While truly self-driving cars are surely just around the corner, for now here's an AI early-warning system for your semi-autonomous ride Self-driving cars could be equipped with a trained early-warning system that alerts the person behind the wheel whenever it realizes it's entering a situation where a human driver has had to take over before. Today's systems, like Tesla’s inappropriately named Autopilot with “full self-driving capability," rely on software to …
So we have a set of known situations on which the driving S/W has been trained and everything else is, therefore the set of unknown situations. AFAICS this then takes some of the unknown situations and creates a new set of known unknown situations. Or have I missed something?
This experimental model was thus able to accurately predict potentially dangerous scenarios with 85 percent accuracy up to seven seconds before they occur.
I'm torn between pointing out that the experimental model has an "up to" 85% probability of passing the hazard perception test a human driver has to pass before being allowed to drive in the UK, and that it takes 6.87 seconds to bring a vehicle to a halt from 60MPH so in many real world circumstances "up to" 7 seconds is too slow. Hence the Hazard Perception test with a minimum pass rate.
I do wonder how many of these autopilots would pass a driving test.
Don't worry, they'll all pass a typical US driving test with flying colors.
Most states don't have anything like a "hazard perception test", the driving part of the test takes less than 10 minutes and usually consists of driving less than a mile, then parking successfully (there are rumors that some people have to parallel park, but I've never known anyone who did). People do still fail, but it's pretty rare.
Oh, and there's a written test, it's typically 10 multiple guess questions taken from a published question pool.
Took the theory test a long time ago, I think it was around 40 multiple choice questions, but what stands out is someone failing and complaining that some of the questions she got wrong were about the different rules that apply to teen drivers and as she didn't have any children she shouldn't be expected to know the answers.
I don't want one just around the next corner I come to.
It's neat (in a Geeky way) that they can do as good as they do now, but it's no where near good enough yet to be driving on the road other than for testing with an alert driver ready to take over and save the day.
Truthfully, I'm sure some self driving cars drive better then some people. But that is a low bar.
But, after acting as a passenger, will they be alert and even itf they're alert will they take over in time to assess the situation and act?
The underlying theory of both this and, in fact, speed limits, is that "we don't trust the driver except in the most difficult circumstances".
Generally the people that see autonomous vehicles as a boon are those who would like private transport but cannot drive.
Currently the world is being offered part-time self-driving with, oops apocalypse, over to you Mr.* Licensed Driver, I'm not taking responsibility for this cluster**** of a situation we're now in... and, 'When did you last read your insurance policy?'
The point being that Mr.* Licensed Driver is still required to babysit the robot, so no private transport for Mr.* Cannot Drive.... so I personally do not see the point of the rush to push a half-baked 'solution' onto the roads of the world when Mr.* Licensed Driver should be happy to have control of his own destination rather than being a passenger effectively taking the role of driving instructor and concentrating on when to step in when something goes wrong (a driving instructor might want to comment here on 'how' stressful that can be).
I reckon most drivers would also (like me) consider themselves to be 'bad' passengers, unable to detach themselves from the traffic situation about them when being chauffeured.
(Mr.* other gender pronouns are available)...
With delight I just purchased a bottom-of-the-range car... the only driver assist toys it comes with are ABS and a thing that goes beep if you get too close to something while reversing.
It occurs to me that in forty years of driving I have never had the ABS come on. I wonder if it works?
I was very pleased to discover that my new-to-me cheap-brand car, while it does have rear parking sensors also has a persistent "off" switch. Hit the button and the screaming banshee is silenced permanently.
Totally unlike my wife's not-quite-so-cheap-brand car, where you can't turn the thing off at all, and which has a near useless 9" screen which, while you can turn it off, comes on again when you hit just about any control, including selecting reverse. As I often have to reverse in a dark car park using the mirrors, this bright light near the windscreen somewhat messes with my night vision. A jab at an onscreen icon will blank it again, until you start getting near an object, when as well as bleeping, the blasted screen comes back on!
The big thing my cheap car is missing is folding mirrors. My last car (sadly deceased at just over 201,000 miles) had electric folding mirrors, very swish. My wife's car has mirrors you can fold over manually, but this new-to-me car's mirrors cannot be folded at all, other than temporarily on a spring. Really odd, that.
ABS? I'm convinced that on one of the rare occasions when I know for sure it came on, it actually made the situation worse. Emergency braking on gravel on top of tarmac. Without ABS the one locked wheel would have ploughed through the gravel to the tarmac underneath. With ABS, the car barely seemed to be slowing at all.
M.
I lapped my miata at a local raceway. Left the braking a bit late on one corner, and managed to lock it up. I was fine, but I have no idea how long the ABS hasn't been working. Tried it a bunch of experiments, and did goad it to engage in an icy parking lot once and never since....
My father drove till he was 86. He was partially sighted. My mother would tell him about hazards in the road. As he was very deaf in his left ear this was not terribly usefull, doubly so as although her eyesight was much better than his she still had cateracts in both eyes.
There is a solid use case for self driving cars.
"There is a solid use case for self driving cars."
There is indeed. But I'm a little hazy on what's supposed to happen when, for example, the car encounters a situation beyond its capability and tries to turn control over to a blind or incapacitated driver.
Self driving is self driving. They crash.
But then how should they crash? Save the passenger(s)? Save maximum number of lives? Start an instantaneous online auction (money to the company, naturally) as to what behaviour should be followed? Oh, the choices.
Looks like an 1 April article but these surreal days, one never knows.
This post has been deleted by its author
1) people who due to age or other infirmity can no longer drive safely / at all
2) people who can't drive safely / at all temporarily due to injury
3) children too young to drive (wouldn't it be nice if your car could pick up your kid at school for you?)
4) drive you home from the pub if you've had too much (or anything?) to drink
5) allow you to work / relax / eat / sleep / etc. during a longer trip
6) people who just view driving as a chore and would rather do other tasks - some of whom do those tasks (i.e. using their phone) while driving already
I think it has been way overhyped and those who thought it would be arriving in just a few years a few years ago were off by a decade, but it will eventually overcome the hurdles in its path. We just need to stop the idiots like Musk who are trying to push something not fit for purpose as "full self driving" (which even their lawyers recently were forced to admit is only SAE level 2) and wait for regulations and insurance to catch up before proceeding. Pushing cars onto the roads before they are ready will set things back by many years by souring the public and government against it.
1) certain disabilities aside, if a person is that old and infirm they probably don't need to be going out alone anyway. By the time my father finally gave up motoring, he couldn't go out safely without my mother in tow; he sometimes "freezes" due to an arthritic spine. She gave up driving at the same time. She wanted to give up earlier, but felt she couldn't. Gatepost scratches on the car suggested she should have.
2) I'll give you
3) No. Absolutely not. If you live too far from the school for the child to walk home, there'll (usually) be a school bus. Help reduce congestion, improve pedestrian safety and save the environment. Exceptions for after-school clubs I suppose, but people have coped since education became universal after the (?)1902 act. If I'm not available to collect my child, chances are my car isn't either as it will be with me at work.
4) Best course of action is to frequent your local and walk home, preferably with company, rather than driving to the trendy pub miles away, but I realise this isn't always possible. Or if so sozzled you can't walk, get a taxi (self-driving?) Bit difficult for me to comment further as a non-regular pubgoer.
5) Hmmm... I can see the attraction, but boy I'd find it difficult. Maybe my children or my grandchildren wouldn't. Trains I can cope with, but I get jittery on busses, particularly on fast roads.
6) Perhaps the real answer is just to reduce the need for lots of travel? Then again, if this last year has taught me anything it's that mail order really isn't the be-all and end-all and I've really come to appreciate "proper" shopping and see it less like the chore it used to be (though the queueing can sometimes be tedious), and I can't be a good advocate for reduced travel while my daily commute is a 90+ mile round-trip.
I totally agree with you that it will come eventually, though I'm not convinced that is necessarily a great thing. It will take a while for legislation, insurance, driver ("supervising driver"?) training and such to catch up, so perhaps it's good that these not-really-ready devices are being touted now, while their flaws are obvious to anyone who isn't blinkered by being invested in the technology.
M.
You should maintain a small portfolio of pubs. Even in the countryside, where the nearest may be several miles away.
Walking for an hour through the rain on a Wednesday night only to find that the after-work crowd have gone home... is not a workable solution.
So we have an SA (Semi Autonomous) vehicle controlled by an AI that has to be watched by an HI (Human Intelligence) who will be told when they need to take over control of the SA from the AI by another AI that is also watching the SA?
I think I may be able to save some time and money here.................
This is warning the wrong person. The owner of the prat-mobile can perish with his vanity robot and not a tear need be shed.
Every automated vehicle (even dicky adaptive cruise control) should require a roof mounted warning light, which signals slowly when the computer idiot is in charge. If it begins to transition to the biological idiot, the warning light should speed up and intensify until the system is reset.
So you have a new Artificial Guesser guessing whether the first Artificial Guesser guessed right. But that Artificial Guesser is going to need it's own Artificial Guesser to guess if it got it right. Ad Infinitum.
Sounds like a full employment plan for Artificial Guessing researchers.
Typically most of these systems (used on fly by wire systems etc) as you say use 3 computers to check the input, and if one is "wrong" then it's input is discarded, and in certain use cases the computer that came to the wrong decision is cut out of the loop as being possibly subverted.
So if they have the ability to detect when things are likely to go wrong, why not feed that back in as an additional input to the primary AI/ML system to tell it not to do what it thought it was initially going to do, hence resulting in hopefully a more positive outcome.
OK, so I guess we could get to double / triple / quadruple negative type situations, but hey, it would be more fun to watch s people try and copy real intelligence (and real stupidity)
Assuming the human passenger/driver is busy doing something else -- working, sleeping, being entertained, whatever -- then when the system decides that its just too difficult and needs the human to take over there will need to be some kind of briefing, getting up to speed, before the human can take over. If the vehicle can stop safely and alert the drive, fine, but if its going at speed then its just not going to work -- by the time the human figures out what's going on and why then that human may well have been killed.
I'm interested in self-driving because as we get older our situational wareness tends to deteriorate (its this that modern driving tests are looking at in those simulators that ask you to spot the hazards). I need the machine to be an assistant, to stop me doing something silly while I figure out what's going on. I don't expect to have to second guess the machine -- if it gets into a situation it can't cope with and expects me to just take over then its worse than useless. There are plenty of ways to mitigate the risk but they involve abstract thinking that is beyond AI such as altering route, ttime of journey and so on (a common trick is that when you're driving at night avoiding unprotected left (US) or right (UK) turns -- adjust where and when you drive.)
but for some reason we seem to expect self-driven cars to be perfect, or at least near perfect. If we allowed most of the developed systems to be used on the roads now, yes there would be issues and yes people might get killed/injured, but a whole load less killed/injured than if we continue to let humans drive those same cars. So, it seems we wait until some indeterminate date in the far future when a country finally takes the plunge and realises that the best way to reduce car accidents is to remove the weakest link (the human) from the decision tree. From what I see on the roads, far too many humans have not evolved sufficiently to deal with moving at anything faster than walking pace (4Mph) but still somehow manage to get driving permits.
Some time in the near past a transportation economic journal remarked on autonomous vehicle in urban areas as a possible source of increased pollution and congestion. You get out to do your business and your vehicles cruises looking for a spot to park. Another offering was banning of private vehicles and the urban authorities use the autonomous vehicles on a roving basis in place of the current transportation systems . Mack Reynolds used this in his stories but one draw back he had pointed out was when the authorities want you the doors seal and delivered to their sanctum .
The current mix of vehicles is most dangerous of the possibilities and until the mix is mostly autonomous vehicles with just a few aberrants with the vehicles (AV) communicating velocity and acceleration will we be close to 1900 prediction of The Scientific American that the streets of the cities will safe from drunken riders and runaway wagons.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
