After a Tesla crash, more doubts drivers can be trusted with self-driving tech like AutoPilot

Tech
7
0

CLOSE

The driver of a Tesla Model S crashed into a fire truck while driving down a California highway.
Buzz60

SAN FRANCISCO — If you want proof that people will push the limits of a technology, even if at risk to their lives, look no farther than last week’s crash of a Tesla Model S in Utah.

According to a report issued Wednesday by police in South Jordan, a suburb of Salt Lake City, the 28-year-old woman at the wheel of the $100,000 electric sedan engaged Autopilot — Tesla’s driver-assist software that requires driver oversight — and then didn’t touch the steering wheel for 80 seconds.

Until she hit a stopped fire struck at 60 mph.

That she walked away with only a broken foot likely warrants a separate story on how the Model S can handle a crash. But for the moment we’re focusing on the fact that for half the length of your average pop song, the driver felt little need to monitor her vehicle’s progress.

The driver, whose name was withheld by police, couldn’t be reached. But she had told police previously that she was looking at her phone while the car sped along on Autopilot.

Police added that “the vehicle registered more than a dozen instances of her hands being off the steering wheel in this drive cycle,” including two instances where she didn’t touch the wheel for more than one minute. She touched the wheel briefly each time only after heeding a visual alert to do so that appeared on her dashboard.

Human drivers’ readiness to trust their cars’ ability to take over some basic driving functions is becoming more apparent as automakers feverishly add driver-assist tech to their fleets.

These features largely consist of basic heads-up systems that warn you about cars in a neighboring lane, but increasingly will consist of so-called Level 2 technology, wherein multiple sensors work in concert to let the car take over for you in bumper to bumper traffic, with your monitoring.

The reasons for this in-car tech arms race range from increasing safety (40,000 people died in U.S. road accidents last year) to keeping up with rivals (once only found in luxury cars, such tech is fast trickling down to entry-level models). 

More: Edmunds takes closer look at Tesla, Cadillac and Nissan semi-autonomous driving systems

More: Tesla driver in Utah crash kept taking her hands off wheel as car sped in Autopilot mode

But can humans be entrusted with tech that can lull us into a false sense of security and cause us to ignore automaker warnings about their fallibility? Last week’s crash suggests maybe not.

In a story last month focusing on the potential negative consequences of increasingly capable driver assist tech — essentially an automotive weigh station between regular cars and self-driving vehicles — an autonomous car expert seemed to predict the Utah incident.

“These systems are designed not only for ideal environments like good weather and clear lane markings, but also for rational behavior, and humans are predictably irrational,” said Bryan Reimer, a research scientist at the Massachusetts Institute of Technology who specializes in self-driving tech.

“The question is, does the automation work so well and require so little of you that in essence a new safety problem is created while solving for another?” Reimer asks.

More: Elon Musk shakes up Tesla as another Model S faces crash queries

More: Raw: Feds probe Fla. fatal Tesla crash and fire

The Utah driver’s reliance on Tesla’s Autopilot system wasn’t limited to the moments before the May 11 accident.

Citing data provided by Tesla engineers who checked the car’s computer logs, police said the driver engaged Autosteer (which keeps a Tesla between lane markings) and Traffic Aware Cruise Control (which will account for moving cars ahead of you, though not a stopped truck) on multiple occasions during her drive, “regularly adjusting the vehicle’s cruising speed” before settling on 60 mph slightly before the crash.

Both the National Highway Traffic Safety Administration and the National Transportation Safety Board are looking into the Utah crash and will make recommendations based on their findings. The investigations comes as Tesla is trying to ramp up production of its entry-level Model 3 sedan, considered critical to both short term cash flow and long term electric car-market strength. 

Both transportation agencies also are examining the details of a March crash of a Tesla Model X in Mountain View, Calif., in which Autopilot also was engaged.

The Model X hit a center median and the driver was killed; Tesla said data shows he ignored multiple warnings to retake control of the car, but his family is considering a lawsuit based on the driver’s past complaints about Autopilot’s erratic behavior over the same stretch of highway.

Tesla CEO Elon Musk recently blasted the media for playing up recent Tesla accidents, arguing that traditional cars kill an average of 100 people a day and those stories go underreported.

“It’s super messed up that a Tesla crash resulting in a broken ankle is front page news,” Musk tweeted earlier this week.

Tesla, along with other automakers with driver-assist tech, stresses in car manuals that Autopilot’s various features — which work leveraging road information collected by on-board radar and cameras — require constant vigilance and failure to do so “may result in death.”

That cautionary stance stands in contrast to the somewhat misleading and omnipotent name for the system as well as various enthusiastic comments made by Musk over the years.

“Long term, it’ll be way better than a person. It never gets tired, never has something to drink, never argues with someone in the car,” Musk said at a press conference when Autopilot was added via a software update to Tesla vehicles in 2015.

Long term, Musk may well be proven correct. Waymo, Google’s self-driving car company, has been testing autonomous vehicles for the better part of nine years now and the few accidents reported typically are the fault of humans in other cars.

But the problem with our current state is that “the most dangerous part (of today’s driver-assist tech) is the handoff between humans and computers, because humans are too complacent and lax if the computer seems like it can do it on its own,” says Karl Bauer, executive publisher of Cox Automotive.

“That’s an unfixable problem,” he says. “There’s almost no way around bad human behavior.”

In fact, he cautions that even if automakers suddenly made it more difficult for drivers to disappear into other tasks while driver-assist tech handles the road, some people will find a way around that. 

“Another bad habit of humans is they revel in finding ways to trick a computer,” he says.

True enough. Consider a product called the Autopilot Buddy, a $179 piece of magnetic plastic that fools a Tesla’s steering wheel into thinking your hands are on it. No more nagging alerts. Or chimes. Or … life?

Follow USA TODAY tech writer Marco della Cava on Twitter.

Facebook Comments

POST A COMMENT.