Semiautonomous Cars Have Flaws. That’s Why They Need Tests

These days, modern cars come with sophisticated driver assistance tools, like adaptive cruise control, which maintains a set distance from the car in front, and active steering, which keeps a vehicle in its lane. They can also brake automatically if the driver doesn’t spot a stopped object ahead, and warn when there’s a motorbike hovering in a driver’s blind spot. These features are the start of the promise of our autonomous future, where cars will drive themselves and we will be but passengers in a self-driving world.

Automakers bundle these systems up with fancy names, like Tesla’s Autopilot, or Mercedes-Benz's Drive Pilot. But they're still being refined and developed—and are at different stages of capabilities. In a series of road tests with five high-end cars, including two Teslas, the Insurance Institute for Highway Safety found that “some systems handled some situations better than others,” according to chief research officer David Zuby. And, more crucially, “today's systems aren't robust substitutes for human drivers.”

So until that moment comes—when fully autonomous cars are predictably safe—the IIHS wants to build a safety ranking system, just like how it rates cars for safety in crash tests. Car buyers can use that information to decide which features suit their driving needs, and they will become an increasingly important purchase factor as semiautonomous features make their way into more and more vehicles.

For instance, twenty automakers have pledged to include automatic emergency braking as standard by 2025, a feature that should help reduce rear-end collisions. But when the IIHS drove its group of cars towards a stationary target, they didn’t all stop. The Mercedes E-class, BMW 5 series, and Volvo S90 managed to brake to a halt from 31 mph. A Tesla Model S and Model 3 didn’t, but they did slow down to reduce the severity of the impact, which is the way Tesla claims its system is designed to work.

When adaptive cruise control was turned on, all the vehicles managed to brake for the stopped object, and the IIHS says the Teslas braked earlier and smoother, whereas the Volvo slammed on the brakes at the last second. All the vehicles also managed to stop when the car in front of them changed lane, revealing a stopped object. This is an important test as it’s something that all the owners manuals warn the systems may not respond properly to. The the cars computers are designed to ignore most stationary objects, to avoid slamming on the brakes for every overhead street sign or roadside bench. But that does mean they can miss obstacles that come out of nowhere, like when a Tesla hit a fire truck, parked on the freeway, in January. (The manuals all say the human has to stay aware and ready to take over.)

Having two Teslas in the test allows for some interesting comparisons between the company’s original and updated Autopilot: the Model S was from 2016, with older hardware and software; the Model 3 was new, with Tesla’s upgraded camera suite and latest operating system. When the researchers took the cars from the track to the street, they noticed instances in all of the cars where they didn’t brake properly for stopped cars ahead, except the Model 3 which did spot the cars and stop. But the flip side is the researchers noted 12 times the Model 3 unexpectedly slowed down for no reason, as if it spotted something in its path. When they reviewed video of the drives they worked out at least seven were probably because of tree shadows on the road.

In steering tests, the Tesla Model 3 and Model S stayed in their lanes well, whereas all the other cars needed the driver to grab the wheel to make it around curves on a track test. On the road, the BMW bounced between the lane lines, and Model S and Volvo S90 also struggled when the road got hilly, and lane markings disappeared over the brow. The Mercedes E-Class did better, and the Model 3 the best.

The IIHS says these tests are just the start, and they’ll expand on them as more cars with more features hit the market, and work with international safety bodies like the UK’s Thatcham. But they already highlight the balancing act that automakers are having to pull off. If their systems are too capable, then they risk the driver’s attention wandering, which is a criticism leveled against Tesla and may have led to a fatal collision in northern California in March.

But if they’re too simplistic, they’re just frustrating to use, and drivers won’t bother. And the basic safety systems, at least, do save lives. IIHS says preventing lane departure crashes alone would save 8000 lives per year. Tesla makes bold safety claims for its Autopilot suite, which are hard to check without data, but the company says it does plan to release regular safety statistics, starting later this quarter, as promised by Elon Musk. He also says his cars are going to get more capable with software updates, changing lane for themselves, for example.

And more advanced assistance tools, that really do allow a driver to switch off, or check their phone, are coming from more mainstream automakers. This round of tests didn’t include Cadillac's capable Super Cruise, which uses a camera to monitors a driver’s gaze. Audi’s Traffic Jam Pilot will take total control of the car under 37 mph, but won’t be available in the US because of the difficulty of dealing with different state laws.

The ultimate reality is most drivers' first encounter with autonomy is going to be with a driver-assistance tool taking on part of the job. So being able to make an informed decision about which car is best, based on how they work, will make everyone feel like a better co-pilot.


More Great WIRED Stories

Read more: https://www.wired.com/story/semi-autonomous-cars-safety-tests/

Posted in PoliticsTagged ,

Post a Comment