Slashdot Mirror
Blind Man Test Drives Google's Autonomous Car
Velcroman1 writes "'This is some of the best driving I've ever done,' Steve Mahan said the other day. Mahan was behind the wheel of a Toyota Prius tooling the small California town of Morgan Hill in late January, a routine trip to pick up the dry cleaning and drop by the Taco Bell drive-in for a snack. He also happens to be 95 percent blind. Mahan, head of the Santa Clara Valley Blind Center, 'drove' along a specially programmed route thanks to Google's autonomous driving technology. Google announced the self-driving car project in 2010. It relies upon laser range finders, radar sensors, and video cameras to navigate the road ahead, in order to make driving safer, more enjoyable and more efficient — and clearly more accessible. In a Wednesday afternoon post on Google+, the company noted that it has hundreds of thousands of miles of testing under its belt, letting the company feel confident enough in the system to put Mahan behind the wheel."
Grow up. They have done 200,000 miles with a person sat in the driver's seat to ensure he can take control if anything went wrong. On a pre-programmed route this is a very stable system and he had someone beside him in the passenger seat (I also wouldn't be surprised if it was dual-control so the passenger has access to a brake pedal). Meanwhile this technology could eventually change the lives of millions upon millions of disabled people, damn right it deserves the publicity. With your attitude we'd never have wheelchairs or crutches or surgery, all things which, the first time out, could have resulted in injury but have been life changing tech for millions.
Boy, if that's not one of the most appropriate metaphors for our time...
Soon, they'll just jack us into our pods, and grow us for the power we generate. :-)
"Flyin' in just a sweet place,
Never been known to fail..."
However, in the UK, eyes are no longer tested after you do your driving test. So in reality there are many drivers on the roads with substandard vision who have not been tested in decades (I got rear-ended on my bike by one on a straight road, in good visibility, while wearing bright clothing. It was an elderly gentleman who had no corrective lenses - he just ploughed into the back of me). At least when I was in Texas you got an eye test for driving every 4 years, not a "squint at this numberplate" eye test, but one using an optician's machine.
Oolite: Elite-like game. For Mac, Linux and Windows
The law currently states that it's illegal to drive when legally blind, which is defined as a visual acuity of 20/200 or less using best correction possible. If your vision is better than 20/200 but still bad, you're assessed on a per-case basis. This suggests that anyone with visual acuity better than 20/200 may be allowed to drive using this technology (or future derivitives of) if it is considered to be a corrective device. How they would measure such improvement is unknown, since visual acuity tests certainly don't involve any driving. This is speculatory, of course, since there will have to be a review of driving law if this kind of thing becomes commonplace.
However, in the UK, eyes are no longer tested after you do your driving test. So in reality there are many drivers on the roads with substandard vision who have not been tested in decades (I got rear-ended on my bike by one on a straight road, in good visibility, while wearing bright clothing. It was an elderly gentleman who had no corrective lenses - he just ploughed into the back of me). At least when I was in Texas you got an eye test for driving every 4 years, not a "squint at this numberplate" eye test, but one using an optician's machine.
True up to the age of 70, after which it is part of the medical carried out every three years. In theory drivers are responsible for getting their eyes tested and reporting themselves to the DVLA if they cannot see well enough to meet the requirements. In practice many people who think that their site is not good enough and cannot be corrected avoid being tested so that they can continue driving. I even knew someone who drove a 3-wheeler car as he had only passed a mtorcycle test before his eyesight deteriorated.
Later she was called to the police station to make a statement. The police had arrested the driver. He said he had not seen the crossing because there was thick fog (mildly overcast). Then they discovered that he was registered partially sighted. He had cataracts.
He was convicted of:
His comment to my wife at the police station? "You've spoiled my day". He simply did not realise how serious his offense was.
So I applaud what Google is doing, because I've worked with computers for nearly 35 years, and human beings for over 40, and if the system is designed I would trust the computer over the human being any day of the week, and double on Sundays (drunks with hangovers).
From scarped cliff or quarried stone she cries "A thousand types are gone, I care for nothing, no not one."
isn't the point that the car is actually driving.
But that would only be legal in Nevada.
"'This is some of the best driving I've ever done,' Steve Mahan said the other day.
I guess he usually uses those pavement reflector thingies to drive by braille.
Driving home tonight there was a young kid playing quite near the road, so I dropped my speed in anticipation of him doing something stupid. He didn't, but I did wonder about the google car making those sorts of calls. I'm sure these google guys are pretty clever and have thought of all these things... are there any video's of self drive cars reacting to these sort of situations?
Like that feeling you get when you see someone else on or near the road and you aren't completely sure that they have seen you and you react by lowering your speed to avoid a potential collision. It's got me out of trouble a few times. If there was an accident you probably wouldn't be at fault, but you've gone one better and seen the accident coming and avoided it.
I'd want to see lots of video evidence of a self drive car doing this sort of thing before I'd be happy sharing the road with one.
Yes, because Google (and the authorities letting these cars on the roads) would have *never* thought of the possibility of pedestrians running in front of these cars.
Quick! Get in touch with them and bring this to their attention!
Most likely an autonomous car can react quicker to an obstacle running in front of it faster than a human can.
And given the average human's driving ability, it probably fares no worse when it comes to being in the correct lane at complex junctions.
Maybe it will need two more orders of magnitude of testing and refinement before it can be included in cars that the blind person can be alone in, but progress is progress, and this is surely a milestone?
Of course I will jailbreak my car when it comes with such technology, so that I can add my own AI modules, such as "HunterKillerMod" that turns the car into a pedestrian killing machine. And "DestroyAllCyclists" too, obviously (who won't have that installed?).
He had a policeman sitting next to him ...
"Mahan has no driver's license, of course -- just one of the hurdles that had to be crossed: Google enlisted the aid of Sergeant Troy Hoefling with the Morgan Hill Police Department to accompany the drive."
Puteulanus fenestra mortis
This is speculatory, of course, since there will have to be a review of driving law if this kind of thing becomes commonplace.
Interesting times ahead. For all my reservations, there will eventually come a time when a self drive car is better under all driving situations than the average road user, and the generation after that actually "driving" a car will be "retro" and a steering wheel will be something kids ask about when they see one in a movie.
Alternatively, by then kids will be plugged into their computers at birth and never move from their beds...
And to go a little further, technology doesn't get sleepy. Technology doesn't get distracted by cell phones, GPS systems, or the radio. Technology won't have a blind spot. This is going to be an incredible advance. I'm much less worried about a driverless car hitting a pedestrian than I am the average driver hitting one.
I'm a leaf on the wind. Watch how I soar.
I'm not sure what the exact cutoff is(probably varies by state); but they do do some eye testing during the licensing process and you can lose your license for doing sufficiently dreadfully on the test. There are also certain conditions, it is my understanding, that can trigger a compulsory re-test.
Trouble is, the licensing tests are quite infrequent and people can go rather rapidly downhill between them(and, in much of the country, once you are too old to drive, you might as well go to the nursing home to die; because you are now about as independent as you were at 14...) The amount of risk that they pose to others is pretty selfish; but not having a license is Serious Shit in large areas of the US, in addition to the direct inconveniences of aging, so it is entirely understandable that people keep doing it for years after it stops being a good idea.
It is probably also the case that voting patterns really don't help: very young and very old drivers are both menaces to themselves and others. However, only the latter group votes. This, I suspect, is why much more scrutiny is given to the former(despite the fact that virtually all of them will become less dangerous as they gain experience), while the latter are substantially 'grandfathered in' by pro-forma renewals of decades-old licenses, despite the fact that they'll keep getting more dangerous until something eventually kills them or makes operating a vehicle physically impossible.
They can also drive safely millimetres (like inches but smaller) apart from each other, massively increasing the capacity of the existing road network.
I've seen that thing MERGE WITH MOTORWAY (freeway right?) TRAFFIC!!! 8@~~
It's bonkers clever. I want one. Where we all just sit around the table inside it having breakfast.
Automated cars are also unlikely to rip along at 80-90 kph in a 50 zone like the psycho-cabbie who nearly ran me down on my wake-up walk half an hour ago, too.
I do not fail; I succeed at finding out what does not work.
Did they think of the possibility of driving over a cliff-edge while out of GPS reception?
If the Internet is to be trusted at all, I'll take the chance of a self-driving car careening off a cliff due to lack of GPS reception over the chance of a human careening off a cliff because of GPS reception.
Does it detect ice, snow, oil, sand before the wheels are there?
Humans certainly don't... but there are already automatic traction control systems that do an excellent job maintaining the vehicle's footing in all but the more extreme situations - I can't imagine it would be that hard to send that data to the pilot AI and have it react by slowing down. Also I'd imagine it would be easier for the computer to detect ice and such using sensor data (IR cameras to detect road surface temp, lasers reveal changes in surface reflective properties, etc.)
What about kids throwing stones off the top of a bridge onto the passing cars (common problem in the UK - someone died just the other month from this)? Is the car looking UP too and determining their intent?
Again, I doubt humans would do much better. The radar systems on an automated car could conceivably be used to detect objects that may hit the car even from above and some evasive/mitigating action could take place - with better reaction times than a human driver.
This is why even a jumbo jet - so of the most highly automated and tested machines in the world - has TWO HUMAN OPERATORS. And even there, they have TWO because the first can't be trusted on their own (proven by that recent thing with the pilot).
Again, though I don't keep careful track of these things, there seems to be more incidents related to human error than automation error. Specifically the humans overriding the automated systems to correct for a problem that didn't actually exist.
If you honestly, seriously, think that you can reliably determine the outcome of a machine complex enough to obtain all that data, you're an idiot.
Humans are essentially machines much more complex than that, and have tens of thousands of years worth of historical precedent for doing incredibly stupid things despite having accurate information - yet somehow they are more trustworthy than a machine just by virtue of not being a machine? This kind of argument instantly refutes itself.
How do you test the system for these things? Tens of thousands of hours of real-world driving. Considering all a human needs to legally operate a 2-ton projectile is roughly twenty minutes worth of testing (if you're lucky!) I'll take my chances with the machine.
=Smidge=
Most of the problems you cite are *highly improbable*. Nobody is claiming that a driverless car will never make a mistake, but the facts are many automotive fatalities each year: http://en.wikipedia.org/wiki/List_of_countries_by_traffic-related_death_rate
This is because humans make mistakes. Our current system is not 100% safe. A replacement system does not have to be 100% safe, just better.
To evaluate a driverless systems success it is not meaningful to look at the least likely cases, but to look at overall whether or not it will reduce of increase number of deaths. If 90% of deaths are due to drunk drivers, and driverless cars only fail when say a bridge collapses (a very rare incident), then this will be a net win.
Why would it need to tell the difference between a human and a non-human? If it's not in the way, it's not in the way. If it's moving in a manner that will cause it to be in the way, then react accordingly. A human standing at the curb and suddenly running out into the street is no different, functionally, than an empty trash can getting thrown into the street by a gust of wind. An obstacle is an obstacle regardless of what it's made of and regardless of whether or not it's sentient.
=Smidge=
I know some human drivers who have no idea how to merge onto the freeway.
Frankly, I have much more faith in a Google computer driving my car then I do the other humans on the road.
Reaction time is also cut considerably, as is the time it takes to physically perform whatever act is deemed the best course of action. If "slam on the breaks" is the action, the car doesn't have to lift its foot off the pedal and move it over to slam the brake -- the car's already braking.
A child running in front of a car is a recipe for disaster either way, but the kid is probably safer with the driverless.
Was that you? Sorry, I was trying to get FP...
I don't know if there is but from what I understand, this thing is incredibly intelligent.
If it was truly smart it would stay in the garage and never come out. Driving is a curious game, the only way to win is not to play.
Did they think of the possibility of driving over a cliff-edge while out of GPS reception?
Or what happens if a bridge collapses? Does the car detect the void underneath it and stop, or just think it's a steep hill and plummet over the edge?
Google cars use only GPS to get direction. The actual driving is done with laser grids and radars (for near distance) and video camera (for long distance). So the car doesn't drive according to what it "thinks" should be there according to the plan, but it drives according to what it "sees" (with its sensors). Any of the situation you mention will end up with the car detecting a lack of drivable surface, stopping, asking its GPS for an alternative route and going another way.
Wherever a human driver will react the same, or will be too busy getting distracted with on-board enterteinment/smartphone/passengers/news paper, etc. and fall into the hole is left as an exercice to the reader. (yup, we've already had stories on /. of clueless drivers wrecking their cars because the GPS told them to go a certain way).
There are a BILLION and one problems, that only happen once in a lifetime.
And that's why you put the stuff on extensive testing. Already in the range of hundreds of miles on actual raods with actual traffic in the case of Google cars.
Yes, it won't take into account some really weird exceptions. But... Humans make mistakes too, and mostly in normal everyday boring situations (because they are boring and the brain kicks into "autopilot" routine mode). (And chance are that some of the really weird situation are going to be "missed" by the human, not because the human wouldn't have had reacted correctly, but because the human was to bored to pay attention). Also even if you, the human driver, think of hundreds of situation which might be missed by an IA, but where you think you'll be able to react correctly, I can probably think of situations where you drove perfectly well, but still got into an accident because of some other driver.
At some point in time, we will reach the situation where an autonomous car (even including the accidents due to weird rare situations) will cause a lot less casualities than a human driver (who might just not be paying attention).
The point of this publicity stunt is to show that, given the current extensive testing the cars have undergone, this point in time is nearing soon.
And, also, the advantage of autonomous cars is, as the wierd situation happens, they can be analysed and the programming can bu updated, leading to even less casualities on the long term. .
Whereas, with human drivers, you can't just magically "programatically remove from the road" asshole, idiot and dristracted drivers
You can't verify a system on this scale. It's like trying to verify a Kinect. You just cannot guarantee what it will detect something as just by a simple test of something similar. And this is orders-of-magnitude more complex, more important and more deadly than a stupid games console.
You can't prove *mathematically* that the autonomous car will be perfect in absolutely every single situation (juste because there is a potential inifity of such situations). But you can prove *statistically* that the autonomous car is better and safer than a human driver based on the number of accidents and casuality caused by both. And overall this *will* mathematically increase the safety on roads.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
They can also drive safely millimetres (like inches but smaller) apart from each other, massively increasing the capacity of the existing road network.
Stopping distance doesn't change that much. The reaction time becomes smaller, but the braking time stays the same. It's fine for normal use, but when the car in front collides with an oncoming vehicle or something falling off bridge and comes to an abrupt stop then your driverless car still needs almost as long as a human-operated car to come to a safe stop without hitting the vehicle in front.
I am TheRaven on Soylent News
The drive-up ATM's at Citibank branches in the NY area have had Braille labels on all the buttons for years. Seemed kind of silly up until now, you certainly would not want a blind person standing in the driveway using the ATM, and I certainly hope a person requiring Braille labels on an ATM would not be behind the wheel of a car. Not knowing Braille myself, I always assumed the labels said "Get out of the way!!! You're standing in a driveway!!!".
But now I realize that Citibank was preparing for the eventual release of the autonomous car.
"We make our world significant by the courage of our questions and by the depth of our answers." Carl Sagan
Why would it need to tell the difference between a human and a non-human?
There are a few situations where this could be important. Consider a cat runs into the road on the right and a child runs into the road from the other side to get the cat out of the road. A human would typically prioritise not hitting the child. If the AI doesn't, and hits the child in preference to the cat then it's not going to look very good. If there's only one obstacle, you want to avoid it. If there are two, then you want to avoid the most valuable ones, and generally we consider humans to be more valuable than anything else you're likely to collide with.
I am TheRaven on Soylent News
As explained by other, the car *does* slow down, and even eventually halt when exposed to situation it thinks it can't handle.
Also, the car has much lower reaction times. So in some situations, it doen't really need to slow down, it will react immediatly if needed, whereas a human driver will need to slow down to make room for slower refelexes.
(The distance between autonomous vs human-driven cars on the motorway, for example).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
I think he's referring to the fact that a digital driver could have as many "eyes" in the form of cameras as it needs, arrayed in whatever way works best. It can have a 360 degree ring of cameras on the top of the car, for example, which has no blind spots at all. (I mean, unless you manage to crawl under the car via a sewer system or something....)
Compare that to a pair of forward facing eyes, with an elaborate system of mirrors to try and allow them to see behind the car as well as in front. Lots more blind spots, and they can only look in one direction at a time.
Additionally, once all cars have this system, it won't even need to allow for lag for the driver behind to react, it can broadcast a signal to brake and a whole row of cars can instantly come to a halt, so an end to one car braking and causing a massive pile up. I suppose the car doesn't care what the obstacle is that's suddenly appeared, an obstacle is an obstacle, but watching some of the videos, the system paints various obstacles in either yellow or red, pedestrians red. I assume this is some kind of risk assessment or prioritisation - i.e. if a child runs out in front of the car the system will prefer to scrape the parked car next to it than hit the person. To make this kind of judgement in sub second time in a 360 degree arc is exactly the reason why computers will make better drivers than humans in the future.
I think it may have something to do with cops legally able to supervise an unlicensed driver during a road test. Usually as part of receiving a driver's license. You should probably get that paranoia checked out, though.
The problem is this is just another nail in our coffins. Human are not the best chess players. Humans are not the best Jeopardy players and now humans are not the best drivers. The list is only going to grow until we find there is nothing we do better than computers. This technology will eliminate millions of jobs. It will even change buildings. What will happen when Wal-Mart finds out that it is cheaper to deliver goods than it is to build huge stores? How many square miles of land are now just parking lots that will not be needed anymore? I am glad that I am over 60 years of age because this and other computer advances will eliminate so many jobs that a person born today will find it extremely hard to find a job or a purpose for living.
When automated cars form convoys, they are in communication with each other. That means if something causes the first car to brake, the rest of the cars know instantly that the lead car has braked, and that they should brake. Insert "you go left I go right" logic (if needed), and these cars are now better than you at doing it. There is nothing human drivers can do that these cars can't, as their sensor packages are far better than ours, and they can communicate with other cars allowing driving formations humans simply can not handle. The logic they've put into these cars is simply staggering, and that's what's taken so long to develop.