Dear Tesla, stop calling it autopilot – and drivers are not your guinea pigs Tesla is misleading drivers about the efficacy of its Autopilot feature and is putting lives at risk, according to Consumer Reports. The automaker's autopilot system, when engaged, is supposed to control the speed of the vehicle and its distance from other rides and objects – but it's more of a super-cruise-control than a …
What are you, under forty? Get off of my lawn, pronto.
* That's what we used to call it for a while. Because you buy it in an unripe state and have to wait for it to mature a bit before it's of any use for you. You know what? It's friday, let's go to the pub and have a few. I'll get my coat.
@Mark 85 - Others are showing the correct way to develop autonomous automobiles; slowly and carefully in specially designed test vehicles. These vehicles are operated under controlled situations even when on public streets by trained staff. They are not operated by untrained owners under any and all conditions which is what Tesla is doing.
The technology is several years away from being deployable safely to the masses. Driver assist systems that handle emergency braking, etc. are deployable because the driver is still actually driving the vehicle.
"The technology is several years away from being deployable safely to the masses. Driver assist systems that handle emergency braking, etc. are deployable because the driver is still actually driving the vehicle."
Clearly having a collision rate lower than that of humans is too dangerous - so we should actually ban all human drivers...
Of course not *you*, you're one of the 90% of drivers who consider themselves to be a 'good' driver.
The technology is very good at what it does - and it's capabilities are improving all the time, unlike human drivers who are generally careless and whose abilities/habits tend to degrade over time (after those first couple of years).
Given a choice - I'd have an autopilot enabled car now, and use it as well. I would be significantly safer as a result of doing so.
As someone who once had a little more money than was good for him (post divorce so no restraining influence) I bought my Tesla about 12 months ago, it's my primary vehicle and has clocked up about 28k since. The 'auto pilot' is excellent as a driving aid - enhanced cruise control. The manual makes this very clear - but I suspect far too many idiots don't RTFM.
As for the car, lovely to drive and a very comfortable large car for long distances. A little planning is needed occasionally, but frequently doing a Scotland-Midlands-Bristol trip, no big deal. Stupidly expensive, but some mugs have to be earlyish adopters or things don't progress.
> Given a choice - I'd have an autopilot enabled car now, and use it as well. I would be significantly safer as a result of doing so.
Are you applying for a Darwin Award?
"> Given a choice - I'd have an autopilot enabled car now, and use it as well. I would be significantly safer as a result of doing so.
Are you applying for a Darwin Award?"
No - I'm not.
I've looked at the state of play, I know people who work at Tesla, I know people who have them.
I have read the user guide.
The addition of autopilot is a net safety enhancement. It is not a license to kip, nor to watch a film/read a book.
Clearly having a collision rate lower than that of humans is too dangerous - so we should actually ban all human drivers...
Are you referring to the autonomous vehicles that get to pick and choose the driving conditions under which they are tested? It's pretty easy when the typical route doesn't change much and the road conditions are almost always near perfect. While humans do worse in inclement weather the autonomous vehicles haven't really been tested in rain, snow, or other storms now have they? Hell, how many miles have they logged in the dark?
If you're referring to Tesla, their autopilot has logged only about 130 million miles and they've already had their first fatality. That gives them a rate of 0.77 per 100 million miles which just about ties them with the drivers in the state of Maryland who logged 56,432 million miles in 2014 alone.
"their autopilot has logged only about 130 million miles and they've already had their first fatality"
In a condition where a truck driver turned across oncoming traffic and it's not even known if there was sufficient time for a human to react and avoid it if he was 100% in control.
Also, where the evidence seems to have been tampered with (the driver was known to use a dashcam at all times, yet the dashcam is missing) and the truck driver claims to have _heard_ a portable DVD player over the noise of his own engine from 20 feet away - which is pretty suspect to say the least.
Not to mention your cherrypicking of statistics. You can't compare long-established rates with 1 incident in this milage. GIve it another 3-6 sampling periods and you might be right, but right now there's insufficient data to draw any conclusion.
I'm not picking any cherries. I'm merely looking in the basket of cherries that has been pre-picked by the many already declaring that the technology is safe. My whole point was to show that there has been nowhere near adequate testing so saying it's safe is a bit premature considering how many miles are driven by actual people. Consider that if there is another Tesla fatality on autopilot in the next week it will change the statistic from being one of the better numbers to being tied with Mississippi which is one of the worst. While everyone may think 130 million miles is a lot, it's a mere drop in the bucket compared to annual road miles.
"if there is another Tesla fatality on autopilot in the next week it will change the statistic from being one of the better numbers to being tied with Mississippi which is one of the worst."
Exactly. It's the statistics of small lumpy numbers. They can be used, or abused, even more than most other statistics can.
"Of course not *you*, you're one of the 90% of drivers who consider themselves to be a 'good' driver.
As someone who's driven over a million miles without an accident, I do consider myself to be quite a decent driver. Although some of the sights I've seen in me rear view mirror have been quite harrowing.
"Yeah... MS has been doing this for years but the computer isn't driving down the road."
And that is the crucial point. How come the software that can control the steering isn't a fully tested and approved bit of code? How come it can be updated without being re-submitted for new approval? I can't seen that happening in the aviation or pharmaceuticals industries where even the slightest update to a product needs official testing and approval, possibly lasting for years.
Is it just that so many people die on US roads that an extra few don't natter? DaaS, Death as a Service, ooooohh it's internetty and cloudy so it must be good.
Nikola Tesla was a genius, he basically invented everything to do with Alternating Current.
But he was also completely bonkers. Mad as a hatter. Fell in love with a pigeon. He is quoted saying things that were just wrong. 'Power the world with six towers...', no. 'Knock down a building with a small clockwork mechanism...', no.
Somehow he lived in a hotel, but died penniless. I assume that the hotelier was extremely generous.
But yes, a great inventor. A genius. A half-mad genius.
This post has been deleted by its author
"But he was also completely bonkers. Mad as a hatter. Fell in love with a pigeon. He is quoted saying things that were just wrong. 'Power the world with six towers...', no. 'Knock down a building with a small clockwork mechanism...', no.
Somehow he lived in a hotel, but died penniless. I assume that the hotelier was extremely generous.
But yes, a great inventor. A genius. A half-mad genius."
You make it sound like his achievements are worth less than those of others because he lived his life differently to what you would consider normal.
Einstien treated his wife like shite, but everyone kisses his arse.
"But he was also completely bonkers. Mad as a hatter."
“No great mind has ever existed without a touch of madness.” -- Aristotle
"Power the world with six towers..., no."
Perfectly possible, but would be hugely wasteful of energy. Maybe one day when we have abundant energy, but I suspect even then we would like efficiency, except for a few things where convenience is more important. For now we use the technology to transfer power between sealed sections of UK submarines, and a previous gen of the technology is used in those "wireless charging" mats.
"'Knock down a building with a small clockwork mechanism...', no."
The theory of resonance is quite well understood now, and a powerful enough mechanism could knock down a building if the correct resonant frequency was found.
Building a small compact mechanism that can do it is tricky, but I don't think there has been much study into it. Primarily because if you want to knock down a building, you might as well skip the intricate clockwork mechanism, and brute force it with a seismic bomb.
"The theory of resonance is quite well understood now, and a powerful enough mechanism could knock down a building if the correct resonant frequency was found."
Rubbish. You're clearly not a physicist.
Let's assume for argument's sake that we find a building with a nicely tuned resonant frequency with near-negligible losses.
The device doing the exciting has to provide the energy into the resonant system. Energy doesn't come from nowhere.
If the exciter doesn't provide enough energy (or, if you will, power) to overcome losses in the resonant system, there will be no visible resonance. Just damped oscillation, maybe.
A small clockwork motor will not provide enough power to overcome losses in a real building.
Next.
First of all, as already stated by Tesla, a car running in Autopilot is far less likely than a car not driven in Autopilot to be involved in a fatal collision.
The fact is that when used properly this technology saves lives, but like many other safety technologies when used incorrectly can cause serious injury or death. i.e. an airbag is more likely to cause injury to a passenger who is not wearing a seat belt involved in a low speed collision than had it not been there. Do we then take that one isolated fact and extrapolate it out to conclude that airbags are bad and should be banned? No we don't, we expect people to adhere to the road rules and not behave in an unsafe manner so that the majority of people can benefit form the net safety benefits of airbags.
Another factor with any safety technology is that people risk compensate, for example this excerpt from the British Medical Journal:
"Compulsion to wear a seatbelt cut deaths among drivers and front seat passengers by 25% in 1983. But in the subsequent years, the long established trend of declining deaths in car accidents reversed, and by 1989 death rates among car drivers were higher than they had been in 1983. Evidently the driving population risk compensated away the substantial benefits of seatbelts by taking extra risks, putting others in more danger. This period saw a jump in deaths of cyclists "
So we need to decide will we all become luddites in the face of advanced safety technology or will we embrace it despite the fact that humans will initially try and find a level of risk that they are comfortable with, usually overshooting the mark and then coming back to some soft of acceptable level.
This issue is not, "is it a good idea," but, "is Tesla misleading drivers with regard to its abilities."
My snake-oil detector goes off whenever I see "intelligence" applied to to IT stuff. "Autopilot" might work well on an aeroplane in an obstacle-clear sky with radar, objects in relatively predictable trajectories, coordinated air-traffic control and two pilots at the ready, but in a cluttered ground environment even speed-maintenance alone is dodgy around town.
First of all the name Autopilot, obviously borrowed from aviation, does not imply that you can have a snooze, hop in the back seat or read the paper. Do you see the pilot and co-pilot just wandering around the plane whilst letting the autopilot on a plane do its thing?
Let's say Tesla caves in and calls it Driver Assist. Do you think that the man watching the DVD would have done anything different? That the lady who accelerated her husbands Tesla into a brick wall wouldn't have? That the man who crashed into a field because he refused to place his hands on the wheel despite the car asking him to do so repeatedly would have put his hands on the wheel?
Look I'm no aviation expert or behavioural expert but I think common sense dictates that the answer to the above questions is no, no and no.
"First of all the name Autopilot, obviously borrowed from aviation, does not imply that you can have a snooze, hop in the back seat or read the paper. Do you see the pilot and co-pilot just wandering around the plane whilst letting the autopilot on a plane do its thing?"
If you like air travel, I might have bad news for you...
I refer you to the FAA, Advanced Avionics Handbook - Chapter 4:
"An autopilot can be capable of many very time intensive tasks, helping the pilot focus on the overall status of the aircraft and flight. Good use of an autopilot helps automate the process of guiding and controlling the aircraft."
I don't read anywhere that a pilot can just go wander around and leave it to George. I see emphasis that the Autopilot is there to help the pilot "focus on the overall status of the aircraft and flight"
So although you may find some case where an asshat pilot may have done so illegally, this does not detract from the overall case that the name Autopilot does not imply that you can divert your attention from the driving task.
There are generally TWO people qualified to fly the metal tube at the front of the plane. One of them can go for a leak/wander around the plane leaving the other one in control
I have been on a scheduled flight where there was only one pilot. Granted, this was in a twin engined Cessna flying from Fort Myers to MIA. This was operated by Air Florida.
As someone who had a job a long time ago designing Autopilots, it rankles me when Tesla calls their thing an Autopilot when it isn't. Driver Assist is a far more accurate description.
First of all the name Autopilot, obviously borrowed from aviation, does not imply that you can have a snooze, hop in the back seat or read the paper. Do you see the pilot and co-pilot just wandering around the plane whilst letting the autopilot on a plane do its thing?
They are fully trained as to what an autopilot actually does. They will get fired if they do. Not the same thing as being alone in your own car.
Let's say Tesla caves in and calls it Driver Assist. Do you think that the man watching the DVD would have done anything different? That the lady who accelerated her husbands Tesla into a brick wall wouldn't have? That the man who crashed into a field because he refused to place his hands on the wheel despite the car asking him to do so repeatedly would have put his hands on the wheel?
Look I'm no aviation expert or behavioural expert but I think common sense dictates that the answer to the above questions is no, no and no.
Indeed, but common sense has to compete against a childhood of watching Knight Rider. Autopilot implies more hands off. Nobody is going to read x pages of EULA to check what the Autopilot can or can't do, they're going to go "yeah, it's got Autopilot, sweet". This is also how Apple's phones are sold, by marketing to people who have no idea what it really means, but they know it's cool. Have you read iOS's EULA? Has anyone? Thought not.
The other difference between Teslas and the rest is you can 'drive' with your hands off the wheel for longer.
It all gives a sensation of autonomy that the rest don't have, but it doesn't mean that Teslas actually have it either. Tesla can't handwave away problems and say it's "Beta" because that's Silly Valley culture, they're in meatspace now. You'd never get a Rainbow Road screen on a Volvo to distract the driver with.
By the way, Consumer Reports copied me.
You're right. In most people imagination, "autopilot" means some kind of sci-fi AI fully able to control a vehicle flawlessly. Tesla cars don't have an R2D2 in the trunk taking control.
While in aviation pilots are trained in understanding their hardware capabilities and limitations. There are also limitation on when and how an autopilot can be used and when not - some rules made after crashes related to too much confidence in autopilots (i.e. the Turkish Airline 737 crash at Schipol - the pilots then took too much to react too, the root cause was a faulty sensor).
Moreover planes fly in a highly controlled environment with far less obstacles (and less morons). I wonder if car makers didn't study what happened in aviation. Anyway, maybe car too need a special license for "instrumental driving rules"....
'A.I. is hard.'
Might be an understatement.
In fact, it might be impossible (except in the sense of neural nets). Coder drones typing in zillions of lines of code? Forget it, not a chance.
A.I. is as allergic to the infinite complexities of the real world as those alien microbes in 'The Andromeda Strain' were to rain.
"Do we want to advance or not?" is apologistic rubbish.
Remember: The stupid thing drove into the side of a truck.
The NHTSA or DOT should order Tesla to turn it off.
Not only does it need to be several times safer than humans, it must be certified not to make stupid mistakes.
They're very far from being done.
Of COURSE cars in autopilot are less likely to be involved in collisions. Because it is basically a super cruise control! The per mile accident rate for cruise control (the old school kind that just sets a speed and doesn't keep following distances or brake) is far lower the regular accident rate for the same reasons - not because using cruise control makes you safer, but because you only use it during situations where you are already safer.
Man, I'm really losing a lot of respect for Musk over this. That guy is willing to twist statistics and say anything instead of admitting to the real problems with autopilot - chief among them the name and the fact that it doesn't enforce any attention from the driver. Hell, it can detect when you take your hands off the steering wheel but DOESN'T DISENGAGE. I hope the NTSB investigation puts some heavy penalties on Tesla, they deserve it.
Musk has probably singlehandedly set back autonomous driving by five years, because the regulations his ineptitude is going to cause will make it tougher for everyone. Even those who are responsible and not trying to push the envelope by using humans as guinea pigs for beta software in the name of publicity.
Your argument makes absolutely no sense.
If you have seen any video on how users are using AutoPilot, you would see that it is being used a lot in circumstances Tesla does not condemn. Maybe get informed before blurting out your venom.
I find it amazing how little accidents have happened *despite* how users use it, far away from the easy, safe highway with simple lanes, etc...
That said, Tesla should invest even more time and energy in informing the users on the limits of the system since there are plenty of daredevils who want to find the outer edge of its limitations...
"Hell, it can detect when you take your hands off the steering wheel but DOESN'T DISENGAGE."
So, just to be clear here, you'd prefer Tesla to design a system that stops trying to drive your car at the point that you remove your hands from the wheel (rendering the vehicle technically out of control at that point). Surely that's the exact point in time at which you want something to be in control of the vehicle?
From what I can tell, Tesla have a series of warnings and alerts that go off if it detects no user input. Actually disengaging the system would be one of the stupidest actions they could possibly take.
Tesla safety risk: a soundbite answer to a question involving dozens of complex variables. It completey evades the driver's risk of adopting a more relaxed attitude to situational awareness because "the autopilot has it covered".
The comment regarding accident stats connecting seat belt usage with increased accident rates is also overly simplistic and a clear example of correlation rather than causation. I could, for example, point out that in the same time period cars began shipping with all-round disc brakes as standard, resulting in drivers using them more agressively in traffic, and turbocharging became ecconomically viable and mechanically reliable, resulting in drivers using accelleration more agressively. I know this because I Was There.
And there is a difference between a Luddite and someone who would rather their kid, raised carefully and dutifully to all the proper standards espoused by SINKS and DINKS posting on El Reg forums these days, not be killed as collateral damage when some fucktard with more money than sense doesn't RTFM or does and fails to recognize the gap between designer hyperbole and engineering reality.
The Tesla is being developed in this case as one would develop software in a post-www world when it should be being developed more traditionally with the knowledge that even fancy expensive cars can kill people, just like the conventional auto industry does. A bad driver for a flatbed scanner will ruin your day. A bad driver for the Tesla "autopilot" will ruin someone else's life.
Also: naming the feature "autopilot" invites stupid people to act stupidly. We live in a world where I dig about a car a year out of my front lawn because young people equate cruise control and automatic gearbox with "I can fiddle with center console gui stereo controls while exceeding the speed limit in perfect safety".
17 seconds? Do they insist on finishing the chapter, or what?
It's the reality that the likes of Tesla are choosing not to acknowledge. A driver who's been lulled into a false sense of security because "the car's driving itself" not only isn't holding the steering wheel, and doesn't have feet anywhere near the brake, but will likely have let their attention drift and have no situational awareness. When the bing-bongs go off first the driver will just be startled, then they'll have to get their hands and feet into the controls, scan their instrument panel to work out what the hell caused that, and then start looking around outside their vehicle to start the process of working out what their vehicle's doing and what's happening around them. Only then can they decide what action to take and effectively take control (assuming they didn't do something daft when first startled). It doesn't at all surprise me that this might take 17 seconds.
And that's exactly the wrong sequence. You should *first* look around to ensure you don't crash into anything or anyone, ensure you have control, and only *then* you can start to assess what's wrong.
That's for example, what pilots are trained to do, while drivers are not.
You're in charge of 3 tons or so of metal; travelling at motorway speeds; being controlled by beta software. You should bloody well know what's going on around you. 17 seconds response time sounds like criminal negligence to me...and as an ex lorry driver, I am aware that you zone out a bit on long motorway stints; but even so.
It also sounds like the alarm should be a little more strident...a discreet dinging bell is fine to call you to yoga lessons; less useful to alert you to a situation that may well involve your fiery death.
Good point...I wasn't saying there should be klaxons though, just something a little more strident than soothing bong noises, which clearly aren't doing the job. Can't use voice because that takes too long. Maybe a buzzing sound with a hint of urgency. Buzz/hiss/white noise combo maybe...I'm no sound engineer. Rapid high-pitched beeps?
If it's too alarming you could maybe give people heart attacks and it would definitely be distracting and also people would disable it if it really annoyed them.
It should be low-key; but it should also command attention. More "You need to be paying attention right now matey" and less "There is something you may care to attend to when you've finished your latte".
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
