Tesla investigation deepens after more than a dozen US ‘Autopilot’ crashes
theguardian.com
theguardian.com
Don't let your hate for Musk as a person cloud your judgement. These things are not comparable.
I don’t dislike the guy, but this is irresponsible.
Do you think that the average consumer knows this much about how plane autopilots work? If not, how is it relevant?
How about calling your system Full self driving and then sticking with a promotional video that says the driver is only there for legal reasons
https://www.teslarati.com/tesla-autopilot-fsd-bias-nhtsa-saf... (obviously a biased source -but worth a read)
"The same goes for her stance regarding CEO Elon Musk. In March 2020, for example, Dr. Cummings seemingly joked about needing someone to stop her from punching Elon Musk in the face. "
"It should be noted that Dr. Cummings’ seat at Veoneer was not disclosed when she published a paper (which was later updated to remove inaccurate details about a fatal Tesla crash) criticizing systems such as Autopilot for their possible dangers. And so far, the incoming NHTSA senior safety adviser has not shared if she would be leaving her post at the Swedish LIDAR company, especially since she would soon be advising a US safety agency on driver-assist systems that adopt both LIDAR and non-LIDAR solutions. "
"Missy Cummings receives around $400,000 worth of stock in Swedish LIDAR company Veoneer per year as compensation for her role as a non-employee member of the board."
Kinda looks like she did:
https://www.automotiveworld.com/news-releases/veoneer-resign...
Tesla gets called out because their autosteer is so good at lane keeping, comparatively speaking. Their new high resolution radar [2] (Phoenix) they’ve registered with the FCC, as well as a retrofit campaign across the fleet, will likely resolve this gap. If so, I’m unsure what happens when other automakers continue to have this deficiency as their autosteer capability improves but they don’t have high res forward facing radar. Will NHTSA hold legacies to the same standard?
[1] https://www.caranddriver.com/features/a24511826/safety-featu...
[2] https://www.autoevolution.com/news/tesla-backpedals-on-the-u...
Is a retrofit confirmed or is it just your assumption? They even sold hundreds of thousands of "vision only" cars (including mine) with Musk and Karpathy claiming radar is useless for them. Funny how they've walked it back with this new radar news.
Tesla Vision was a successful strategy to keep shipping units during supply chain issues (inability to source radar units), but it is clearly not sustainable based on the Vision codebase maturity reflected by operational performance.
The driving dynamics of different vehicles result in different outcomes not just due to those dynamics themselves, but also in the driving behavior they elicit. (and even further, in that drivers of different driving habits often exhibit selection bias in the vehicles they choose)
Yeah, Tesla gets "called out" because their technology is so good, not because their implicit marketing claims the cars can drive themselves.
There are full time shills whose entire focus is demonstrating on YouTube how the cars can drive themselves. You don't think that has something to do with this? Arguing "That's FSD this is Autopilot" misses the entire point. Elon has 100M followers on Twitter; go scan through any thread. People think they've solved FSD, people think they have humanoid robots, people think he already owns Twitter for crying out loud.
So, yes, if Mary Barra got in her Chevy Bolt and told a reported the car can drive itself and someone died doing the same thing, I think GM would be held to the same standard.
Tesla's don't.
If the other manufacturers start throwing “full self-driving” around to describe their cruise control/lane keeping/automatic breaking, and people start crashing because they rely on that, then yes. I expect there would be some regulatory activity.
Tesla articles get views because of both the Tesla haters and fanboys, makes financial sense for news sites to publish them. That increased press leads to increased investigations.
Those other manufacturers have more robust legal teams, so they advertise their cars capabilities a bit more honestly.
I don’t think the average person knows that an aircrafts autopilot requires constant monitoring.
If they're not the same thing why equate them?
Except plenty of modern cars don't just have fixed velocity cruise control. They have adaptive cruise control that will speed up or slow down to match the velocity of the car ahead of you. They also have lane centering that will keep your car in its lane. That sounds an awful lot like Teslas.
Here's the thing - people get in accidents all the time when assistance features like ABS, lane keeping, traction control, TACC, good-old-fashioned dumb cruise control, etc. are actively altering the vehicle's performance and behavior.
So when we consider Tesla's AP system, it's really just TACC + lane keeping. Something that is common on a wide variety of vehicles today. Why then is Tesla specifically being focused on? All these other systems have similar failure modes, including, but not limited to: Not stopping for a stopped vehicle far ahead, not slowing for or avoiding objects in or near the lane, etc.
These are assistance features. Thus, in all of these scenarios, it is likely that chief responsibility for the accident lies with the human behind the wheel. We should be studying all OEM systems, not just Tesla's. And specifically, we should be looking at whether the system made the situation worse through its actions than if it had taken no action (akin to a distracted human driving on dumb cruise control).
If it is the case that a Tesla on AP will ram right in to a firetruck flashing its lights while stopped in the lane directly ahead, it is quite likely that many other systems do the same. We just don't really talk about them, because it casually falls under the realm of "human wasn't paying attention". They're not special. We all sort of implicitly accept that. But when it's a Tesla, it's special.
In any case, we'll get over it eventually. This technology is not unique to Tesla and will only become more commonplace. It will continue to get better at avoiding things despite the hoards of distracted drivers.
https://electrek.co/2022/06/08/tesla-files-use-new-radar-con...
Because before Tesla launched the feature, "Autopilot" was known in the general consciousness to mean a specific thing. Autopilot is the thing on planes that lets the pilot do other tasks (like navigational routing).
Other car manufactures are smart enough to not brand their cruise control as "Self Driving" or the like.
If I sold a brand of drain cleaner under the name "Laundry Detergent" that would rightly result in some consumer protection investigations.
Not that I agree with your statement, but the "general consciousness" would be wrong in that case.
Autopilot in most aircraft will gladly fly you into the side of a mountain or other aircraft.
Um, what? A driver can override Tesla autopilot at any time.
I was talking about freeing the pilot from maintaining situational awareness. This is not supported for drivers of Tesla vehicles; the term "autopilot" is misleading in that context.
Should there be in investigations for motorcycle companies making unsafe bikes? Or is it more the case of certain people in general being predisposed to taking more risks and not realizing the consequences?
Why? Because most anyone already understands that motorcycles are dangerous, and those that choose to ignore risks get fucked. Why should Tesla be any different?
I haven't used the Ford system but I have used the Tesla system and it requires you to move the wheel at a pretty annoyingly and regular frequency to remain in autopilot mode. I believe the cabin camera can also detect when you are on your phone or looking away - but i'm not sure if this is rolled out to non-FSD cars.
I think the frequency that you have to prove you are paying attention is about every 30 seconds
This is so disingenuous.
Go on YouTube and search for people "self-driving" in their Teslas, then do the same for people doing anything remotely the same with standard cruise control.
Tesla and its fanatics want it both ways: it's both the most sophisticated self-driving software on the planet, and only regulations are preventing it from being widely used and reducing traffic accidents, but when accidents happen it's just "cruise control". Pick one.
>so much a company can do
Yes, therefore the product should not be legal if these error's can't be minimized or the magnitude of the consequences are disastrous (random unpredictable crashing etc). "Errors" are not all created equal. Some are catastrophic and occur at random.
Tesla does have safeguards to prevent people from dozing off on autopilot: there's sensors on the steering wheel, and apparently some models have front facing cameras. Some people put weights on the steering wheels, but at that point people are taking deliberate efforts to bypass safeguards. If someone jams a shim in their microwave latch, they can turn it on while the door is open and fry themselves. That's comparable to putting weights on Tesla's steering wheel.
What are these features for, if they require 100% attention from the driver to monitor? At that point, why not just drive the car? Is there a goal for these features to take incrementally more and more proportion of the responsibility away from the driver, such that the driver can incrementally devote less attention to monitoring? We all know the answers to these questions. We know what the goal is here.
But the problem is, human's cannot sort of, kind of, partially pay attention to the road and the car. And any feature that propagandizes them that it will take care of incrementally more of the driving responsibility will result in less (rounded to none) attention devoted to by the driver. It's a binary proposition beyond some threshold as mutli-tasking is a myth. Do you see the problem?
As the features appear to be working better and better, the more people will devote less attention to the road even in some bounded range of their drive but this is all it takes for catastrophic errors (that will and are happening because of the nature of failure modes of AI) to occur. And these errors are qualitatively different than the ones humans make. At any given point in time the care can and will just do something completely stupid and dangerous. Swerve into a median, crash into an emergency vehicle, turn sharply into a car in another lane. There's documented occurrences of all of this. When these random catastrophic errors occur, the human will be on the margin paying less attention to the situation than if it was simple cruise control or normal driving.
Over the last 6 years, there's been 16 collisions, one of them fatal. How many of those accidents would have still happened if people were using adaptive cruise control instead of Autopilot? How many were averted because autopilot is more sophisticated than traditional cruise control. Jumping straight to the conclusion that all of these collisions were due to autopilot, and not considering the fact that it might have been the same or even better than the status quo is naive.
Its funny how people rightfully criticize the government and laugh at videos of senator grilling the CEOs of Facebook or Google with stupid questions, but then as soon as a company does something that they don't like, they somehow expect those people to get their shit together and sensibly regulate something as nuanced as autopilot?
For technology to get better, it needs to be stress tested - in this case that means accidents and people dying. Which is pretty much on par with development of car technology since its origin. The warnings on autopilot are there. If you don't want to take the risk, don't use it.
You don't buckle people in and yeet them against the wall that is your optimism! Half the bloody challenge of testing is setting up the conditions for actually getting meaningful data!
This won't be a problem when the regulation equates to: "get these death traps off the road, now". Or, "disable your not-intelligent pseudo-self driving now".
I'm not talking about working on the technology, I'm saying if it doesn't work get it off the market.
It is an absolutely ridiculous view. I really don’t want a shitty company be able to freely test on the streets their shitty technology that can cause the death of innocents. Are you seriously saying that they are free to increase their profits over the lives of people? On what basis is that ethical or moral? Especially when we talk about a company that jumped into an arguably technically inferior solution by having an incentive to develop their “self-driving” based on visual input only. Their competitors at least take it seriously enough and realize that just because humans can manage with a vision-only system (though a goddamn moving one, not static), lidar and other sensors are required on cars, because ML is nowhere near the capacity of even a worm’s brain, let alone what a human has.
Tesla labeling a feature "Full Self Driving" certainly isn't helping the situation when the feature clearly does not deliver what the name of the functionality implies.
Popular understanding of the word aside, Tesla often throws around phrases like "full self driving" in the same sentence as "autopilot".
Or did they get rid of that? It sounded like a good compromise to me.
There are tutorials on social media for how to circumvent the driver monitoring for 'autopilot'. See for example, the instance in the article of a Tesla going down the highway with the driver in the back seat.
Tesla touts (and sells as a separate option) "full self-driving" capability. They tout that it can make turns on city streets, obey street lights, yield to city traffic, change lanes, park, etc. This is not intended for the use case of on the highway, well-supervised, between exits.
IMHO Tesla has repeatedly overmarketed their actual capabilities. It's clearly starting to catch up with them.
It is, definitely. How come they advertise it as FSD (and I know they very much circle around the exact wording, but a good Litmus test to see what is the actual expected output from the ads is to ask a random person on the street and they will definitely think that they are full self driving, while most other cars have some basic lane assist, etc)?
Their marketing is absolutely malicious, what else would keep up their massive hype bubble?
If you are a company building cooking knifes, but sell knifes as sharp as razor blades and people start injuring themselves super badly in their kitchen, saying that "people should be more careful with their kitchen knives" won't get you very far in court and in the public's eyes.
> Tesla cars come standard with advanced hardware capable of providing Autopilot features, and full self-driving capabilities—through software updates designed to improve functionality over time.
And this is the stuff that is filtered through their communications staff; the stuff Musk has said is even more egregious.
An Apple employee was killed by his Tesla, which always veered into a highway barrier after an OTA update (he was distracted one day and didn't prevent the crash, died as a result).[0] The video above also involves the results of an OTA update.
There are more reports of dangerous Autopilot behavior than any other brand's cruise control that I'm aware of.
The problem here isn't neglectful driving, it's that you may be driving in a straight line on a highway for half an hour, distracted and "lulled" by the monotony and predictability of the road, and in a split second your car veers off and tries to kill you because it no longer knows where the lane is.
Now drivers need to not only pay attention to the road and over cars, but also be ready to react instantly with zero warning if the car does something unexpected and dangerous.
[0]: https://www.kqed.org/news/11801138/apple-engineer-killed-in-...
Essentially, autopilot's main value proposition is that you don't need to pay attention to it -- and that's what makes it so dangerous.
She actually ended up disabling some of them. She explained that they "made driving too easy" and it was too hard for her to keep her attention on the road while using them.
There is a continuum between actively driving the car and falling asleep in the back seat, and AP puts you somewhere on the left side of the middle. It's generally great stuff.
All the point-to-point full-self-driving BS is meh, I don't really care about it, but give me world-class TACC and lane keeping (original autopilot) and I'm a happy commuter.
I largely agree with you here, having driven a car with adaptive cruise control in stop-and-go traffic and finding it to be a total game changer.
That said, I'd argue there's a huge difference between having the car micromanage the accelerator+brake versus having it micromanage the accelerator+brake+steering. With the former, the driver must remain engaged with the driving task (even though the most tedious aspect is taken care of), but with the latter, no engagement whatsoever is required.
Tesla maliciously advertises as much much more than adaptive cruise control, lane assist, etc, which is disgusting on their part.
Some luxury car models have incredibly low driver fatality rates regardless of cruise control activation. Just the fatal stories I've seen reported on hackernews take tesla out of the top bracket for safety.
Does it concern you that it might provide a false sense of security?
Let’s say that people are not good at determining a black-box system’s reliability (as it is hard enough to do so with even statistics and hard data)