After partially faulting Tesla Inc.’s automation system for a 2016 fatal crash, U.S. safety investigators last year called on carmakers to do more to ensure drivers stay engaged as next-generation cars start to steer themselves.
Since then, the National Transportation Safety Board has opened investigations of three new cases — two of them involving Tesla vehicles — that call into question the progress that’s been made in guarding against motorist misuse of semi-autonomous driving technology.
Tesla, a pioneer in driver-assistance technology with its Autopilot system, has lagged behind automakers including General Motors Co. in embracing driver monitoring. While the electric-car maker still relies on technology that federal investigators said was too easy to sidestep, it’s now working on unspecified improvements to its vehicles, according to the NTSB.
“They have indicated that they have already made some improvements and are working on additional improvements,” agency spokesman Peter Knudson said, in the first indication that the company is contemplating more changes to its driver-assistance system. NTSB highway investigators have been in contact with Tesla technical staff, he added.
A Tesla spokeswoman declined to comment on the NTSB’s recommendations for improved driver monitoring.
The difficulty of keeping drivers in automated vehicles engaged — combined with the broader safety benefits of ensuring people in traditional cars and trucks aren’t peering at electronic devices or nodding off — is a growing safety concern that’s spurred several car companies, including GM and Subaru Corp., to deploy infrared cameras in the cockpit trained on the driver to track head and eye movement.
Driver-monitoring technology is needed for any vehicle that needs humans to handle part of the driving task, said Bryan Reimer, who studies driver behavior at the Massachusetts Institute of Technology. This includes conventional vehicles without driver-assist systems, cars that guide themselves for some periods without human inputs, such as Cadillac models featuring GM’s Super Cruise, and self-driving cars with people serving as safety monitors.
Motorists today are bombarded by distractions, from mobile phones to in-dash navigation systems, Reimer said. “Drivers need help making better decisions,” he said.
The NTSB is investigating two crashes this year in which Tesla drivers were using Autopilot. The system can automate steering and follow traffic in some conditions, but the company warns drivers they must monitor it at all times. The system isn’t designed to be fully autonomous and can’t detect some objects in its path, according to the NTSB.
In the most recent case, a Model X slammed into a concrete highway barrier on March 23 in Mountain View, California, killing the driver Walter Huang. His family has hired Minami Tamaki LLP to explore legal options, the firm said on April 11.
Tesla said in a blog post last month that Huang, 38, didn’t have his hands on the wheel for six seconds prior to striking the barrier where lanes split on the freeway.
“The driver had received several visual and one audible hands-on warning earlier in the drive,” the company said in the March 30 blog post.
The NTSB is also investigating a Model S crash that occurred on Jan. 22 near Los Angeles. In that case, the car rammed into a fire truck that was parked on a freeway after responding to an emergency. The driver, who wasn’t hurt, told people on the scene he was using Autopilot, according to a tweet by the union for Culver City firefighters.
After the 2016 fatal crash involving a Model S driver, Tesla modified Autopilot to improve sensors and also increased the frequency of warnings to drivers who didn’t have their hands on the wheel.
“What Tesla has is basically a sensor that just detects whether your hands are on the wheel,” said Mike Ramsey, an analyst at researcher Gartner Inc. “If it doesn’t detect anything on the wheel for a certain amount of time, it first gives a visual warning, then an audible warning, then the car starts slowing down. It’s somewhere in the neighborhood of 10 seconds or longer. At 70 miles per hour, that’s a long time — a lot can happen in that period of time.”
The NTSB found last year that the driver in the 2016 crash had barely touched the steering wheel and ignored multiple warnings over a 37-minute period before his Model S plowed into the side of a truck. The agency — which has no regulatory powers but makes recommendations to government and industry — wrote to Tesla and five other car companies to recommend better means of driver monitoring.
Tesla has installed an inward-facing camera above the rear-view mirror in its new Model 3 sedan, but hasn’t said whether it could be used to monitor drivers.
The issue of attentiveness also arose in the first U.S. pedestrian fatality involving a self-driving car. In that case, an Uber Technologies Inc. sport utility vehicle killed a woman in Tempe, Arizona, on March 18. The NTSB is looking at multiple issues related to how the car’s software and sensors performed, and an interior video showed that the safety driver in the SUV wasn’t watching the road for about five seconds leading up to the collision.
Uber’s self-driving system wasn’t covered under the NTSB’s recommendations, but the agency believes its call for better monitoring is relevant for all autonomous vehicles that still keep a human in the loop, according a March 8 letter to Daimler AG’s Mercedes-Benz.
Using technology to monitor driver behavior is beginning to gain traction with some automakers and safety organizations. The European New Car Assessment Programme plans to add incentives to its safety ratings regime for carmakers that equip new vehicles with driver-monitoring systems starting in 2020.
In March, Subaru said it will equip higher-end versions of its redesigned 2019 Forester with an infrared camera facing the driver. That will be paired with facial recognition software to monitor drivers for signs of fatigue or distraction, and to warn drivers with a beeping noise after three seconds of inattention. Ignoring the warnings will prime the Forester’s suite of driver-assist technologies, such as automatic emergency braking, to slow or stop the vehicle to avoid or mitigate a collision.
The issue of driver attentiveness “is really a critical area,” said MIT’s Reimer, whose team is testing a fleet of vehicles rigged with cameras to observe how drivers perform in real-world situations. “We are clearly picking up technology in the car, portable phones, et cetera, at rates that are far and above what should be socially acceptable.”
By Alan Levin, Ryan Beene and Keith Naughton