Google: Our self-driving cars would be tip-top if you meatheads didn’t crash into them Google's self-driving car engineers have released their first progress report, and they said they will publish monthly updates from now on. "We’ve made a lot of progress with our self-driving technology over the past six years, and we’re still learning," the report [PDF] stated. "Every day we head out onto public streets so we …
"You're driving it wrong."
Seriously though, the real question is: would these accidents have been less likely with a human behind the wheel?
From the sounds of it, not really. About the only iffy one is the rear enders, mainly because people drive by pattern and if the person behind the gCar was thrown off of unexpected behaviour by the gCar, they may have reacted incorrectly - which might not have happened if they were behind a regular car.
Sure, but can the self driving car anticipate the "unexpected" behaviour? Quite often a real driver can tell what another driver might or might not do based on how they are driving,where they are looking/not looking and just their general driving style. I've avoided plenty of accidents because I could see another driver was an inattentive idiot so I was prepared even before they pulled their bonehead move for the day.
Let's give it their first law: Always behave as if the other vehicles are out to kill me. Worked for me as a cyclist, motor or bicycle. In one case I was run off the road in a rocky part of Southern California by this idiot lady. Her son had died riding a motorcycle. The motorcycle California Highway Patrol officer was not amused. Neither was I but that's beside the point as I was never called for the trial.
Get programming people!
That's a thought. A computer-driven car can be programmed to assume the worst: that a car might suddenly stop in front of them, swerve into them, cut into the narrow gap you normally leave for the first instance, assume the end of a blind curve can be roadblocked, and so on. And make all the car's driving actions work under those assumptions. That way you don't need cues to be prepared for trouble: you're prepared in any event.
So the car leaves a safe gap behind the car in front, which is promptly filled by another car, so it has to slow to establish a safe distance again, and another car cuts in ... all the while it's decreasing the gap and increasing the closing speed of vehicles behind it.
As a biker, you develop a "sixth" sense as to the behaviour of other road users very quickly - either that or you become road-kill.
I've had my fair share of accidents over the years (and yes, some were my fault), but I've avoided far more simply by being able to anticipate when that driver in the 4x4 with the "Child on board" sticker is about to do something stupid.
Many of these things will be down to almost imperceptible behaviour, something that will be impossible to encode into the self-driving car simply because as humans, we would find it very hard to describe quite why we knew the muppet was about pull that bonehead move of the day.
As a biker, you develop a "sixth" sense as to the behaviour of other road users very quickly - either that or you become road-kill.
I agree. It is amazing how much the realisation of imminent pain/death can improve your perception.
My bike instructor put it this way: If you have an accident in a car, you'll dent the bodywork. If you have one on a bike, it's going to f*****g hurt! This is a great motivator to be aware of your surroundings, notice the guy on the roundabout who has not spotted you, back off approaching a blind bend, and beware of sheep who think that the best place to be on a foggy day is sat in the middle of the road.
Still there may be an advantage to the GoogleAI: it constantly checks for texting and call behavior from all nearby cars and via comparison with registration information identifies vehicles with smartphone-distracted drivers so as to maintain increased separation.
Or at any rate, that's why Google says they want all that data, right?
Decently well, I would think. Radar doesn't rely on metal, and to stealth a vehicle requires a combination of radar-deflecting design and radar-absorbing paint, and they'd still be of limited use in multistatic (they'd spot the dead zone against the background) or mobile (the case here, it can hit differing angles) detectors.
I wouldn't have thought they'd bother anyone for long - being only milliseconds from death, their numbers must be decreasing pretty rapidly. I mean, they're no more visible than not-particularly-visible pedestrians, and if they dared venture out on the roads there'd be a bloodbath.
I wouldn't have thought they'd bother anyone for long - being only milliseconds from death, their numbers must be decreasing pretty rapidly. I mean, they're no more visible than not-particularly-visible pedestrians, and if they dared venture out on the roads there'd be a bloodbath.
I like your pure Darwinistic view of the world. Of course, the insurance may look at this differently (and so would the stealth cyclist) but as a long term problem solving strategy it's flawless.
maybe he was a fan of Stephenson's protagonist in Zodiac - "if they can't see they can't hit you" school of thought.
I'd like to know how the google cars compare, statistically, to human drivers for same mileage & same conditions. And let's not forget statistical confidence, mostly wrt sample size.
Pretty bloody good.
If memory serves me right, the probability of a new driver to have an accident in the first year in a metropolitan area is > 50%. This is for London which is of course infinitely worse than mountain view.
As far as Two more were down to other cars not obeying stop signs, my late dad used to have a couple of sayings: "There is a special alley in the graveyard for those that thought they have priority" and "I have priority, but does the moron on the other side of the intersection know that".
The arrogant writer has his blame game on. Google's approach seem to be getting the rest of the world to conform to its ideas about roads and how cars should behave, so that their ideas will "work" - but that won't work because, unlike the search space, they have no monopoly. What they didn't report here was that on their first cross-country trip, they "discovered" that the road surface markings were different in each state, read; the guys developing this haven't a clue about the larger world of driving outside of their home town, which is typical of CA parochialism, and they want you to trust your life to them. They can barely manage well paved orderly thoroughfares. They don't have a monopoly on the roadways, although they're trying desperately to buy influence in Sacramento in order to get permission to further menace the already menacing freeways of CA with yet another home-brew custom car.
There are several types of contributors to this section, those who feel they have additional insights to a situation, those who have alternative experiences and those who seek additional information. I was unable to decide which you tried to be. I was left with the impression that you simply wanted to expose your prejudice and preconceived idea of how the world should be.
Put simply if you are starting from any location, it does not matter which you will initially profile that location not say what New York or for that matter New Delhi has if you are in California, Surely it would be basic common sense to start of with the right profile? Have done that and got the thing working, expansion to areas where all sorts of variables add complexity would be the next stage. Should new profiles be kept as a monster data base or paged in and out according to how many specific data points are noted. Road marking may well vary from State to State, that is a variable that will need to be covered, the style of trees the dampness from bone dry to water logged, the closeness of buildings indicating congest town or open country all have to be mapped in some way to build a risk profile. Hopefully new drivers live long enough to acquire the skills that will allow them to take such issues in their stride. Returning for a moment to road markings, they may well be part of a set standard per local variation. However, wear and road works can also affect their presence and appearance as can the presence of high sided vehicles and other variables. All such variables do have to be considered when controlling a vehicle. Knowledge of wind speed should also alert drivers to the risks of different situations, how many do take account of such issues - not many if they are of the wetwear variety that I see weaving about in cross winds. Some, (many?) cannot even detect changes in the road layout ahead causing precipitous manoeuvres when roads narrow. One ended up facing fully the wrong way after braking too hard while doing two or three other changes of direction. They did not hit anyone.
A while later close by similar stupidity resulted in five deaths. No autonomous driver-less control required, that was down to wetwear meathead driving.
This is a research project, for such someone to complain that research finds things out is does not show a high level of understanding of why research is carried out. I may be the only one who expects research to discover aspects that are not obviously apparent to desk jockeys, but someone will probably tell me I am wrong on that point.
I did once avoid being rear ended on a country road, I stopped due to stationary farm traffic, the meathead behind was talking, not looking I could see they would not stop so I drove onto the grass verge, my movement drew the other driver's attention to the road, he braked and steered out towards oncoming traffic stopping two cars in front of me along side the stationary traffic. Would his error have happened with a Google car - I doubt it would.
Put simply if you are starting from any location, it does not matter which you will initially profile that location not say what New York or for that matter New Delhi has if you are in California, Surely it would be basic common sense to start of with the right profile?
Interesting that you're fully prepared to overlook the fact that Google happens to have world's largest database of road layouts in Streetview.
Dear Ac,
Having considered your narrative, may I make a few observations:
- if googles plan to develop autonomous cars is so shit, why don't you show them where they went wrong? I am sure they would appreciate your sage thoughts
- were it not for homo sapiens pushing the boundaries (and occasionally tripping over them) WE WOULD BE STILL LIVING IN CAVES watching TV by candlelight...
It is called 'evolution'
Btw how did your last Nobel committee nomination go??
Regards,
Jay
This post has been deleted by its author
The arrogant writer has his blame game on. Google's approach seem to be getting the rest of the world to conform to its ideas about roads and how cars should behave, so that their ideas will "work" - but that won't work because, unlike the search space, they have no monopoly
I noticed the downvotes here, but in support of your statement it's worth observing that other, non-Google efforts don't appear to have Google's problems of realising roads may differ from those at home. This may be because those companies already build that rather important part, the car, and thus have a far more developed insight in what a car may encounter on a global basis because they already sell globally and as a consequence had to deal with roadside issues.
Not that that is forgivable for Google: as it created world's largest database of actual street footage in Streetview I would like to hear how it is possible that it appears they entirely ignored that in their development.
To me it sounds like they are starting simple and adding complexity as they need too - the benefit of this is if they only add more complexity once the previous stuff is working then they know in which area they need to look to find the problem.
There is a possibility of painting yourself in a corner with this approach so it will be interesting to see how they (or anyone else building self-drive cars) deal with this.
Are Google's self-driving cars clearly marked? I don't know, but I assume they are. Even if not, I suspect the folks in Mountain View (where they test the cars) can recognize them by now.
How does that change the behaviour of other (human) drivers nearby. If you see a self-driving car a bit ahead in the next lane, would you instinctively steer away to reduce the chance that something unexpected happens? An experienced driver can usually guess the intentions of other drivers by observing them - we all do that instinctively. If that Toyota in the next lane looks suspicious, let me give it some space, eh? If you don't know how the robot will behave, will that make you behave more carefully? And if you get used to it?
In other words, is there a bias that makes accidents involving Google self-driving cars less likely?
I have never driven next to a self-driving car, so I have no idea how I would change my behaviour.
I would have thought most people would take the opposite approach, driverless cars will be fair game for cutting up, tailgating, pulling out in front of, etc, even fun games seeing how much you can force them to take evasive action, swerve dramatically and so on.
While I can see the fun in playing chicken with a google car I suspect google would have very good documentary evidence of your behaviour.
In local town, the school bus drivers had a simple arrangement with the local police. If a car was seen driving carelessly around the bus, the local constabulary would pay a visit to the driver and scrutinise the car for defects. For most kids their first car was barely roadworthy so finding defects was trivial.
This post has been deleted by its author
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
