location technologies. Some of these systems, known as Level 2, are actually able to drive for a human on highways, says Amnon Shashua, chief executive of Intel Corp. subsidiary Mobileye, which provides the majority of cameras and processors used in today’s driver-assistance systems. These maps are what enable partial self-driving in the newest Nissan Rogue, Leaf and Altima models with the ProPilot Assist system, and in the new Audi A8, he says.
These systems can drive for you but they require you to pay attention to the road, and keep your hands on the wheel. (Sometimes they even use cameras to check on you.) They’re a kind of enhanced cruise control that can steer to keep the car in the lane and maintain a safe following distance from the car ahead. Camera-based sensors can sometimes have trouble spotting poor lane markings, but systems like Cadillac’s Super Cruise use maps so detailed, they know where the lane is, regardless of whether they can sense the markings.
The challenge is measuring the effectiveness of Level 2 self-driving technology. It will in theory lead to even safer vehicles, in part because they’re programmed to drive more conservatively than humans tend to, says Ian Reagan, a senior research scientist at the Insurance Institute for Highway Safety. But there isn’t enough data to know the magnitude of the effect it will have, he says.
The data will likely arrive soon. Nissan’s aggressive expansion of ProPilot will make it a standard feature on all vehicles. In 2021, about 1 million vehicles with ProPilot will ship world-wide, says Andy Christensen, senior manager at Nissan’s North American technical center. GM is planning to roll out its Super Cruise system—which uses an ultra-detailed road map similar to the one built by Mobileye—for all Cadillac models in 2020.
Ford has just started offering its first Level 2 self-driving system, adaptive cruise control with lane centering, as an option. Honda and Toyota offer ADAS technologies on the majority of their new vehicles and some form of adaptive cruise control as an option on many.
For decades ahead, this fusion of human minds and machine reflexes will likely be the norm. Researchers at Cleveland State University estimate that only 10 to 30 percent of all vehicles will be fully self driving by 2030. That’s in line with predictions from others—a PwC analysis estimates that 12% of all vehicles will be fully autonomous by then. Initially, all fully self-driving vehicles will be Level 4—that is, they have to be in geographically constrained areas, and will only operate in good weather, as does Waymo’s fleet of self-driving vans that it is testing in Phoenix. Truly autonomous, aka Level 5, cars are still science fiction.
Waymo CEO John Krafcik told an audience at The Wall Street Journal’s annual tech conference in November that self-driving is “really, really hard” and that such vehicles won’t be ubiquitous for decades. He said cars might never be able to drive autonomously in all weather conditions.
There are many unknowns in this transition that favor the status quo, from consumer acceptance to the high regulatory bar that it must meet, not to mention legal liability in the event of an accident.
It’s also not clear whether current sensors will be enough to accomplish full autonomy. If it requires pricey laser-based sensor equipment and even transponders and sensors in the roadway, then cost will remain a significant barrier to its adoption. If car companies and their customers come to be satisfied with the safety and comfort of human-driven vehicles they might lose the impetus to invest the vast quantities required to push self-driving cars to their theoretical limits. Other investors’ eagerness might also wane.
That could have huge implications for the fortunes of companies like Tesla, whose CEO Elon Musk said would transform its existing fleet of cars into a 1 million-strong robo-taxi armada by 2020—something few analysts believe is feasible. It could also spell doom for companies such as Uber and Lyft, which aren’t yet profitable and might not be until they can cut out their high human costs—that is, removing drivers from vehicles.
While self-driving technology might not be able to rescue the fortunes of some of tech’s biggest transportation visionaries, it does have the potential—here and now, implemented in human-driven cars—to reduce or even nearly eliminate road deaths.
Corrections & Amplifications
At the WSJ tech conference in November, John Krafcik, chief executive of Alphabet Inc.’s Waymo, said that self-driving cars might never be able to drive autonomously in all weather conditions. An earlier version of this article incorrectly quoted Mr. Krafcik saying self-driving cars might always require some kind of “user interaction” in inclement weather. (June 18, 2019)


Post a Comment