As driverless cars falter, are 'Driver Assistance' systems in closer reach?
With investigations and lawsuits over accidents adding scepticism towards fully driverless technology, car companies are betting on systems that take some, but not all, control
Published: Wed 19 Oct 2022, 9:31 PM
Matt Williams/The New York Times
Imagine heading eastbound on, say, I-95 when you and your pickup encounter red brake lights for miles ahead. Now imagine not touching the brakes or steering wheel and, instead, sitting back and letting the car deal with it.
For the next hour of stop-and-go slog, the truck's system does the driving: anticipating slowdowns, accelerating, braking and steering on its own. When traffic eases up, the pickup climbs to a selected 70-mph speed and executes automated lane changes. The system checks blind spots and flashes turn signals.
But this truck isn't designed to be an entirely driverless one. The truck's infrared driver-monitoring camera watches for eye and head position. You can glance at a passenger or consult a navigation screen - but if you look away for more than a few seconds, LEDs illuminate blue on the steering wheel rim, a transparent command to get your eyes back on the road. If you ignore prompts, the rim flashes red, and the system disengages and reverts to hands-on control.
As Tesla faces a federal investigation and lawsuits over fatal accidents involving its Autopilot system, shaking public confidence in robotic cars, could a pared-down approach like the one described - variously called “partial autonomy” or “driver assistance” systems - be the more realistic future of hands-free driving?
This type of system, more like a no-nonsense chaperone than one you would find in a fully robotic car, is a necessary component for top scores from the Insurance Institute for Highway Safety's forthcoming ratings of partial-autonomous tech; high ratings from the independent nonprofit are prized. And though General Motors is taking the lead with its Super Cruise system, GM is not alone; Ford, BMW and Mercedes-Benz are making similar attempts.
Super Cruise combines minutely detailed, 3D laser-scanned roadway maps with cameras, radar and onboard GPS. By the end of this year, the company intends to expand the system's network to two-way undivided highways for the first time and double its total operational domain to 400,000 miles. Doing so would allow hands-free driving on some of North America's most epic byways, such as the Pacific Coast Highway, Route 66 and the Trans-Canada Highway.
None of this means that car companies are abandoning the dream of fully autonomous cars. In addition to Tesla, GM's Cruise division, Alphabet's Waymo and Argo AI continue to develop and test robotaxis, with human safety operators aboard, in cities including Miami and Austin, Texas. Cruise has begun charging fares for robotaxi rides in modified Chevy Bolt EVs in San Francisco and is mapping Dubai with the hope of starting a robotaxi programme there next year.
But as fully driverless technology has faltered, so has faith in such technology.
“The systems work great, right up until they don't,” said Bryant Walker Smith, an associate professor in the Schools of Law and Engineering at the University of South Carolina, who has advised the federal government on autonomous vehicles. “We don't have a full sense of the winning combo to cover most of the driving people do.”
In addition, Cruise temporarily halted and recalled its 80-car fleet for a software fix following a two-car collision that injured two occupants in June. A GM public filing noted that law enforcement had cited the human-driven car for being mostly at fault, including for speeding, and that the company's robotaxis had, before the collision, safely executed nearly 125,000 left-hand turns through gaps in oncoming traffic.
David Harkey, president of the Insurance Institute for Highway Safety, said the industry's reality check over the technical challenges, and attendant public disillusionment, is masking genuine progress. For one, the building blocks of partial-autonomy cars are already in every showroom. Automated emergency braking is standard on every new car as of September, thanks to a voluntary agreement forged in 2016 among automakers, the Insurance Institute and the National Highway Traffic Safety Administration.
Such radar- or camera-linked brakes have cut police-reported rear-end collisions by a striking 50 per cent, Harkey said, according to their research, adding that automated braking for pedestrians has reduced the number of car-human collisions by 30 per cent versus cars without the feature. And anti-lock brakes; cameras, radar and ultrasonic sensors to manage blind spot and lane departure monitors; and adaptive cruise control have become standard as well.
“We saw that as beneficial tech, and the same will be true for some new tech. We will continue to push to get more features on more models to save more lives and prevent crashes,” Harkey said.
The trick, he said, is to build on that promise, with systems that measurably boost safety but keep human drivers in the loop.
“These are driver assistance systems, not driver replacement systems. Some consumers don't know the difference,” he said.
In Germany, Mercedes has begun pushing boundaries with its new Drive Pilot, which legally allows a driver to perform nondriving tasks - checking email, even watching a movie - but monitors the driver and alerts when to retake the wheel.
The nonprofit institute divides these sorts of systems into levels of automation, from zero (no automation) to five (full automation). Experts see Level 3 (some automation, but with a driver at the ready) as the diciest of the levels, a limbo zone compared with Level 5 cars that are truly robotic. For now, Drive Pilot can operate only on certain highways at up to 37 mph.
Mercedes is seeking certification to offer the system in the United States next year.
The next test is GM's Ultra Cruise, which the company intends to debut on the Cadillac Celestiq, a six-figure electric flagship sedan, late next year. The system is designed to ultimately deliver hands-free driving on 3.2 million miles of roadway - nearly every inch of paved road in the United States and Canada.
Jason Ditman, Ultra Cruise's chief engineer, said the systems must work with full transparency and consistency to instill confidence among owners and the public.
“If you think it's hard to get someone to let go of the steering wheel on highways,” Ditman said, imagine a snowy country lane or crowded city street.
GM says Ultra Cruise will stop and start at traffic lights and stop signs, autonomously follow navigation routes, do close-object avoidance of vehicles and pedestrians, even self-park in driveways. The machine learning system will identify dicey scenarios and upload data to continuously improve performance, and GM can remotely shut down use of the system on any road where the company is not confident of performance.
GM says the system will eventually handle about 95 per cent of driving, aside from complex scenarios such as multilane roundabouts.
Despite high-profile crashes, Smith, the professor, believes that excessive focus on drawbacks of driver-assistance systems distorts the true crisis: Nearly 43,000 Americans died last year in motor-vehicle crashes, which kill roughly 1.3 million people worldwide annually.
At least 100 people will die on US roads today, and we're not going to hear about them,” he said. “Chances are that not one will be killed in connection with a driver-assistance system.”
This article originally appeared in The New York Times