New cars can stay in their lane—but might not stop for parked cars

In recent years, a number of car companies have—like Tesla—begun offering driver assistance systems that offer lane-keeping as well as adaptive cruise control. This might seem like a big step toward a “self-driving car,” since a system like this can travel down the freeway for miles without human intervention. But a new report from AAA underscores the limitations of these systems.

Its most dramatic finding: the advanced driver-assistance systems (ADAS) on the latest cars still struggle to avoid collisions with parked vehicles. They tested cars from BMW, Kia, and Subaru; none consistently avoided running into a fake car partially blocking the travel lane.

The researchers also examined the ADAS in the Cadillac CT6 and the Ford Edge, but these cars’ systems weren’t included in the parked-vehicle test because their driver assistance systems wouldn’t engage on AAA’s closed course. They were included in other tests conducted on public highways.

“All test drivers reached a general consensus that combining adaptive cruise and lane-keeping functionalities in a single system did not consistently enhance the driving experience,” the report said. The vehicles made mistakes often enough that drivers often found the experience nerve-wracking rather than relaxing.

Greg Brannon, a co-author of the AAA report, argues that a fundamental challenge with this kind of system is the need to maintain alertness. Human beings are terrible at paying continued attention behind the wheel of a car that mostly drives itself. So when (not if) these vehicles make a mistake, there’s a heightened risk that the driver won’t be paying close enough attention to recover safely.

Why ADAS systems keep hitting parked cars

You might think that cars with adaptive cruise control or emergency braking capabilities would be able to stop for parked cars and other obstacles in the roadway. But often they fail to do so. To see why, it’s helpful to understand a bit about how these systems work.

Most of the early adaptive cruise control systems sold 10 to 20 years ago were based on radar. And as we explained two years ago, radar has some important limitations:

Radar has low angular resolution, so it had only a crude idea of the environment around the vehicle. What radar is quite good at, however, is figuring out how fast objects are moving. And so a key strategy for making the technology work was to ignore anything that wasn’t moving. A car’s radar will detect a lot of stationary objects located somewhere ahead of the car: these might be trees, parked cars, bridges, overhead signs, and so forth.

These systems were designed to work on controlled-access freeways, and, in the vast majority of cases, stationary objects near a freeway would be on the side of the road (or suspended above it) rather than directly in the car’s path. Early adaptive cruise control systems simply didn’t have the capability to distinguish the vast majority of objects that were near the road from the tiny minority that were on the road.

So cars were programmed to focus on maintaining a safe distance from other moving objects—cars—and to ignore stationary objects. Designers assumed it would still be the job of the human driver to pay attention to the road and intervene if there was an obstacle directly in the roadway.

Today’s driver assistance systems are more sophisticated than early systems that came on the market 10 or 20 years ago. They’re no longer just radar-based; most have cameras, and, presumably, they try to use both radar and cameras to identify potential objects in the road ahead.

And indeed, the newest cars do seem to do better on this score than past systems. The best system AAA tested in this respect was a 2020 Subaru Outback. AAA performed three test runs on the Outback with a starting speed of 30 miles per hour. The vehicle stopped in two out of three test runs. In the third run, the vehicle at least slowed down before striking the dummy car.

On the same test, a 2019 BMW X7 stopped in one out of three runs, while a 2020 Kia Telluride hit the dummy vehicle in all three 30mph runs. Both cars decelerated somewhat prior to impact in the other runs.

In other words, all three vehicles have some capacity to detect imminent collisions and at least mitigate them—which represents an improvement over early radar-based systems. But none of them was sophisticated enough to consistently prevent crashes with parked cars.

Tesla Autopilot isn’t the only system struggling with parked vehicles

This provides a bit of vindication for fans of Tesla, which has gotten the most media attention for this kind of ADAS failure. At least three Tesla owners have died when Autopilot crashed into a concrete lane divider (in one case) or a semi crossing the road (in two others). Several other Tesla cars have had non-fatal crashes into parked police cars or fire trucks. AAA’s tests make it clear that this isn’t a problem that’s limited only to Tesla vehicles.

I asked AAA’s Greg Brannon if he knew of ADAS-related crashes involving other car brands; he said he didn’t. It’s not clear why. Perhaps Autopilot has simply been on the market longer. Or maybe Tesla crashes get more media coverage, and crashes involving other automakers have flown under the radar. Brannon noted that the National Transportation Safety Board has investigated several fatal crashes involving Tesla vehicles using Autopilot. The agency doesn’t seem to have conducted any investigations involving other carmakers’ ADAS technology.

That may change in the coming years as other companies sell more and more cars with Autopilot-like capabilities—since other car models seem to have the same basic limitation as Tesla’s Autopilot.

Highway performance

The parked car test was the most clear-cut failure of these ADAS systems. But AAA’s testing went far beyond that scenario. On its closed course, AAA tested whether the Subaru, BMW, and Kia vehicles could stay in their lanes and respond when a vehicle ahead hit the brakes. All three cars handled these situations with little difficulty. (Again, cars from GM and Ford had geofences that prevented the ADAS from engaging.)

The researchers also performed qualitative testing of the five cars’ performance on public highways. Over 800 miles of testing, drivers in the BMW, Ford, Kia, and Subaru cars intervened dozens of times—most often because the vehicle seemed to be drifting out of its lane.

Cadillac’s Super Cruise was different. Its drivers intervened only eight times in 800 miles of driving over concerns that the vehicle wasn’t staying in its lane properly. However, the system seemed overly conservative about proactive disengagement. There were 37 times when Super Cruise had “unexpected or erroneous system disengagements.”

“Test drivers were sometimes taken by surprise and were required to retake full control in the middle of critical situations with little to no advance notice,” the AAA report says.

Super Cruise is different from most other driver assistance systems because it uses a user-facing camera to ensure drivers are paying attention. (Subaru has a similar system called DriverFocus, but Brannon said drivers saw little indication the camera was being used.) The Cadillac’s camera monitored the driver’s eyes and tried to estimate whether the driver was paying attention to the road. AAA found that this system was far from perfect; on a number of occasions, Super Cruise forced the driver to take over because it wrongly thought the driver wasn’t paying attention.

All that said, there’s a lot to be said for Cadillac’s approach, which we’ve praised before. Super Cruise is conservative about its own capabilities, handing back control to the driver at the first sign of trouble. This not only minimizes the chances that the system itself will make a mistake, it may also prompt drivers to pay closer attention to what the vehicle is doing.

“Constant monitoring and intervention”

By contrast, AAA found the Kia Telluride allowed users to go as long as two minutes without touching the steering wheel. Even worse, dashboard indicators were “easy to miss due to inconspicuous icons,” creating the risk that drivers could become confused about whether the ADAS was active.

The Subaru Outback would “frequently issue a lane departure warning while disengaging simultaneously.” This would sometimes occur on a sharp curve, creating a heightened risk of a crash if the driver didn’t take control quickly.

Ultimately, none of these systems earned rave reviews from AAA’s drivers. Drivers reported that the systems “could sometimes increase perceived driver workload because constant monitoring and intervention in many cases was required.”

This may be a fundamental problem with this approach to driver assistance technology. The ADAS is supposed to do most of the driving, but the human driver is supposed to still monitor the system and make sure it doesn’t make mistakes. But our brains aren’t wired for this level of monotony. Monitoring a system that works correctly 99 percent of the time is in some ways harder—not easier—than just driving the car yourself. And monitoring a system that works correctly 99.9 percent of the time is even harder, because it’s that much easier for our brains to get distracted by something else.

https://arstechnica.com/?p=1697228