Slashdot Mirror
Hidden Obstacles For Google's Self-Driving Cars
Paul Fernhout writes: Lee Gomes at MIT's Technology Review wrote an article on the current limits of Google self-driving car technology: "Would you buy a self-driving car that couldn't drive itself in 99 percent of the country? Or that knew nearly nothing about parking, couldn't be taken out in snow or heavy rain, and would drive straight over a gaping pothole? If your answer is yes, then check out the Google Self-Driving Car, model year 2014. Google often leaves the impression that, as a Google executive once wrote, the cars can 'drive anywhere a car can legally drive.' However, that's true only if intricate preparations have been made beforehand, with the car's exact route, including driveways, extensively mapped. Data from multiple passes by a special sensor vehicle must later be pored over, meter by meter, by both computers and humans. It's vastly more effort than what's needed for Google Maps. ... Among other unsolved problems, Google has yet to drive in snow, and Urmson says safety concerns preclude testing during heavy rains. Nor has it tackled big, open parking lots or multilevel garages. ... Pedestrians are detected simply as moving, column-shaped blurs of pixels — meaning, Urmson agrees, that the car wouldn't be able to spot a police officer at the side of the road frantically waving for traffic to stop."
Paul continues, 'A deeper issue I wrote about in 2001 is whether such software and data will be FOSS or proprietary? As I wrote there: "We are about to see the emergence of companies licensing that publicly funded software and selling modified versions of such software as proprietary products. There will eventually be hundreds or thousands of paid automotive software engineers working on such software no matter how it is funded, because there will be great value in having such self-driving vehicles given the result of America's horrendous urban planning policies leaving the car as generally the most efficient means of transport in the suburb. The question is, will the results of the work be open for inspection and contribution by the public? Essentially, will those engineers and their employers be "owners" of the software, or will they instead be "stewards" of a larger free and open community development process?"'
I will only buy a Google pod or whatever they're going to call it when it can safely and legally get me home from a night of alcoholic excess.
my blog of work misery - http://beastofbaystreet.com
The Google car has to be shown how to get to the garage, on your property, behind your house.
But seriously, if they'd known the way already, some people would have a heart attack.
GoogleCar: Please select the destination:
A. Before the garage where you cook your meth?
B. Before the garage where you distill your moonshine?
C. Before the garage where you grow your weed?
It's almost as if it's a research model and not intended for mass-market use. It's the car equivalent of an early alpha, get a grip. If you RTFA, they're not even selling it yet.
I just played with one of these at the California Academy of Sciences, and waving at it was one of the things I did to see whether the visual representation of the lidar's output was real. It had no problem detecting that I was waving, or the movement of individuals in the crowd around me.
No, homez, this isn't anywhere near "early alpha" analogy. This is like saying you're well on your way to producing a written a web server, when in fact what you've built is something which can deliver a single web page to a single client at once, and requires editing of configuration files to deliver another page.
The Not-Google-But-They-Slapped-Their-Brand-Name-On-It Self Driving Car is more a self-driving tram (where the rails are clear road markers) with basic collision avoidance. I will not be looking forward to when they're licensed to run in the UK. Although to be honest if I saw one I'd probably step out in front of it, because the earlier this dangerous project is aborted the better.
I have worked 20 years for a major auto OEM. Every time this site runs a Google car article (and there are too many) I cringe.
The first autonomous vehicles will only operate on controlled access expressways, and upon exiting there will be areas where the driver will have to take over or the vehicles will stop.
It will be decades before these vehicles can handle real life situations. You will need AI that can improvise as well as a human. Good luck with that.
I know a LOT of people that do drive over them.
So, I see no problem with that.
"nearly nothing about parking"??? There are PEOPLE that can't park - just look at any shopping parking lot and you will find a lot of vehicles that aren't parked between the lines... taking up two or more parking spaces. And yet, some of the advertisements for cars are now for "self parking" ability so that the driver doesn't have to.
No problem there either.
As for the rest... DARPA has already given contests (which have been won) about driving without a road map.
No problem there either.
Where I live girls get drunk, too.
The technology is in it's infancy stages. Why the media keeps hounding Google on all these issues seems immature. I don't see any other competing company attempting to do the same thing, and if there is, they are definitely staying clear of the media spotlight.
I see Google making some great progress in this area, but give it time people - they will work out the kinks, but it won't be done in year.. lets realistically say that maybe in 5-10 years from now we might fathom the idea that the car is safe enough for whatever weather and situations we can throw at it.
To get truly self-driving cars there needs to be a lot more tech in place. We need to develop a Roadmaster computer system that 'controls' all the cars by feeding them road parameters. WE need vehicle to vehicle(V2V) comms as well. The idea that self-driving cars are going to happen without beacons and roadmaster computers and V2V is silly. All the preparations that go into the google car to make it work needs to be baked into the road system to make it work en masse. We have long way to go, and google's car is a tiny baby step. In no way should be it considered ANYTHING but a pure research platform.
Good-bye
If I get drunk, and my Google car drives me home and overrun a red light or something like that. Who the cops should blame? Who will get the ticket?
Lisias@Earth.SolarSystem.OrionArm.MilkyWay.Local.Virgo.Universe.org
They are the best kind of girls
Even if it can only self-drive on a routes I've already driven manually, it would still be extremely useful.
Judging by how badly TFA was written.
Got it. So the cars cannot handle changes in traffic markers.
So they cannot deal with new stop LIGHTS but they can deal with new stop SIGNS. WTF?
And it would be "unmapped" for the first attempt. Right? Because the cars should be sending back data on road conditions and such to HQ. Right?
And the car needs the map to drive, right?
So they just drove over the same "few thousand miles of roadway" again and again and again and again? Until they got to 700,000 miles?
As it should. Because you don't know if that piece of paper is covering a rock or a pothole or whatever.
Isn't that one of the easier problems? The car waits until it detects a gap of X size where X is dependent upon the speed of oncoming vehicles and the distance it needs to cross PLUS a pre-set "safety margin".
Google often leaves the impression that, as a Google executive once wrote, the cars can “drive anywhere a car can legally drive.”
Wow, that's just a big ol' lie isn't it?
Brave Sir Robin ran away. ("No!") Bravely ran away away. ("I didn't!")
Bitch, whine, moan. Autonomous cars are a work in progress. Didn't anyone think they weren't?
Would you buy a self-driving car that couldn't drive itself in 99 percent of the country?
Of course not. But then, no-one's selling them yet, because they're STILL DEVELOPING THEM.
systemd is Roko's Basilisk.
Simple. The owner of the car will get the ticket.
Google's autonomous car technology is in its infancy. There have been other company's doing this for many many years with more advanced code / refined. The company I used to work for was able to detect negative obstacles (holes) water and humans. Their technology also was able to operate without a predefined map unlike Google's car.
In areas with fast growth detailed maps of every inch of the roads will require an accumulation of data almost hourly. For a specific map to be updated hour by hour may offer a billing opportunity for a major data service. Assume your car drives you to work at 8 am.. When it is time to return home the computer will need to know if any alterations or conditions have occurred during the day. Since most of us repeat patterns in daily driving those routes may update themselves while the vehicle is at rest. But any variation in routes would call for some sort of advanced route request so that one could get in the car and be transported without delay. Many sales organisations have craved a new street or new dwelling directory for decades and it has never happened yet. I live in a region in which new roads and homes crop up daily and it often takes maps months before those new streets appear. One can get a list of new building permits from the city or county but there is no map to guide one and the new places may be months from completion. The sales community would love daily mapping of their region.
I'm going to map my drive to work, by driving it a few dozen times. Then the car can take over. I don't care if it's no good in parking garages or my own driveway. I'll spend 3 minutes driving from my house, let the car take over, let the car do the boring freeway driving, and it can alert me when I'm 3 minutes from work. Then I'll take over and get into the parking garage and park my car.
Are we really whining because a brand new technology can't do EVERYTHING for us? Because it only takes care of MOST of the drudgery?
and I'll 3D print my self-driving car at home.
"Would you buy a self-driving car that couldn't drive itself in 99 percent of the country? Or that knew nearly nothing about parking, couldn't be taken out in snow or heavy rain, and would drive straight over a gaping pothole?"
Google is just trying to accurately mimic the real driving population.
How's it going to park in a dirt lot? Recognize which spots are likely to be too muddy? How's it going to park behind the barn?
Don't some of the breathless stories talk about whether or not these cars will need steering wheels? What's going to happen when the steering wheel-less models meet something they can't deal with? Blue screen of death?
just take uber
Google often leaves the impression that, as a Google executive once wrote, the cars can 'drive anywhere a car can legally drive.' However, that's true only if intricate preparations have been made beforehand, with the car's exact route, including driveways, extensively mapped.
Sounds like a bunch of FUD.
I can go to my provincial (Canada) government website and download the entire traffic map, with exact distances (vector drawn to scale) and including every entry to every driveway, be that city or rural. Where do you think Google get's its map data?
Avoiding potholes, sinkholes, obstructions, and other cars is part of the algorithm. If you mean the car will not swerve madly into the other lane, potentially creating a dangerous situation but drive over a hole in the road, well, good!! If it also does not drive over barricades, that's good too.
So yes, bunch of FUD. If you need car to park itself, there is already technology out there TODAY on the roads.
https://www.youtube.com/watch?...
It's 3 years old already.
You want a car that detects if road surface is more slippery and adjusts traction control and/or warns you? Out there already.
If a car can drive itself 90% of the way, and it requires me to park on a parking lot, what's the big deal? I don't want to drive during the tedious part of the route, where most of the fatal crashes happen because of DUIs or distracted or speeding/dangerous drivers.
So, we are just as far from self-driving cars as we were 50 years ago.
And a DUI for not overriding the car in time to prevent running the red light.
And that is if you are the ONLY person with a robot car on that road. Which may be correct for the initial roll-out. But this is a great example of the "network effect". If 100 people in your state own robot cars then a LOT of your state will be continuously mapped / re-mapped / re-re-mapped / etc.
There is space to be filled and page hits to be collected. Demanding instant perfection for every edge-case is a good way of doing both.
Google has logged over 700,000 miles in those vehicles. Without a single robot-controlled accident.
There might be problems in certain weather conditions. Or in certain other conditions. Or whatever. In which case the driver should take over.
And since it is software, eventually those problems should be solved.
I agree (article submitter here). I submitted the article mostly not to complain about lack of progress but because the article covered a lot of interesting details about how the Google technology worked in discussing the limits of the current system. I have little doubt such systems will continue to rapidly improve.
I was involved briefly on a project for self-driving cars in the late 1980s at Princeton involving neural network ideas for image processing, and I suggested we could just train the cars to drive specific routes. However, that suggestion was scoffed at (and I did not try hard to push it). My argument was that most driving is stuff like daily commutes or runs to well known stores, and so pretty much the car could drive exactly the same way every time, seeing the exact same sights. That might make it feasible to train the neural networks from just a few video recordings of drives over the same stretch of roadway. Granted, lighting conditions, weather, other cars, pedestrians, and possible lane changes make that harder -- but is seemed like a good place to start, rather than try to create a car driving system that could drive in arbitrary new circumstances where it has never seen the road before. Solar panels have succeeded much the same way -- the early ones were niche (like in calculators or satellites), but sales drove more R&D that lead to better and cheaper panels in more and more applications. A self-driving car that could only drive me from home to a few local towns and back on fixed routes (safely, while, say, I surfed the web) would still be tremendously valuable to me. Think of how many people commute the same routes every day for years and could use that commuting time more productively in other ways via the internet. If people with an hour commute could use that time to answer email, then maybe they could work one hour less in the office? Also, a car that just knew how to park itself in a standard location and come back to pick you up in front of some building you work at or apartment you live in would be very useful in cities.
Another idea I had several years ago is that we could have an open source software effort to drive cars in various simulated racing games like "Gran Turismo" or other free play driving games like "Driver: San Francisco" or various off-road sims. That would be a inexpensive and safe challenge for college students. Those driving simulators go to great lengths to make realistic looking images (including things like dust clouds and vehicle dynamics), and they continue to improve. You just feed the first-person video generated the game into the car-driving visual processing algorithms, and you have the software control the game via USB outputs. As the software gets better, then you can fuzz up the image more and more by adding more white noise to it, or whatever other distortions you wanted (like bug white blotches over parts of the image) to challenge the algorithms. Or you could introduce delays and noise in how commands for steering were processed. Such an approach makes writing such software feasible for the average software developer without a special car. Granted, the software would have to focus on processing 2D images instead of 3D laser ranging data. Even Google has talked about testing their software in simulations regarding certification. Ideally, the simulations used for testing would be open source too, like Rigs of Rods (or even more realistic) and if so, things like 3D ranging data could probably be extracted too: http://www.rigsofrods.com/
A 21st century issue: the irony of technologies of abundance in the hands of those still thinking in terms of scarcity.
and the DUI
"Isn't that one of the easier problems? The car waits until it detects a gap of X size where X is dependent upon the speed of oncoming vehicles and the distance it needs to cross PLUS a pre-set "safety margin"."
You've just excluded a huge part of the world. Where I live the right of way is not a right but something you have to fight to take. Google car would be waiting for years trying to make the simplest turn or merge.
They'll have a nice big BETA on the sides and the public will be very understanding of little bugs here and there.
Of course manufacturers will need a little bit of time to integrate their value-added enhancements so you may want to wait for the Nexus cars for trouble-free firmware updates. Or if your model can't be updated simply get a new car every 2 years.
Cwm, fjord-bank glyphs vext quiz
Look this is a question of cost. Self driving car must be able to detect the end of road or a hole or whatever so if google did not already install a radar or similar to check the road status ahead, it would be incredibly dangerous. And a radar or similar detection device would be far better at detecting :
1) pothole
2) slowdown speedbump
3) whatever the state of the road
better than human say, at night. Or in the fog. Or distracted.
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
When you look at the problems they have yet to solve, compared to the problems they've already solved, they don't look that menacing. To me it looks like a prototype that has been fantastically successful.
that the car wouldn't be able to spot a police officer at the side of the road frantically waving for traffic to stop
Well, can't solve that problem so lets hang up the entire concept of self-driving cars because of a handful of hypothetical obstacles. Never mind the lives and money saved, never mind the productivity salvaged by all that extra time. Can't see a cop waving so hang it up. No progress for you!
That's our life, the big wheel of shit. - The Fat Man, Blue Tango Salvage
Google isn't detecting potholes? Back in 1985, we had that on our DARPA Grand Challenge vehicle. The LIDAR on top of the vehicle was generating a ground profile. This was for off-road driving, where that's essential. I'd assumed Google was doing that; they have a Velodyne laser scanner that provides enough information.
In traffic, sometimes you can't see a pothole because it's obscured by a vehicle ahead, but if the vehicle ahead doesn't change speed, direction, or attitude, it's probably safe to proceed over the ground it just covered. On high speed roads, you can't see distant potholes clearly because the angle is unfavorable, but if the road ahead looks like the near road, and the near road profiles OK with the LIDAR, the far road is probably good. That's what the Stanford team used to out-drive their LIDAR range. (We didn't do that and were limited to 17MPH).
Fixed road components should be handleable. People, bicycles, and animals are tough.
There really is not going to be a need for an individual to own a Google car. You will have an app on your phone that you will schedule a car for when you leave, and if you know when you are leaving, schedule that as well. Otherwise, when you have a better idea to return home or go to the next place, you then schedule that with your app. No maintenance to be involved in, no insurance to carry on it. There won't even be a need for parking, no need to even worry about handicapped parking, you will be left off at the place you indicate (and perhaps busy places will have to be modified for having people just dropped off at the door instead of walking through the parking lot), and it will go off and pick up it's next fare, or decide to go back to the garage to be cleaned/maintained/fueled. It will be the Uber/Lyft/cab company nightmare, because no drivers will be involved, just companies with their fleets of autodriving cars. The average individual will not own a car anymore, except perhaps for far rural areas where it would be impractical to be running a fleet of autodriving cars, and those that need a specialized vehicle like a pickup truck/towing vehicle for a trailer.
A selfdriving car changes the paradigm, at least for a large swath of Americans. It also won't change for some, either.
Bryan
I ride a bike to work and I indicate for right and left turns as well as stopping under some conditions. I doubt that the systems on a (semi) automatic car will understand that.
http://michaelsmith.id.au
by the third paragraph.
"Of course, Google isn’t yet selling its now-famous robotic vehicle and has said that its technology will be thoroughly tested before it ever does. But the car clearly isn’t ready yet, as evidenced by the list of things it can’t currently do—volunteered by Chris Urmson, director of the Google car team."
So the author is complaining the car isn't ready to be used on the streets of today because.....the car isn't ready yet to be used on the streets of today.
Don't forget that a lot of beat-groovin' texting and drunk drivers wouldn't correctly identify that police officer either. BTW in 500,000 miles of driving I've never seen one frantically waving me to stop. So the question is, are the Google cars good enough, not can they handle some far-fetched scenario. The longer people drive familiar same-old same-old routes, the less attention they pay; OTOH unfamiliar roads are dangerous too. Google cars have a chance of solving at least some of these *non-far-fetched* scenarios. Better doesn't have to be perfect.
I'll guess allergies, and building those little straw men triggered them.
-The wise argue that there are few absolutes, the fool argues that there are no probabilities.
It doesn't matter if the cars are 99% accurate. A software glitch causing the car to run over a pedestrian in a crosswalk will cost Google millions. A fully automated car HAS to be 100% fool proof or the manufacturer is just stupid in our litigious society.
I have been saying this for years. We are nowhere near the human-level AI that would be required for an actual self-driving car - there are an infinite number of scenarios that are impossible for any modern computer to handle - and when we do achieve that true AI, in decades or maybe hundreds of years, a self-driving car will be one of the least important possibilities. I will utterly stand by my prediction right up until I see a self-driving car in any real-world variable conditions. I will admit, I do not understand why Google is pretending they can achieve this; they should know better than anyone all the impossibilities.
Put another way, if autonomous cars started off working on 0% of roads and you want them to eventually work on 100% of roads, well somewhere in between you have to pass through 1%, 5%, 10%, 25%, 50%, 75%, and 90%. It's rather disingenuous to criticize them for not getting all the way to 100% in one fell swoop.
So who decides which 50% (or whatever) of the road is suitable for auto control? My daily drive takes me through easy bits and much trickier bits, sometimes changing within 100 yards and back again, and how tricky depends on what other drivers are doing. What will the "50% competent" car do? Will it be saying :-
"Quick Dave, it's tricky, take over ! - It's Ok now Dave, let go the steering wheel - Oh hang on it's tricky again! - Now it's OK again - Oh jeez, some idiot's just pulled out in front I can't cope, LOOK OUT !! where the fuck are you Dave? TAKE THE FUCKING STEERING WHEEL DAVE !!! Aaarrrgghh !!! "
Sure you might get the stray deer hopping through traffic that requires a human to take control and improvise.
So the human must sit there with just as much attention, and with just as much skill as if he were driving anyway. So much for those hopes of drunks/non-drivers/blind people etc.
"[Chris] Urmson says these sorts of questions might be unresolved simply because engineers haven’t yet gotten to them."
Chris Urmson is the "director of the Google car team." He's guessing as to why his team hasn't solved problems yet, or if they've even attempted? It's like Google isn't even really taking this seriously.
Particle beam deflector cannons mounted behind the grill.
-- I ignore anonymous replies to my comments and postings.
The problem is not that YOU can overcome car limitations.
The problem is that OTHERS will not - all those lowest-common-denominator- people who will be told that this is a self-driving car will expect it to self-drive in all conditions.
If you don't think it's real, just read stories of how "GPS crashed a car" because some of the lowest-common-denominator drivers switched off their brains and just drove car into a river or off a cliff.
Just wait until a drunk sues car manufacturer because their car crashed or stopped in a middle of road instead of parking safely. Since those cars are going to first appear in USA, their usage will be dictated by famous USA lawyers based on experience of the least smart drivers, not those who understand what such a car can or cannot do.
On a pure technical front what Google does is a technical advancement, but it's not going to be a universal solution. Especially not when advertised as a car that drives itself.
Gather 'round kids, I'll tell you of the time when Google Self-Driving Cars were just around the corner. It was considered to be one of the 'last' frontiers in artificial intelligence, because AIs had already been tried -- and proved successful -- in other venues. A great many people, our smartest people, were concentrating on this and similar problems. AI was not a transformational experience.
In fact, this fixation with AI turned out to be a big mistake. You'd love to hear about some great struggle between man and machine, how we were brought low by our own cleverness. Here's the real why.
These AIs were not smarter or capable by any means. People just re-arranged their environment to make these machines comfortable, much as you would clean house and rearrange the furniture to better accommodate a handicapped guest. In manufacturing, specialized robots proved very adept for the most tedious and repetitive tasks of assembly but general object manipulation such as unpacking, but in sorting and placing parts the clumsiest of humans excelled. So the world became a place of conveyor belts and hoppers and jigs. Humans loaded the jigs, verified the proper operation of conveyors with a deft glance, and reigned supreme in the packing and shipping department. Once everything was in place and arrived at the proper moment, intelligent robots were able to construct incredible devices in seconds, where there once had been hundreds of steps spanning several days. So far so good.
If cleverer and more intricate devices were all it took to survive, the wheels of progress would still be turning.
No emergent intelligence, no revolution. The only machines that turned on their masters, turns out, were those specifically instructed to do so. Runaway killer drones suffering from software bugs, malfunctioning friend-or-foe systems, some hacked by dysfunctional or suicidal humans into becoming killing machines. But it was in the end quite impersonal, even boring. The machines did not seek to overrun Earth or join forces any more than a nail gun goes off in search of human wrists once its safety catch is removed. Except for that little skirmish where one million people were murdered in cold blood, but the machines doing it were too busy to notice the Corrective Forces moving in, they lacked the cleverness to hide or disguise themselves and... problem solved. A bit late for that One Million as they are called now but far less dangerous than say, a pandemic.
So Humans Need Not Apply . The automation of everything progressed. Clever humans do not need a reason to be clever because cleverness is its own reward -- something to do with endorphins and the "Ah Ha!" moment which we won't get into here -- we were clever. The most fascinating problems to solve were those which, when solved, put more people out of work. This happened gradually and mostly to other people, and no one shed a tear over it because it was easy to imagine liberated humans enjoying a life of leisure somewhere. A half-remembered snippet of an old film or Utopian anime was enough to do the trick.
So when it came to self-driving cars our best and brightest were right there. It is a sexy problem, successful negotiation of a task that even the dullest of humans seem to manage pretty well. And they were making progress, and predictably as in all these dystopian outcomes, the laziest among them would say, we are so close. But we'd be even closer if we could just get the humans off the roads.
They were screwed because there were more humans all over the roads then ever before, more than anyone could remember. These were people displaced by technology, the jobless seeking jobs, the homeless moving from place to place. While they had been busy making cell phones smaller and web apps more numerous, some of the 'true' engineers among them shaving
<blink>down the rabbit hole</blink>
When a human drives in a new area she must observer traffic signs, lights, markings, etc. She must become aware of all routes, access and egress with out help from some GPS system. Yeah it may be used to find a location, but beond that she must do the driving with out help of specialized directions. The self driving car (SDC) MUST do the same thing, Yeah one can provide the SDC with a set of regular traffic signs, lights, markings, etc, but it must be alble to handle accidents, construction, new and closing of old entrances, exits, etc. So knowing where every corner and driveway exists is over kill, it must think for it self and figure out the way with out human intervention.
I can't wait to attach a push-button deployable cardboard cutout of a person on the rear of my car to freak out these autocars into thinking a pedestrian suddenly appeared directly in front of them on the freeway at 70 mph.
the police officer is doing it wrong, he should run in circles to get recognized by the google car
Now can they teach cyclists to understand hand turn signals?
Simple. The owner of the car will get the ticket.
Interesting. A very nice and new way to get "pwned" by someone. :-)
"But sir, my car got hacked!!!" :-)
Lisias@Earth.SolarSystem.OrionArm.MilkyWay.Local.Virgo.Universe.org