Tesla reports two more fatal Autopilot accidents to the NHTSA Tesla's automated driver assistance system (ADAS) is coming under fire again, with the National Highway Traffic Safety Administration reporting two new fatal Tesla accidents in its monthly ADAS crash report. Since mandatory reporting began in June 2021, the NHTSA has recorded 18 fatal accidents it said involved ADAS systems. …
You really can't say people will die without Tesla's ADAS but, you can say people will die with it. Can you use a "glitch" for accidental excuse, especially with the insurance/lawyers? Also, how can you tell if the ADAS failure was "user error" or a glitch, is that reported?
We do know that people will die without ADAS. People die in non-ADAS accidents by the thousands each day, I don't think anyone doubts that. Statistically, people die both with or without ADAS. Individually, people might die with or without ADAS.
The big question is whether the chances for an accident are better with or without ADAS. AFAIK, there's still not a truly reliable source with an answer to that.
Autopilot (Tesla's ADAS) = look, ma, no hands!
Other manufacturer's ADAS = lane departure/emergency braking/cruise control/collision avoidance. No freaking 'autopilot' anywhere.
Should Tesla drop the BS marketing 'autopilot' term you can bet your ... whatever... Tesla's beta testers... sorry, customers, would keep at least one hand on the wheel and one eye on the road.
And, with that statement, comes to solution: Telsa should put touch sensors on the steering wheel, thereby requiring the driver maintain their hands on the wheel whenever "Autopilot" is active. Remove your hands, Autopilot automatically disengages.
That simple.
But they probably won't do it, until and unless backed into their last, ultimate corner. Because people want to believe the "Autopilot" magic and Telsa is happy to feed these [stupid people's] delusions, because it sells more cars. The legalese of the fine print of what it actually does, "driver assistance", is lost inside the Owner's Manual, the ToS, and the brochure fine print. But as long as it exists, somewhere, Telsa can hide behind said legalese, sell the lie, and make money.
> Telsa should put touch sensors on the steering wheel...
It has to be much easier to monitor the driver than all the other vehicles in the vicinity and the other things an autopilot must do to drive safely so surely the car should be monitoring the driver and if they aren't paying attention it should start by sounding an alarm. Failure to respond should be treated as a deadman's handle type event.
I believe they do have sensors in the steering wheel. Drivers are given several warnings to put their hands back on the wheel which if ignored require the driver to come to a stop, turn the car off and on again before the 'auto'pilot can be started again.
At least, that's what I've heard. Apologies if i'm getting it mixed up with another manufacturer.
Err, haven't you missed several clauses from that IF statement ?
How about other road users - pedestrians, animals, etc.
And add in things like roadworks, accidents, incidents (such as fires) which don't involve a vehicle or road users but do involve (e.g.) a fire engine parked in the road with bright flashing lights on it (which apparently Teslas can't see).
And there's a problem with requiring all road vehicles to be fully automated with no over-ride. One class of vehicle I drive fairly regularly is an agricultural vehicle, and I can tell you there's no way that's going to be fully automated - even just for on-road use. So there I am, driving along in my "automated but can't be over-ridden mode" on a tractor - and I want to turn into a field. It's not going to let me is it ? And if you work around that one, I'll have to manually steer it and decide the speed as only I know exactly what the load is (the system could know the weight, but not necessarily the characteristics such as CoG or dynamics (ever carried a large liquid load !)) and things like the ground softness once I leave the tarmac. And considering that some of the tractors I've driven are older than I am (and I'm in that phase of life where I can tell you how long till retirement), it'll be a long long long time before all the non-automated vehicles are gone.
So for the time being, any automation has to work in a world where a very large proportion of what it has to interact with is neither automated nor predictable.
I had the "pleasure'' of driving a hire vehicle with "Lane Warning".
Fine when driving initially around Toronto staying to lane, but YE Gods the full blown panic mode it went into when in tight fast moving traffic, when having spotted a gap between vehicles to safely move into & I used the indicators to signal that intention.
"Autopilot" is actually a pretty good term for it. Using autopilot in a plane doesn't take any responsibility away from the pilot, they still have to be on alert monitoring the instruments, looking out the windows, and being ready to take over control of the plane. The public being idiots doesn't mean the term is incorrectly used.
Where it gets REALLY problematic is when Tesla calls it "full self driving" which is a product that they're currently advertising and selling, but which CANNOT currently do what the name implies it can, and when used improperly it's been known to kill people.
I just watched a documentary about the Airbus that went down flying from South America back to France. Pilots being interviews talked about the "startle factor" when auto-pilot cuts out or switches to "emergency" mode such that not only are alarms going off, but the handling may change and limits normally enforced by the computers may no longer be enforced.
Pilots flying on autopilot are highly trained individuals and can still get it wrong when "startled" by the sudden change in circumstances and they usually have at least minutes to deal with the issue. A car driver on "autopilot" will be nowhere near as highly trained and probably only have seconds, if that, to deal with an emergency in a massively more complex environment than being at 36,000 feet with no other aircraft closer than about three miles.
The big question is whether the chances for an accident are better with or without ADAS
I don't give a shit if my odds of dying are lower with ADAS if it has a much higher chance of killing me in a way I never would have died if I was driving without it.
If I get myself in a situation where an AI with faster than human recognition of the problem and ability to change lanes, correct a skid, slam on brakes, etc. could save me from something that would have otherwise killed me due to my puny human reflexes, but the price is that it might do something stupid I would never do myself like just drive straight into a stopped vehicle without even slowing down then I will take the first option every time.
I've always said that before people will be comfortable letting a car drive for them it has to be TEN TIMES better at avoiding accidents and death than the per mile averages for human driven vehicles - across ALL conditions because that's where those human numbers came from.
Why 10x? Because a lot of those deaths are avoidable if you simply don't drive drunk, drugged, distracted, or so tired you can't stay awake, which greatly inflates those stats. More are avoidable if you pay full attention ALL the time which humans can't do so they save it for particularly dicey situations. Plus those statistics when averaged give a picture of the "average" driver, when most believe they are above average.
Add all that up and an autonomous vehicle needs to do 10x better than human drivers before it will receive wide acceptance. And can't make any ridiculous fatal 'unforced error' type mistakes a human driver would never make.
"I don't give a shit if my odds of dying are lower with ADAS if it has a much higher chance of killing me in a way I never would have died if I was driving without it."
And that, ladies and gentlemen, is why we can't have nice things. You're talking about being (hypothetically) offered something that will LOWER YOUR CHANCES OF DYING, and you're turning it down because you want to control how you die. That is just stupid. Would you turn down cancer drugs you hadn't developed yourself?
You didn't read their posting properly. The point, well made, is that a huge proportion of the risk of driving can be eliminated by competent driving, so the self-driving system has to be better than a good driver to be worth it for them. Would you want a self driving system in the car if it had half as much risk of crashing as a drunk driver? Because that would still be an improvement overall.
Cancer? Would you take a preventative drug which was half as effective as "not smoking" but doubled your chances of getting arthritis?
I was driving through a town centre at a reasonable speed, well below the limit. A pedestrian stepped into the road without looking, and I went for the brake which I was ready covering. But by the time I got on the brake the car was already braking and stopping.
Though I believe I would have stopped in time, the car actually stopped about 3 metres before where the impact would have been.
In the event, the pedestrian hopped back out of the road, but I was very impressed by the car response and I always will have my own driving augmented by these aids - which will continue to improve. I will always use them because I am only human and not arrogant enough to believe I am an exceptional driver who will never make a mistake. Anything that adds to safety is to be welcomed.
And remember it's not all about you. Any time a poor driver or impaired driver could ram your car if there are no safety systems to prevent it.
The trouble is with stories like this, people also assume that the the deaths where ADAS was engaged were due to ADAS and wouldn't have happened otherwise. I'd guess that in a considerable number of these, the accident would still have occurred, it was more an issue that even ADAS couldn't prevent it. Any maybe ADAS in fact lessened it, reducing the consequences.
I'd guess otherwise - the problem with increased levels of driver automation/assistance is that, unless it's done VERY well, it risks causing the driver to lose focus on the fundamental act of getting their vehicle safely from A to B - the less work you require them to do to complete that act, the easier it becomes for the mind to drift, attention to be given to someth....ohhHHHH FUUUU *BANG*
*sounds of approaching sirens wailing in the distance*
There are undoubtedly cases where ADAS has prevented/significantly reduced the severity of an accident, and there'll be a few cases where ADAS was involved in an accident where the scenario was such that nothing other than not being on that stretch of road at that moment in time would have prevented it. But when we hear about these headline failures of ADAS, there does seem to be a recurring theme to them - namely that by the end of the report there's a good chance you'll be left thinking to yourself "WTF was the car thinking, WHY would it have done that?" - and that's definitely not because ADAS hadn't been given a fighting chance to avoid the crash, but because it did something fundamentally stupid whilst being left to its own devices by a driver who wasn't paying attention.
"If I get myself in a situation where an AI with faster than human recognition of the problem and ability to change lanes, correct a skid, slam on brakes, etc. could save me from something that would have otherwise killed me due to my puny human reflexes, but the price is that it might do something stupid I would never do myself like just drive straight into a stopped vehicle without even slowing down then I will take the first option every time."
Last big accident I witnessed a bloke drove straight in to the back of a stationary car without slowing down.
That's the point, yeah. ADAS is worth using for me if its accident rate is less than my own non-ADAS accident rate.
Problem is, nobody knows any of the numbers there. I don't know what my own non-ADAS accident rate is, beyond "low", because I don't have enough samples. I believe I am a good driver, but it's well known that everyone overestimates their driving skills.
I also don't know what the ADAS accident rate is, but that's more annoying, because whereas my own numbers are mathematically unknowable due to not enough samples, ADAS numbers should in principle be possible to have with some confidence. But we still don't have them.
Which leads me to conclude that I have to agree with you: with this much smoke, ADAS needs to actually be overwhelmingly better than average, in order for me to be able to be certain that it is at least somewhat better than average. I won't use a system that's "quite a bit better than a drunken madman". Actually, just like everyone else, I believe I'm a solidly better than average driver, so I won't even use a system that's "somewhat better than average". It doesn't take much to get to requiring 10x better, yeah.
The only bit I disagree with is that it's better to die by own stupidity than by a computer's, because my own stupidity is preventable. Yes, it's all preventable, but no, statistically, it doesn't all get prevented. That number can be made very low by careful driving, but it's not zero and only a fool would pretend it's zero. It is a positive number, which could in principle be compared to another positive number. The color of the dice I'm about to roll doesn't really matter.
The very short version of this: yeah, I agree with you, but what if autonomous vehicles were already 10x better? I don't think they are, but if/when they will be, I still won't know, because the damn numbers are not available anywhere.
Total US deaths by car accident in the US, 2020: 42,338. (https://injuryfacts.nsc.org/motor-vehicle/historical-fatality-trends/deaths-and-rates/) That's an average of slightly less than 116/day. Hardly "thousands every day".
But even if deaths/mile driven (a FAR more meaningful stat) were, say 1% of what you have with humans driving, you still have to contend with the problem of remote control assassination or mass murder that simply cannot exist without these interposing systems.
If you've been paying attention, these issues have been raised here continuously.
Why would I be talking about the US alone?
https://www.who.int/news-room/fact-sheets/detail/road-traffic-injuries
3500 deaths per day. So, yeah, thousands.
The possibility of dangerous remote car hijack very much should be factored in, when figuring out ADAS accident rate. I don't think it's going to move the scales very much, though. 3500 per day is a big number.
In the UK, as far as I'm aware, advertising cannot be misleading.
If Tesla called their system ADAS, it'd be fine - Automated Driver Assistance, yeah, got it. Assists me. Instead, Tesla called it "Autopilot" and the name alone certainly gives the impression of being rather more than just assistance. I'm pretty sure their marketing-of-the-time was trying to give a self-driving impression too, but they've obviously reigned it in a lot since.
Lately, however, Tesla have had something which surely crosses they line - they call it Full Self Driving. That is literally its name. It is not full self driving at all; it's just ADAS. They warn you that full self driving is not full self driving as if that's somehow supposed to alleviate them of any responsibility...
https://www.tesla.com/support/full-self-driving-subscriptions
It's bizarre. How can they be so completely misleading with a product named in a way that specifically says it is something it is not and this be allowed? Even in the USA, it seems like a stretch.
I think you are getting confused by the wording (which is exactly the point), Basic Autopilot, Enhanced Autopilot, Full Self Driving Capability and Full Self Driving are separate things.
Only a handful of beta testers have access to Full Self Driving.
If you car has the FSD computer then you can buy a subscription to Full Self Driving Capability which is not FSD and is just some extra driver assist features for autopilot.
Once you have access to Full Self Driving Capability you can request access to the Full Self Driving Beta if you are in a location that allows self driving cars.
What do most people understand if a car has "Full Self Driving Capability"?
They've gone and done it again - first they sold something called Autopilot which was not even close to what people understood it to mean and now they've done the same with "Full Self Driving Capability".
And they're still flogging the snake oil with what their cars will be able to do as they have done for the last decade and still want their customers to beta test unfinished software, which is fine if you're beta testing a web browser but not fine at 70mph down a motorway.
Icon is what Teslas seem to be doing lately.
"What do most people understand if a car has "Full Self Driving Capability"?"
Like I said the point is to confuse people. People will assume that the car being having 'Full Self Driving Capability' means that they can use it as a self driving car, when Tesla mean it to be 'this car can self drive, but it won't because that feature is disabled'. Using misleading and similar sounding terms is the most basic way of getting people to buy things they don't need and probably don't even want.
You have a point but most of us only see the word 'autopilot'. Some of us remember the first (I think) Trident landing in fog on autopilot and the knowledge today that most airline aircraft fly and land on autopilot - with pilots only sometimes landing manually to keep their hand in. Because I understand that even the finest avionic autopilot with over 50 years of development can't cope with everything and will hand back control to the pilot. As professionals they are expected to be on the ball and already well aware of the situation where human control might be able to cope with the unexpected better than a box designed to cope with the expected.
Even then some have failed spectaculary. The problem is we, as drivers, do not approach driving with the same level of professionlism and dedication. Released from the chores of routine driving the handover would probably occur when the driver's spatial awareness had become too relaxed to quickly realise and take emergency action in a second or so or detect that 'autopilot' was screwing up.
Having said that - just programming 'autopilot' to drive within speed limits and other traffic laws could have a significant potential to reduce KSIs. The answer lies in the stats yet to come.
I think one of the key non-technical differences between autopilot on planes and 'autopilot' on cars is that the added risk is worth it with planes. With a car if something goes wrong you can pull over and try and diagnose it, with a plane you need to diagnose and fix the issue whilst also keeping the plane in the air. The workload is so much higher in a plane so using automation to reduce it is worth the risk of the automation making the occasional mistake (which pilots are trained to spot).
Some of us remember the first (I think) Trident landing in fog on autopilot and the knowledge today that most airline aircraft fly and land on autopilot - with pilots only sometimes landing manually to keep their hand in.
On the contrary. Unless things have changed a lot since I last had a jump seat (pre 9/11), the actual landing is almost always flown by a human but they occasionally practice a full Cat 3 autoland in case they need to use it in bad visibility. It's VERY much more complicated than landing manually - although even manually the autopilot takes it down to 50' or so.
I find the whole Tesla marketing ideology around "Autopilot" to be incredible - even marketers (should) have ethics.
My car has driver assist with lane-keeping, active cruise control, vehicle detection, pedestrian detection, object detection, traffic light detection, collision avoidance and self-steering, but if I take my hands off the steering wheel there's a three stage alert:-
1 The dash and steering wheel light up like a Christmas tree
2 The car gets annoyed, beeps loudly at me and vibrates the steering wheel
3 The system cuts out, slows the vehicle and (I think, not going to test it) tries to pull over safely and comes a complete stop.
The user manual is covered in warnings that the driver must be alert and is at all times in control and fully responsible for driving.
It's still a great car to drive and on motorways it does do most of the driving itself, subject to the above, which makes it a less exhausting task.
What they're saying there is that their figures might not be accurate because a manufacturer might currently be breaking the rules.
Also I guess there's going to be problems with detection, threshholds and reporting. How serious does an event have to be to define it as an accident, and even if you can agree on that, are the cars capable of detecting and reporting that. Plus if the owner has disabled all communications on the car, it can't report back to base what just happened.
There's a book by the creators from the Red Dwarf series set in the future where there are fully automatic cars. The hero tries to take control of the car and after fighting through about 2 dozen prompts of "Are you sure you want to do this? Really sure?" he gets control of the car. Only to discover that the steering wheel is just a prop, and he can't steer, accelerate or brake.
For some reason, that story always reminds me of something Tesla would do, once they've cracked this fully automated driving thing of course...
One problem with modern cars is that they work too well. They're quiet, comfortable and so they isolate you from the road environment. Add in decent tires, anti-lock and vehicle stability systems and you can drive like a fool and still get away with it most of the time. Driver safety systems are a great idea but not if they encourage the less responsible to be even more foolish. This is the problem with full self driving. It isn't foolproof so its not going to stop us from being foolish and may actually make things worse.
Car safety systems like lane maintenance and adaptive cruise control are very useful for catching momentary inattention so I'd want any new vehicle I buy to have them. I don't want full self-driving, at least not at highway speeds (around town in dense traffic is another story -- but then I avoid this if at all possible). Self-driving only reacts to what it sees; it might see better than me but its perceptions, and in particular its ability to anticipate problematic situations, is severely limited. At best its going to be an extremely bad driver, on total alert when situations are straightforward and completely missing building hazards.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
