Even After $100 Billion, Self-Driving Cars Are Going Nowhere

T-Shirt with Stop Sign can possibly make driverless cars stop

(Changed the title b/c it was needlessly confusing)

Justin Banner

It is sort of funny that the Waymos can be fooled by such a simple “trick,” but we’re also talking about something that is programmed to recognize the construction sign. While it seems dumb on the surface, it also means that the Waymo driverless system is working mostly as intended—it's seeing a seemingly temporary stop sign (we think a guy wearing a stop sign is technically temporary, as he isn't rooted into the earth with a signpost) and stopping. It does beat the alternative, like that self-driving Cruise vehicle that ran from the cops and Tesla’s Full Self Driving (FSD)—that really isn’t self-driving at all—that runs into curbs and crashes into emergency vehicles. Now, it stands to reason if the Waymo can be tricked by fake stop signs, it could potentially be tricked into, say, exceeding local speed limits by a T-shirt with a higher number on it—if, you know, Waymo and other testers didn't limit their vehicles' top speeds regardless of signage—we'll leave you with one more kernel of knowledge from jasonbcarr, conveyed in the text from misterphx: “Don’t use this magic trick for evil, Stay safe out there, kids!”

For self-driving cars, the free ride is over

Andrew J. Hawkins

In the span of a few weeks, federal safety regulators have opened investigations into top operators of driver assistance and autonomous vehicles.

For years, autonomous vehicles have operated in relative obscurity. With few vehicles on the road and a laissez-faire attitude among government regulators, automakers and big tech firms have been free to test — and even commercially deploy — with little oversight.
Well, those days are done. In rapid succession, the National Highway Traffic Safety Administration (NHTSA) has opened investigations into almost all the major companies testing autonomous vehicles as well as those that offer advanced driver-assist systems in their production cars. Tesla, Ford, Waymo, Cruise, and Zoox are all being probed for alleged safety lapses, with the agency examining hundreds of crashes, some of which have been fatal.

The new investigations signal a new — and perhaps more antagonistic — phase in the relationship between safety regulators and the private sector. The government is requiring more data from companies, especially around crashes, in order to determine whether the industry’s safety claims live up to their hype. And the companies are finding that the proliferation of smartphones with cameras is working against them, as more videos of their vehicles behaving unpredictably go viral.

New concerns about crossing guards' close calls with driverless cars

An NBC News investigation is raising safety questions about the ability of driverless cars to recognize pedestrians in crosswalks, and a member of Congress is demanding action because of it.

NBC Bay Area's Bigad Shaban has the story.

Waymo recalls 600+ self-driving vehicles for software, map updates after one strikes a pole in Phoenix

Dominique Newland

Waymo is recalling more than 600 self-driving vehicles after one of them struck a telephone pole in Arizona.

The recall includes 672 vehicles, which is the entire fleet.

...So you will continue to see Waymo vehicles on Valley streets. In fact, just last week, Waymo expanded its service into North Phoenix, where it now goes as far as the Desert Ridge area.

Waymo recalls 600+ self-driving vehicles for software, map updates after one strikes a pole in Phoenix

Dominique Newland

It's funny, early in my driving career, not long after passing my test, I managed to reverse into a pole, even though I knew it was there. It was a case of not having sufficient awareness of the size of my vehicle, nor of its position relative to this object. I would have assumed this would be a primary advantage of a self-driving vehicle, equipped with cameras and sensors, avoidance of elementary mistakes would be a reason to wish to own one. Or not. 

[Laird]

Self-driving cars 'generally' safer but more at risk of accidents at dusk and dawn, US researchers find

By Brianna Morris-Grant for the ABC on 19 June, 2024.

Self-driving cars are better at routine tasks but are five times more likely to be involved in an accident in low-light, new research has found.

The comparison of accident data from 2,100 self-driving vehicles (SDVs) and 35,133 human-driven vehicles (HDVs) found the automated cars were generally safer when staying in lane or adjusting to traffic.

The peer-reviewed study, conducted by researchers at the University of Central Florida, also found SDVs were less likely to be involved in rear-end and sideswipe accidents.

"However the odds ratio of an [self-driving vehicle] accident happening under dawn/dusk or turning conditions is 5.250 and 1.988 times higher, respectively," the report said.

Self-driving cars 'generally' safer but more at risk of accidents at dusk and dawn, US researchers find

By Brianna Morris-Grant for the ABC on 19 June, 2024.

I can't find a link to the study these US researchers did?

Would be interesting to see where they get their money...

edit: Ah I think this is the paper.

[Laird]

Ah I think this is the paper.

Oh dear. Unless I'm misinterpreting that paper, the news article misreported its results. The paper started with a set of self-driving accidents, and a set of human-driving accidents, and then compared the two to find, essentially, the ratios of different types of accident between the two sets. All it can tell us in terms of likelihood, then, is, given any random self-driving accident, and any random human-driving accident, how much likelier (or less likely) the self-driving accident is to have certain characteristics than the human-driving accident. It can't tell us anything about the absolute likelihood (i.e., per million driving miles) of self-driving versus human-driving accidents, or even the absolute likelihood of the occurrence of an accident having certain characteristics, [whether] for self-driving versus [or] human-driving accidents, as the article mistakenly seems to imply it does.

Is that your take too?

[Edited for clarity - editing brackets and strikeout indicate edits]

Oh dear. Unless I'm misinterpreting that paper, the news article misreported its results. The paper started with a set of self-driving accidents, and a set of human-driving accidents, and then compared the two to find, essentially, the ratios of different types of accident between the two sets. All it can tell us in terms of likelihood, then, is, given any random self-driving accident, and any random human-driving accident, how much likelier (or less likely) the self-driving accident is to have certain characteristics than the human-driving accident. It can't tell us anything about the absolute likelihood (i.e., per million driving miles) of self-driving versus human-driving accidents, or even the absolute likelihood of the occurrence of an accident having certain characteristics for self-driving versus human-driving accidents, as the article mistakenly seems to imply it does.

Is that your take too?

Yeah, this is where I also felt the paper was flawed, though you looked deeper into it than I did admittedly. 

But of course any report that gives some credit to driverless cars will be spread around, too much money on a questionable idea to back out now...

[Laird]

Aren't you of the view that eventually, self-driving cars will be capable though? I ask because if that is the case, then I don't know why you'd expect anybody to be backing out, unless you mean pausing real-world activity until the technology is perfected.

By the way, I slightly edited my last post to clarify a badly-worded part.