Federal safety official slams Tesla, regulators for misuse of its Autopilot tech
latimes.com
latimes.com
Im a safe driver, and autopilot makes me better at driving. I am not surprised that if you don't follow the instructions bad things can happen. (I believe statistically less than meat powered driving). That is on the human, not the technology.
The guy in the article was playing a phone game. He crashed into a barrier he had had problems with before.
Same guy, playing a phone game, whilst on cruise control in a BMW. Should we regulate that? No, we should regulate idiots in cars.
I couldn’t own a car without it anymore.
I definitely don’t feel Tesla is responsible for the crashes caused by people ignoring all the warnings and notices. People have been ignoring warnings and dying in stupid ways since warnings first existed.
When I read the manual it specifically called out several situations where ACC could cause a problem. Completely stationary objects and using it around a sharp corner can cause it to lose sight of the car in front of you.
As you said people have found stupid ways of dying and not paying attention while driving is one of them. I keep an eye on ACC and will keep my foot above the break if my intuition tells me to.
Super dangerous and stupid, but people do it.
You've got quite a bit of time before it actually does this. As long as you jiggle the steering wheel every ~20 seconds or so, it shouldn't actually stop the car or disable autopilot.
I definitely think Tesla works hard to keep things this way, and my comment was mostly neutral.
[0] Well think and say you do. Most people overestimate how much attention they're actually paying.
[1] And may never be with their camera system but that's another story entirely. I'm still dubious that cameras alone provide a reliable enough picture of the environment for CV systems to go 100% self driving.
Everything about this line is unsupported supposition.
[0] https://twitter.com/DakRandall/status/1170777292768985089
This sounds like a false sense of security. I find it extremely difficult to believe that a typical human can maintain their attention glued to the road in a purely passive manner for hours on end. Even if you were paying attention, without your hands constantly on the steering wheel you would have a slowed reaction time where milliseconds matter.
More importantly this kind of vigilance defeats the purpose of autopilot. I'm not necessarily saying you're doing anything dangerous since the autopilot function seems to be fairly safe, but your apologism comes across a bit like like an alcoholic claiming they drive better when drunk.
With manual driving you have to focus predominantly on one specific task - keeping your car between the lines. Bends in the road, poorly drawn lines, irregularities in the road surface, gusts of wind, all conspire to make this task require constant attention. Look in the rear view mirror for a couple seconds and then look back ahead and you might find that you've drifted, etc.
With AutoPilot the task has changed. You are no longer concerned with minute adjustments to the wheel, and you no longer have to maintain a constant high priority task of staying centered in the lines. Now, instead, you perform higher order functions of route planning, estimating traffic flow, observing other drivers and whether they are paying attention or drifting toward you, etc.
So by no means are you passive at all. It's actually extremely engaging, but significantly less monotonous form of driving. I find it quite pleasant, and not at all hard to stay engaged.
If I'm going to play a game on my phone, it's because I'm choosing to risk my life and the lives of the drivers/passengers around me, not because I couldn't handle looking out the window to pay attention.
i dunno, if you have trouble doing both the minute adjustments (which is usually a mostly subconscious task) and the higher-order functions while driving, maybe you need more practice/training.
there are 30-40k motor vehicle deaths per year in the US (and over a million worldwide). distracted driving is the main culprit (not speed, as most people assume).
we should expect no less than this kind of active driving from every driver.
Staying centered in the lane should be muscle memory for an experienced driver. The higher order functions you mentioned are all things you should be doing regardless, with or without autopilot. I can perform those tasks just fine while driving a stick shift, most drivers can.
The other side of the coin is that reduced engagement in the driving process can lead to lapses in concentration. That's not just my experience, it's the same opinion expressed by F1 drivers tasked with bringing the lead car home in an unchallenging grand prix.
As others have stated, it makes you quite a bit less irritable when the drive is over. That shit wears you out.
Next time you rent a car, get one with adaptive cruise control and give it a try. It is very handy!
Mentally, I find auditing the system while it drives way more fun and less exhausting than actually driving on an uneventful highway.
Apparently it takes more than that to be an "experienced driver", because lanekeeping and speed control leave me very fatigued after a 3 hour drive. I became accustomed to eating in at lunch diners and sitting down for dinner halfway home but other co-workers were not fatigued and could go through the drive through and eat in the truck.
It's definitely easier than it was, and I don't have to think about the mechanics much at all anymore when backing up a 5th wheel or controlling a truck sliding in the snow, but this human doesn't have a chunk of my brain that can do lane centering automatically. That is not a universal brain function, every human can drive but it takes a lot more energy and concentration for some of us. My stress levels have gotten lower and my life has gotten a lot better since I realized that and cut my commute down to a 2.5 mile trip with ~monthly travel instead...
The system is supposed to immediately start heading when it detects you're not responding. Keeping hands on wheel is not good enough, an active attention system similar to train ones should be employed and sharper at that.
Something as simple as a steering wheel button which is required to press in half second after lighting a signal or automatic emergency system is engaged.
Perhaps even in all cars, not just automated, though it's hard to implement safely without automatic steering and lane system.
Staying centered in a lane requires constant minor adjustments which are not entirely predictable. It’s perfectly easy and totally muscle memory when you keep a distant gaze, as they teach in driver school.
Because of this we are taught to take quick glances in your mirrors, and to check blind spots before changing lanes, etc. But you have to maintain focus ahead pretty much constantly. Everything else is secondary because it will cause drifting in your lane while you address it.
That same distant gaze—which is taught because it increases peripheral vision—can also, by the way, contribute to “drivers stare” or “blind staring” where your eyes are open and focused on the horizon but your brain isn’t really registering what’s happening around you.
I’ll also add that observing other drivers on the road, it is rare to see someone perfectly centered and maintaining that perfect center for mile after mile. They will drift left and right in the lane, sometimes enough to make you wonder what their plan is, or what else they may be doing in there. AutoPilot doesn’t just make driving easier for the driver, it makes driving easier for cars around you because the car moves more predictably.
My main point, speaking from experience, is that you actually can be totally engaged in the driving process even without your hand on the wheel. It may sound paradoxical to you just because it’s the only way you’ve ever driven.
Between the 1) constant touches that Tesla needs to feel on the wheel, the 2) guidance computer showing your immediate surroundings, upcoming traffic conditions, and route planning, and 3) the ability to spend more time on higher order driver functions... basically without trying to nerd out, it’s more like piloting a spacecraft than driving a car. It’s fun, and engaging, and IMO a lot safer. By no means is it boring.
Here’s what I will say — that someone who would drive recklessly by using a phone while driving, they may feel like they can “get away with it” better with AutoPilot, just because the car isn’t swerving in the lane when they take their eyes off the road. And you know what, most of the time they actually are safer making that reckless choice with AP than without it.
So the issue is actually incredibly nuanced. Cars today are not built to always prevent drivers from driving recklessly. We do this because driving is a form of freedom, and being subjugated to a machine which is supposed to “enforce perfect driving hygiene” is going to be a very frustrating and at times a buggy and unbearable process.
Instead, with or without ADAS, we rely on personal responsibility. This is a terribly flawed and imperfect solution, but until we actually have FSD and humans are totally out of the loop it is IMO the only option.
Once true FSD is live, we can regulate a date when manual driving can become illegal.
The amount of effort needed for this can vary wildly depending on the vehicle, it’s mechanical condition, and the speed it’s being driven at. Some vehicles are more prone to being bothered by wind, or grooves in the road, or require more constant adjustments to maintain lane position.
My tiredness is dependant on the vehicle.
I’ve had cars I’m exhausted from driving just after a 1 hour commute, and I’ve had a luxury sedan I could drive basically indefinitely (once did 3000 miles in 3 days, I didn’t mind it at all).
I wonder if the person you’re replying to previously drive a much less refined car, and would see most of the same benefits even when driving the Tesla manually.
Honestly the second part of their focus, simply being aware of their surroundings, should also be muscle memory for any experienced driver. That’s probably why they find driving so tiring, because they don’t plan ahead enough to keep a peaceful flow, rather than slightly too late scramble to gather information when they have to change lanes or something.
Yeah, I’ve driven shit boxes (see my comment about the Toyota Tercel stick shift).
But I still much prefer long drives in my Model 3 rather than my wife’s GL450. Before I got the Model 3 I drove an Infiniti G35 for 12 years, which was a perfectly “refined” car IMO.
> That’s probably why they find driving so tiring, because...
I guess you just have to try it. Again, I’m not saying driving is this exhausting thing that requires white-knuckled concentration. I’m saying that with AP it is just better in every way. You’re totally free to disagree if your personal experience with AP is different, but I think speculating on my driving ineptitude or poor choice of vehicle is kind of pointless / shooting the messenger.
I can say, without a doubt, I am less fatigued and more alert after even an hour drive in the Honda. I was skeptical before I had it, but I'm sold now. It's not that you "aren't driving" or paying attention, but not having to deal with the most basic aspects of driving makes a surprisingly difference.
I'm no longer exhausted after a long drive and it's amazing.
I've heard people claim to drive better drunk, and justify it with their perfect crash record. I remain fairly dubious.
And you've very clearly never tried autopilot.
"Where milliseconds matter"? You are wildly overestimating the abilities of literally any driver. No one reacts that fast, in any situation.
Which is precisely the problem. When autopilot goes wrong at high speed the driver needs to react that quickly and that's an unrealistic expectation to put on the driver.
This fundamental mis-branding is why various regulators across the globe are cracking down on Tesla marketing.
https://www.chicagotribune.com/business/ct-biz-lion-air-cras...
> Lemme described "a deadly game of tag" in which the plane pointed down, the pilots countered by manually aiming the nose higher, only for the sequence to repeat about five seconds later. That happened 26 times during the 11-minute flight...
https://nymag.com/intelligencer/2019/04/what-passengers-expe...
8:39:55 a.m.: In a clear sign that something is amiss, the autopilot turns itself off.
8:40:00 a.m.: The MCAS activates.
8:43:04 a.m.: For minutes now, the captain has been using brute physical force to pull the control yoke back in order to keep the plane’s nose from sinking.
8:43:20 a.m.: The demon awakened by the restoration of electric trim reappears. MCAS kicks back in, pushing nose steeply down.
8:43:45 a.m.: Ethiopian Airlines Flight 302 impacts a farm field at nearly 700 miles per hour
That's more than three minutes during which properly disabling MCAS would've been successful and saved the aircraft.
The nature of the MCAS issue is it kept firing every 5 seconds, making things progressively harder to counteract. It takes a while to get to the “can’t counteract” point.
https://www.quora.com/Why-did-the-pilots-on-the-Lion-Air-and...
To return back to the original point for a moment:
40 seconds is a shitload more time to react than the 1-2 seconds you've got when oncoming car traffic veers into your lane unexpectedly. Things in cars happen faster than in planes. A couple seconds of inattention in a plane won't kill you in the same way they will in a car.
autopilot in aviation requires the pilot to be attentive, and maintain situational awareness.
it’s not autonomous flying.
This is uncommon in planes.
TCAS, for example: https://aviation.stackexchange.com/questions/64847/how-much-...
> Traffic advisories are provided 35-48 seconds out and a Resolution will be provided 20-30 seconds out.
It's exceedingly rare you have 20-30 seconds warning of an impending car crash.
When autopilot gives up in a plane, the PIC will respond in seconds. But anything dangerous is typically minutes or at least tens of seconds away from happening at that moment, if at all.
When tesla 'autopilot' gives up in a car, the driver may have less than a second before a collision. Or worse, as we've seen with the 'tesla tries to drive into a barrier' cases, the 'autopilot' might not give up at all and just straight up try to murder you with the only safety remaining being your own attention to the situation.
The fundamental problem with <99.99% reliable autonomous driving is that the driver's attention is guaranteed to stray due to lack of stimulation, and their reaction time will be at ~3s whereas the impeding crash might be only 0.3s away. In that regard, it's much better to just be manually driving and have your attention on the road.
And the consumer flip side of this is, what the fuck is the point of a self driving car if it requires you to fully pay attention and be ready to intervene at any given moment on such short notice? Waymo observed that their test drivers get bored and tired really quickly when operating like that, due to the simultaneous lack of stimulation but need to be able to take over on short notice. The answer is that SAE level 2 autonomy shouldn't be labelled as self driving because it isn't.
Tesla 'autopilot' is an over-marketed lane-keeping cruise control system which has virtually no chance of ever becoming SAE level 3. Tesla marketing has mislead a significant percentage of owners that they're in fact buying a level 3 system, which is why various regulators across the world are cracking down on Tesla now.
Disclosure: I own Tesla stock.
That's a very strong claim. Can you provide any evidence?
I own neither Tesla stock or car.
Fewer things to hit in the sky but you can never be complacent.
Imagine playing an iPhone game while PIC and accidentally flying into the clouds? The CFIT stats are terrifying.
There is almost never a situation where a plane on autopilot can face catastrophe in <20 seconds.
Midair collisions are why we have ADS-B (even in my dad's little four-seater prop plane) and TCAS. The are also exceedingly uncommon.
People driving into a ditch, pedestrian, barrier, or oncoming traffic is commonplace.
IIHS did a survey of 2000 people and a driver assistance system named "Autopilot" received the most responses that overestimated the system's capabilities. For example, "Nearly half — 48% — thought it would be safe to take hands off the steering wheel when using the system." In general, about 50% more people overestimated the capabilities of Autopilot compared to the names used by industry leaders (ex. SuperCruise, ProPilot Assist). Other things users were more likely to think they could do with "AutoPilot" included texting and watching a video.
See https://www.iihs.org/news/detail/new-studies-highlight-drive...
Are you serious?? Literally everyone.
Definitions are what things mean. People incorrectly interpreting a word is their own fault and are in need of education.
Also, that's a terrible study to highlight. It is totally safe to take your hands off the steering wheel as long as you are ready to react.
Do you honestly think Tesla called it "auto pilot" with your pedantic, "technically correct" definition in mind? Or do you think they intentionally called it "auto pilot" to invoke a sense of "hey, this thing can drive itself!". I mean, in some of their marketing they even have a whimsical blurb about how "the driver is only present to make the lawyers happy".
Let's be honest here. Product naming is marketing designed to invoke certain emotions. Calling it "auto pilot" instead of "driver assist" or whatever most other manufacturers call it was very intentionally done.
Exactly. You can read one piece of copy about Summon, that says "Have your finger on this button. Maintain visual contact and pay attention with vehicle at all times", and on the same page. "Have your car come to you while you are distracted by a fussing child".
It's all very nudge nudge wink wink.
You misused "Literally".
Words mean what people think they mean. Definitions are what the author of the dictionary thinks people think they mean.
Unless you're going to spend the time and money and effort educating people on what a word might have once meant and what you would like it to mean again, you're kind of stuck with it. Tesla has absolutely chosen AutoPilot as their name because they know some people will think it has more capability than it does because some people don't know most plane's autopilot is fairly dumb.
People think autopilots do a lot more than they actually do, which is part of the problem with Tesla's naming. They probably should've called it "assisted cruise" or "driver assist" or something, but that's less sexy on the marketing side.
It doesn't help that their marketing pages for Autopilot have done a clever job of blurring the line between what's currently implemented and what's aspirational. https://www.tesla.com/autopilot has a giant header titled "Full Self-Driving Capability". You have to read more closely to see the big caveats.
The only reason it isn't automated is because there's no monetary gain in doing so, the people you are paying is still sitting around for the same amount of time.
https://dictionary.cambridge.org/dictionary/english/autopilo...
"a device that keeps aircraft, spacecraft, and ships moving in a particular direction without human involvement"
Either way, I think the wording causes people to trust it too much.
I mean absolutely no disrespect to you and I understand your feelings are probably coming straight out your heart. However I would like to point out that when you say that the world does not meet your expectation and that somehow is a problem with the world and not your own self is very arrogant.
Furthermore, in bumper to bumper traffic, you're better off staying stopped and letting the gap in front of you grow, only taking your foot off the brake and letting your engine idle move you forward at the average speed of traffic once the interval is large enough. Helps destroy traffic waves by keeping everyone moving forward slowly rather than having to constantly stop and go.
Try it some time. It takes some getting used to, but once you bust your first traffic clot, it starts to feel like a game where the objective is to keep everyone moving forward for the longest time without everyone having to stop. Completely counter-intuitive, but I'll be damned if it doesn't work.
Enjoy. The point isn't to modify the behavior of those in front of you, But to keep traffic behind you moving forward even if only slowly.
This keeps getting repeated and is complete misdirection. The barrier isn't intended to save lives from a headon crash, its to improve the slim odds of survival. Had the vehicle not accelerated toward it, chances would have still been grim.
Imo, noone should be operating a vehicle in this mode at high speeds.
I think thats a bit of an overstatement. Try it.
It's all well and good saying the driver should pay more attention - almost treating it like a normal tech product, I suppose, but most "tech" products don't control almost a Megajoule (mental maths might be wrong but you get the idea) of kinetic energy and the lives of the occupants of both the car and any others it hits.
I'm not saying that it's an impossibility to get it right, just that it needs to be held to even higher standards than the existing engineering present in the car (and planes if I'm being honest), not Sillicon Valley. Luckily this doesn't seem to be too much the case, but it could happen - the few people I know who own Tesla's are all non-car people ("What's suspension?"), i.e. The gadget-factor of the car is definitely attractive. Tesla do seem to be doing a pretty good job, so hat's off to them so far.
Almost every study on human behavior contradicts this. When you have no load, your monitoring and responsiveness gets poorer, not better. (Now, you can certainly go to the other extreme and get overloaded, with the same effect).
You may have a _perception_ that you're doing better, because you feel like you're glancing around a lot, and consider that equivalent to "monitoring", but most likely you're admiring / distracted by scenery, and you wouldn't have those immediate reaction times that (remember) you're meant to have, even with Autopilot engaged.
> Almost every study on human behavior contradicts this. When you have no load, your monitoring and responsiveness gets poorer, not better. (Now, you can certainly go to the other extreme and get overloaded, with the same effect)
With both a plane and a car you still have some load; you are supposed to be monitoring the performance of the systems.
I perceive myself to be safer when I have autopilot enabled on a Cirrus. My thinking goes from having to manipulate the controls to monitoring the performance of the aircraft.
I could be derelict in my duties at ignore the plane and play a game on my iPhone.
I perceive that experience to be very similar to when I engage Autopilot on my Tesla.
I go from having to manipulate the control to monitoring the performance of the car. With over 110k miles on my Tesla I really do feel with autopilot.
Autopilot is normally engaged when there's "nothing". Clear skies rarely have "sudden" obstacles. It's a no-no during severe weather or flying through mountains. I had a fascination with watching airline disaster investigations when I was making regular flights for a year. Around 70-80% of accidents were caused by relying on autopilot during sub-perfect conditions.
Using airplane autopilot as a case to use Tesla autopilot is a terrible, terrible analogy. Let's throw in an extra terrible.
My perception in sense of safety and workload reduction is about the same for aviation vs automative.
I’m a instrument rated pilot that mostly flys for fun on the weekend.
The big danger is folks disconnect from their duty to actively monitor the systems.
I would never play an iPhone game with either system.
A car in "autopilot" is more akin to the terminal environment, though. Things happen faster. Pedestrians run out in front of the car, oncoming traffic swerves to miss a pothole, and a second later you're smashing into them.
What you perceive may not be born out by the stats. It is very important to realize that humans are simply poor at self evaluation. Half the drivers think they are better than average, plenty of people think they do not suffer any adverse effects of consuming a glass of alcohol with respect to their situational awareness and reaction speed and so on.
Same goes for when I’m in the passenger seat of a car - I still monitor. I still look left/right when turning.
I don’t know if there is any additional education involved when buying a car with autopilot, but maybe that would be a good thing to learn about the risks and limitations using the equipment.
I don't doubt you, but do you know of any particular studies regarding this? I am kind of curious about how these studies are designed.
This is akin to the argument that we should replace steering wheel airbags with massive spikes in the center to promote safe driving. Could it work? Maybe, but seatbelts and airbags are more effective than regulating human behavior. In the same vein, there needs to be innovation in the self-driving safety space until we actually achieve Level 5. Autopilot operates in a weird chasm where the general public understands that it's not there yet, but expects it to be very soon. This leads to false assumptions about the capabilities of the system.
There is an interesting implication here, a point that I've tried to make a couple of times but this anecdote does a better job of it. Automation of self driving will lead to much more driving than we have today because today the requirement is that someone needs to be willing to make the trip as well as the vehicle. With automated driving that requirement drops away so there will be many more trips that do not happen now. This will lead to more congestion and will to some extent undo the 'green' effects of the electric revolution.
> With autopilot engaged, I am monitoring everything - probably with more accuracy then if I was driving, due to the decreased cognitive load.
But you've lost situational awareness and won't be able to recover it in time if the autopilot decides to call it a day due to the complexity of the situation. This has been studied countless times for airliner pilots and operators of rail vehicles.
Hopefully the NTSB recommendations on this include better and clearer explanations that full self-driving is anything but, and that those recommendations get implemented into serious laws with serious penalties.
I get that it is easy to outsell your rivals when you can market fairy tales. I just don't think it's okay when people start getting killed because of it.
I asked why wouldn't they just wait until it was actually available and the claimed that at that point tesla would raise the prices to hundreds of thousands per car because of all the income they earn.
It took me a little effort to not respond by backing away slowly.
Are these beliefs widespread and is tesla promoting them? Or was I just talking to an isolated weirdo?
Thankfully, this line of thinking is far from the majority.
I especially think the "your car becomes a taxi and pays for itself" gimmick is beyond fantasy. I chalk it up to Musk being an actually mildly insane CEO, which has benefits to go with its downsides.
As for the cost of the car, I wanted an EV with more range than other manufacturers provided. (I also make use of the longer range on a regular basis--probably about 2 times a month?) (edit:) I also wanted a car that Went Fast. Not practical in any real sense, but a lot of fun.
There are around a dozen Tesla owners at my company. As far as I know, none of them are banking on their cars being a taxi and earning tons of dough. At least one has the FSD upgrade, and uses it primarily to back his car out of the garage on its own so his kids don't bang up the doors when they get in.
Elon sees FSD happening between 2020 and 12020. There is no know reason that it can't happen. Well let's assume if we work real hard we can do it in 5 years instead of 5000. If it takes 15 or 50 or 100 years, no matter. That's basically the same, but the earlier we start the sooner we have it.
Of course, there are those who think there is some spiritual/religious component to true AI that we'll never be able to simulate. But Elon either disagrees or things that level of AI isn't required for FSD.
Is tesla accounting for this liability correctly?
All new Tesla cars have the hardware needed in the future for full self-driving in almost all circumstances.
...[blahblahblah FSD features]...
The future use of these features without supervision is dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving capabilities are introduced, your car will be continuously upgraded through over-the-air software updates.
https://www.tesla.com/autopilot
If FSD as described doesn't happen, I'm sure there will be lawsuits and refunds just like there was with AP1.
https://www.reuters.com/article/us-tesla-autopilot-lawsuit/t...
https://techcrunch.com/2018/05/25/tesla-settles-class-action...
Any early taxicab self-driving (e.g. no responsible operator present) is almost certainly going to be bristling with additional sensors for safety CYA.
Elon has been quoted, multiple times:
"FSD will be available, this year, 2016".
then:
"FSD will be available, this year, 2018".
and the latest:
"You will be able to FSD, coast to coast, completely hands-free, by the end of this year [2020]".
Turning it into him just "visioneering" is an entirely overly-complimentary viewpoint of what he has said and done.
And if it really did happen the way these people imagine, competition in the market against everyone else who has a Tesla would likely slash taxi fares down to within a few percent of your operating costs.
Well selling something called "full self driving" sure sounds like it promotes that view.
These are the same people who volunteer their time to help a multi-billion dollar company deliver cars. The same people who devote every waking hour to viciously attacking anyone who is merely neutral about Tesla or Elon Musk, instead of excitingly approving and encouraging of Tesla and Musk.
It's an absolute cult and I honestly think this is worst than the Glasshole scenario from a decade ago.
I think Elon is underestimating the complexity of actually shipping FSD and I don't think the Tesla robo-taxi idea is likely to work anytime soon, but in a world where FSD did work I think the robo-taxi idea makes sense as an extension of what's possible in that world.
I suspect part of the reason Elon remains optimistic about this is because pretty much everything he's done has been done while a choir of people around him and in the press go on about how it can't work and will fail. The automation at the factory, the gigafactory for batteries, EVs in general, the model 3 etc. If I was in his position I'd discount what all of these people think too.
When you say everything will fail all the time people just ignore you. Eventually you may end up right about something, but it's not because of any ability to predict outcomes.
I'm happy to pay the $3k though even if it's an interest free loan because Elon and Tesla pulled EVs from an uncertain future to the modern day while building up supercharging infrastructure. In addition to that the software/UI of driving the Tesla is leagues ahead of anyone else in the industry.
We need more of this kind of 'failure', not less.
I agree with your broader point. I just don't think that's a great example.
I can't help but feel like that $7000 is an interest-free loan to Tesla for a feature that's only promised to be delivered in the future.
Applying this line of thinking here: paying 50000 € for a Tesla might not be affordable to a lot of people ever. Paying a low amount to be driven by a Tesla taxi service? Absolutely possible.
Autonomous driving means no drivers to pay, also no driving times to observe or cars to transfer between drivers when shifts change. Provided there are enough customers, such an autonomous taxi car could drive all day long, except for charging and maintenance. Depending on the price of rides and the lifetime of such a heavily used car, it might indeed be a lot more profitable to include it into a pool of autonomous taxis instead of selling it to a customer at current prices. The prediction of this acquaintance might not be too far off if Tesla ever manages to make full self-driving work.
Like a lot of things Musk says it's unclear whether it was a lie, whether he believed it, and whether that particular idea lasted more than half an hour.
Here are the numbers: https://blog.kowalczyk.info/article/ac23f6cdd3b543b3b89d9f68...
With very conservative numbers I got yearly profit of $20k. Over 5-10 years it's $100-$200k. This is after paying off the car, insurance and operating costs.
So yeah, a car that potentially makes you $200k is worth at least $100k + $30k price of the car, given that it's money for almost no effort and you also get to use the car for yourself.
You're free to disagree with some of my assumptions but those are the numbers that other people are getting as well, when they bother to actually do an analysis as opposed to argument-less dismissals.
And I don't really feel the need to treat every brain fart of Musk's seriously. After all, he was supposed t deliver cars that could drive themselves from the factory to the customer in 2016.
Edit: Oops, 2017. My mistake; I sometimes lose track of CEO fantasies.
I first heard it on an unrelated youtube video from a popular videogame streamer who was talking about their new purchase. I’ve since heard it several other times.
I am guessing the salespeople at the test drives have learned saying it gets them sales and, since nothing is in writing, it won’t come back to bite them.
Wonder if they learnt that from their CEO, who said "this year" in 2014, 2016, 2018 and again about this year?
https://twitter.com/elonmusk/status/1151163201569972225
https://twitter.com/elonmusk/status/1117116982778679297
https://twitter.com/FredericLambert/status/11171180686547845...
https://twitter.com/elonmusk/status/1182826581070245888
https://twitter.com/elonmusk/status/1151180650033995776
Whether he's right depends almost entirely on how effective Tesla's heavily AI-focused approach to self driving turns out to be. If they can indeed reach the point where Tesla vehicles can operate fully autonomously with no one in the driver's seat, he may very well be right. It would be a long time before Tesla would be able to make enough cars to meet demand and a robotaxi service would be highly profitable in the mean time, so market rates may very well reach upwards of $200k per car.
That's all conditional on Tesla being able to develop and deploy a L4 driving system before everyone else though. So far, early results on that front have not been encouraging. I suspect Elon has underestimated the difficulty of training modern AI systems to reach the level of consistency and reliability necessary for safety critical real-world applications.
I don't know if you're about to back away from me slowly, being an isolated weirdo, but this is the company's stated ambition. Not with these specific numbers, but with the same broad strategic ideas.
It's probably a bit premature and in the region of fanboyism to buy into this vision with certainty today. Tesla certainly has its share of fanboys. But there's been many naysayers claiming impossibility of technological progress only to be proved hilariously wrong in hindsight.
Don't know whether you would accept a 1% or 50% probability of the above scenario happening in the timeframe I specified. But it's not zero, and the payoff would be huge if near-term robust self-driving turns out to hinge on sufficiently diverse, long-tail training data for computer vision neural networks. Tesla would basically replace Uber, sans diver payouts, with no near-term competition. No other company is near a similar position to gather real-world training data on a similar scale.
Tesla is currently betting their autonomous driving strategy on this. Obviously a big gamble, but not lunacy.
They very cleverly formulate it, and you fell for it, because their ambition has been adjusted to "being able to drive coast to coast with no hands on the wheel" and you took it as "create a robot taxi".
What it really means is that their car can do lane-following and adaptive cruise well enough to follow a few highways coast to coast. With a driver fully awake and responsible sitting in the seat.
That's miles and miles away from a robot-taxi and they know it.
I didn't take it anywhere - this is the company's stated ambition. Timeline - meh. Everyone knows Elon Musk overpromises on super-optimistic timelines to drive his teams harder. Especially on things that have never been achieved before. It's practically a textbook management practice at this point.
Maybe it's impossible and your derision will be warranted, we'll see. I love a good drama, especially when it comes to technology.
Unfortunately, if you are hoping for fixes to that, you are likely looking in the wrong place. The safety officials have instead focused on two problems. The first is that Autopilot can be activated anywhere, so long as it can see the lines. The NTSB seems to think that limiting it to some subset of streets will fix something[1]. The second is that Tesla's driver monitoring is insufficient and Tesla has not promised to do what some other manufacturer's have promised (but not necessarily implemented).
Notably, neither of these relate to the accident in question. The driver in this scenario was warned repeatedly that his attention was required. Regardless of the overall effectiveness of Tesla's solution, it is clear that Tesla's current solution was detecting inattention of this driver. Driver attention cameras would have behaved similarly.
Notably absent are clear and convincing evidence that one system is more or less safe than another, despite systems of various designs being in current deployment on our public streets[2].
[1] TBH, I partially feel like this complaint is also a result of the marketing that you are complaining about. Its almost as if the regulators themselves have been fooled into believing that there are places where less attention is require than others and this geofencing will solve something. In reality, attention is required in all places, even with this requirement, making this almost completely redundant.
[2] Please don't read this as a defense of autopilot. It isn't. It may very well be less safe than SuperCruise and other systems, I simply do not see the evidence proving either side right now. Considering the widespread deployments, I think a regulatory change in course should by its nature be evidence based rather than based on conjecture.
That's Tesla's spin on the events of the crash. Per the NTSB preliminary report, the last attention notification occurred some 18 minutes prior to the crash.
https://www.ntsb.gov/news/press-releases/Pages/nr20180607.as...
I own a Tesla Model 3, going in for the hardware 3.0 upgrade tomorrow, and I use the auto steer feature a lot more than I used to. The traffic aware cruise control portion is very good, as robust as any other maker. The automatic steering is LKA on steroids and does fine on open highway and back roads. It tends to caution, slowing for curves I would not even bat an eye at, and for cars turning off the road it does not even begin to accelerate until they well off the road. It is simply magical at night even in the rain.
It is not without its faults. I have a few spots on my drive I know it will mess up, mainly with horridly marked exits that will want to follow instead of maintaining its lane.
The worst part, except on interstates it is limited to five miles an hour over the marked limit. Which means on many parts of my commute I would either annoy a lot of people or get run down by some. Still I consider it as assistant that I meddle with.
I have never ever liked the name autopilot and really had hoped they would shutter the name until it was much closer to level 4. I was surprised no government has forced them to change the name, it implies far more ability than it has and worse too many people have this near magical connotation of what the term implies.
Yes it is dreadful some have died while driving this car or any other. The simple fact here is, regardless of features you are damn idiot if you are using your phone while driving let alone watching videos. I don't care what you think you were promised you know you are in the wrong so stop it.
Did you just schedule a service appointment?
People make the same decision as this driver does every day, and over 1,000 accidents a day result. AutoPilot unfortunately did not save this driver from their mistake in this case. But it wasn't a marketing problem or operator error.
This is incorrect (specifically, it's Tesla's mis-characterization of the situation). The last attention notification was over 15 minute prior to the crash.
https://www.ntsb.gov/news/press-releases/Pages/nr20180607.as...
That's really the issue. You have to pay attention to use Autopilot safely. If you're using it as a way to reduce cognitive load about speed/lane control so you can pay better attention to drivers around you it can improve safety. If you're using it as a way to avoid driving the car it decreases safety. The current design lets drivers get away with not driving far too much.
Autopilot has killed no one. They were driving a car and taking on all of the responsibility that that requires and were involved in unfortunate accidents that happen literally every second of every day. Whether they were distracted by a bird that flew by or by their false assumption they could just stop driving the car entirely doesn't really make a difference.
Musk's decision not to use LIDAR is paid for in blood.
[1] https://www.ntsb.gov/news/events/Pages/2020-HWY18FH011-BMG.a...
LIDAR is not the only way to record depth information. Saying that LIDAR is some kind of savior device is heavily into "solve all your fusion problems with this one weird trick" territory.
When you mix two or more different ways of measuring depth you can increase the reliability of your system dramatically. See "sensor fusion". Optical-only has weaknesses. LIDAR-only has weaknesses. Optical plus LIDAR would be far more robust than either alone.
if lidar and radar don't detect an object, but stereo vision do, what do? you use consensus, or you go with at least one sensor.
there's a lot of tradeoffs involved, including stacking false positives.
LiDAR can distinguish between on and off-road objects, so it doesn't have to ignore stationary objects. I don't know of a LiDAR-based highway assist that's commercially available, though.
Finally, because we're a bunch of pedants here, radar-based systems will register stationary objects that were formerly moving. This is how "follow to stop" systems know when to stop. :)
> The NTSB also called out Huang’s employer, Apple, for failing to set a strict policy for its employees banning non-emergency use of mobile devices while driving
Yeah the real problem here is insufficient employer paternalism during commutes.
[1] https://www.cnbc.com/2020/02/25/ntsb-calls-out-tesla-and-app...
Never underestimate the ingenuity of idiots. No matter how carefully you design your safety protocols, some genius-level moron will find a way of getting themselves killed.
I think it's quite reasonable, from a point of view of preventing false advertising to stop this. Removing all references to "Autopilot" and similar and replacing them with sensible descriptions like "Driver Steering Assist" like the other car manufacturers are using would definitely satisfy me.
While this may be irrelevant to the current Tesla case, it's a big problem in general with all office workers. I know several people who will "join the meeting from their car" and set up a huge threat to their own safety, and that of people sharing the road with them.
If anyone reading this does it as well, or puts people working under them in a position where they feel they have to do it – please stop! The NTSB says so as well.
It's like comparing your primary care doc's outcomes against a world-class brain surgeon. More people come out of the PCP's office alive, but that doesn't necessarily mean they're the better doctor.
“we registered one accident for every 3.07 million miles driven in which drivers had Autopilot engaged.”
And then: “By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 479,000 miles”
This is true. It’s also a complete lie.
It’s set up to make you think that this proves Autopilot is safer than human drivers, but it compares a national stat for all cars in all possible driving conditions to a Tesla stat for high end cars in good weather freeway driving. The latter are inherently much safer.
But so many people are falling for this blatant lying - plenty of examples in this thread alone.
I hate that this works for them, and that it just encourages others to lie similarly. I wish we had higher standards, as a society.
Does the car disengaging autopilot and smashing into something a quarter second later count as an "autopilot engaged" crash?
I don’t mean to give Tesla a pass mor marketing autopilot, but that section of road is hard for humans too.
That's probably why ML gets used so much, you can't really prove it's incorrect but it works a lot of the time so you can get away with selling it until someone dies.
Maybe the repairing crash mitigation devices of frequent & deadly hazards should be a higher priority.
Or maybe better marking the 'gore' lane as in its current state it's simply an empty lane that would look identical to a real lane (i.e if you can't see the divider then it looks like a normal lane).
I don't get why we aren't looking into slightly changing our roads to better work with self driving cars when all it takes is a little more paint.
Because 99.9999% of drivers are human and have no issues seeing the concrete barriers and not hitting them head on at full speed. Until self driving is better than humans in all cases it will not be acceptable.
Nobody will spend money to adapt roads to self driving cars until they comprise a significant percent of cars (read: decades from now, if ever)
In this particular case, a human driver did crash head-on into the barrier somewhat recently.
It also saves more lives than it kills people. So?
Do you have data that the rest of us don't - that is data which identifies incursions (that is near-crash interactions which do not end in a crash) where vehicular automation acts correctly and human drivers do not?
If not, than you really can't make that statement.
You can say that "Tesla's luxury cars running under autopilot show fewer crashes per mile than average for all cars", and be accurate, but that's not "saving lives".
I mean, seriously: if this technology "kills!" we really should have some statistics on that by now, right?
With cars you may need a response in seconds or less in case of a disconnect. Humans suck at that.
Some sources: https://www.google.com/search?q=autopilot+disconnect+airplan...
There's also a lot of interesting research from Missy Cummings' Humans and Autonomy Lab at Duke. https://hal.pratt.duke.edu/
She's been very critical of a lot of self-driving systems that mostly work but still require drivers to pay attention. (She's also a former US Navy fighter pilot.)
Of course, when you consider that an overhead road sign and a stopped semi trailer "look" the same to the systems involved, it's not surprising that they think that everything's OK.
I expect autopilot cars could follow similar regulations as aircraft: "if you want to keep your license, you can prove you've driven in manual mode for x hours every month", where x is enough to know how to control the car in an autopilot disengage.
[0] https://99percentinvisible.org/episode/children-of-the-magen...
The "slamming" of Apple seems particularly lame. Partially it's because they make iPhones, but the head of the NTSB also seems to be implying that an employer has a responsibility to prevent its employees from engaging in risky behaviors when they're off the clock.
_IF_ (and I absolutely don't know in this case, but we've all certainly heard of it before) an employer expects that you will be responsive just because you have a phone with email / Slack access, etc., let alone if they factor that into performance, whether it be informal or formal, then perhaps they do (though certainly a minority, and, again, I have no idea about this particular instance and employer).
Guess we are really going to need eye tracking in any proper system that does this much 'autopilot' for drivers.
> Do Not Use the AutopilotBuddy on public streets. The Autopilot Buddy is not for street use, it is for closed track use only. The AutopilotBuddy Nag Reduction Device is no longer available to USA Residents.
Look, ma! I'm driving on a track with no hands! What a joke.
(Yes, obviously you wouldn't be going anywhere near full speed.)
Alerts mean nothing if you ignore them
>Before enabling Autopilot, you must agree to “keep your hands on the steering wheel at all times” and to always “maintain control and responsibility for your car.” Once engaged, if insufficient torque is applied, Autopilot will also deliver an escalating series of visual and audio warnings, reminding you to place your hands on the wheel if insufficient torque is applied. If you repeatedly ignore these warnings, you will be locked out from using Autopilot during that trip.
Because while they were able to get away with it, and before the lawyers tapped Elon on the shoulder, those warnings only happened and only required your hands on the steering wheel _four times an hour_.
There's TV interviews with Elon doing exactly that.
The last notification for the cited wreck was 15 minutes prior to the crash.
https://www.ntsb.gov/news/press-releases/Pages/nr20180607.as...
I can't say I use it all that much, or like it all that much, but I do so occasionally. Just got home from a roadtrip last night. Frankly, because I'm watching where I'm going anyway, I feel the traffic aware cruise control (available on lots of other cars as well) is the best part. If you're looking down the road the steering input from your body is basically automatic anyway. And it's smoother than autopilot.
(if you're not looking where you're going, hopefully you're not in the driver's seat)
Driving is both the source of trillions of dollars of economic value annually, as well as the cause of ~$1 trillion a year in economic damages (including pain, suffering, and loss of life).
A focused and alert human is, overall, fairly awesome at driving, even in surprisingly adverse conditions. Humans, however, can also be lazy, easily distracted, and make irresponsible choices.
Cars to some extent try to protect humans against making these bad choices (like seat belt warnings) but to other extents respect that humans are not to be subjugated by a machine, and this means even being allowed to make bad decisions.
ADAS systems like AutoPilot and Level 3-5 self-driving are the most promising solution to the daily devastation that occurs on our roadways. This is literally a multi-trillion dollar problem. For an ADAS system to be most effective, it must be reliable and predictable, and it also must be easy to use. An ADAS system cannot increase driver safety if the driver turns it off, or won't turn it on because it's too annoying.
As a Tesla owner I can definitively state, that the safest way for me to drive is with AutoPilot enabled, with my hands on the wheel, and looking out the damn windows. This is how cars are supposed to be driven, and it's how Tesla consistently tells its drivers to drive the car. While Tesla consistently and repeatedly instructs drivers to keep their hands on the wheel and to remain alert, it doesn't stop humans from taking their focus off the road, just like humans do consistently with or without AutoPilot.
My point is basically this; it does not necessarily make driving more safe to make AutoPilot impossible to abuse. The way most drivers that I see on the road are operating their vehicle, it should be illegal to drive without AutoPilot.
In short, I think it's important to understand that a technology can literally save lives while still not saving every life. It's a tragedy every time someone dies on the roads, and it will be a miracle when we have technology fully deployed that can prevent 99% of those deaths.
[1] - https://www.cdc.gov/motorvehiclesafety/distracted_driving/in...
The name seems to imply something the tech does something that it's not, though in all fairness to Tesla, they are very clear about the fact that Autopilot does require an attentive human at the wheel.
Tesla does not want to fix this issue- because they have to know that a large number of their buyers are wanting the car specifically to abuse the assertive technology.
When a rental Ford Fusion yells at me using Adaptive Cruise without sensing my interaction with the wheel you have got to be kidding me that Tesla cannot take the lead on this technology.
Although that is difficult, as in many cases the "cooperative" manufacturers have not yet implemented such features. They have only promised to implement them, sometimes in vehicles years away from production.
One very common scenario (that I have no idea how it would manage to handle) is someone driving with sunglasses on.