"The only exception is when the lines do not exist, or are not visible, in which case the driver will have to take the old-fashioned route and drive with both hands."
That sounds like a severely limited AI.
I wonder if some of these half solutions won't actually be worse for safety. Is someone a great deal more likely to completely shift attention from the road if the car is driving? Maybe even by dozing off? When that happens, how long will it take to regain sufficient focus if the white lines disappear on a stretch of road? How robust is the AI for a car that could be fooled by deceptively-painted lines?
Google's car is much more sophisticated. Given what Google has demonstrated, I'd rather we skip any intervening generations of half-self-driving cars. These pseudo driving cars seem just as likely to give the whole AI vehicle concept a black eye and set the industry back 10 years.
I was just thinking the same thing -- "this will never work in Boston. We fear and hate painted lane markers."
But it turns out that something like 90% of miles traveled are on highways.[1] So what if you just limit it to limited-access highways recognized by the GPS? That covers a huge chunk of miles traveled, and you have actual vehicles out there a lot earlier, which will teach you a lot. And highways kill (I think) tens of thousands of people a year, so getting more reliable drivers out there can't come any too soon.
I'm also thinking that the AI will have to be pretty damn robust to go into production, whatever this article says. Whether or not there are painted lines, it will have to share the road with other cars, so it will have to be pretty smart in order to do anything at all.
No matter how limited the technology, it's good to have automotive companies clearing some of the regulatory hurdles, working on hardware integration, introducing this to consumers, etc.
The principle road tracking was to look for any sets of lines that moved in parallel. The primitive hardware was a series of horizontal line scan cameras aimed at increasing distances in front of the car and a simple sideways shift+correlation to decide which direction.
This way that tracked not just white lines, but tire tracks, kerbs, road wear, roadside barriers etc. The main problem we had was on a clean unmarked newly laid road surface that was perfectly uniform.
There is a highway I commute on daily which has been undergoing roadworks for the last 3years and has so many painted out lines, repairs, temporary markings, filled in trenches etc that in the rain I have no idea where the lanes are.
Made me think of those sections of freeway (I mostly notice this in SoCal, not sure if anyone is familiar w this) where there are grooves or gradations that sort of slowly drift off of the lane dividers on a curve. People often start to follow the false lanes at first. I imagine such a system would have trouble with that.
Who really believe Cadillac will release such car in 3 years? This news is missing the following statement:
This release contains "forward-looking-statements".
Forward-looking statements can generally be identified
by words such as "believe," "appears," "may," "could,"
"will," "estimate," "continue," "anticipate," "intend,"
"should," "plan," "expect," and other words and terms of
similar meaning in connection with any discussion of
projections, future performance or expectations, beliefs,
plans or objectives for future operations (including
statements of assumption underlying or relating to any of
the foregoing). Actual results may differ materially from
those reflected in these forward-looking statements
Have insurance companies figured out how accidents, tickets and liability are being handled?
If there is anything I've learned in my career it is that building systems that are 100% completely error-free is pie in the sky thinking even in moderately controlled environments. Suffice it to say, I will be a very late adopter although I think the efforts are very exciting.
Automated cars are one of those things that is slowly going to creep up on us all since it is a range of technologies working together, until one day we realize we didn't touch the wheel for a whole trip.
We don't really think about it like this, but things like anti-lock brakes and automatic transmissions are all on this same continuum towards "driverless" cars.
A next step might be something like adaptive cruise control: your CC is set at 60 but the car ahead of you slows rapidly and your car automatically applies the brakes.
> A next step might be something like adaptive cruise control: your CC is set at 60 but the car ahead of you slows rapidly and your car automatically applies the brakes.
It doesn't need to be 100% error-free, it just needs to be on average way better than human drivers. Even if there are occasional major catastrophes (e.g., 1000 cars drive off a collapsed bridge) it will still be better than what we have now: http://en.wikipedia.org/wiki/List_of_motor_vehicle_deaths_in...
From a statistical, macro perspective you are correct. But from the perspective of a driver this could be a hard sell.
I would say it is similar in essence to air travel. Even though commercial airline travel is way safer than driving, I'm pretty sure more people fear airline travel due to the element of having no control. Unfortunately when they release probably any automated car crash, as rare as it may be, will be played up as death by machine.
I think they will have to really play up the ability to easily override the automation (even though I doubt it will mitigate automated car crashes in that once you get used to it being successful you'll probably be reading a book when the override is actually needed)
I think they will have to really play up the ability to easily override the automation. I agree. Steamships retained masts for sails into the twentieth century, and I believe horseless carriages could still be hooked up to a team, for a while after their introduction.
Old person hits gas instead of break, crashes into robot car. What would a human do differently? teenager runs red light, gets tboned by robot car. What would a human do differently? Child runs in front of car. What would a human do differently?
It might take a few years, but thousands of hours of video of humans making avoidable mistakes will probably change people's minds.
That would be the case if people are rational. However, people are not rational, and don't like not being in control, even if the chances of them dying are far lower. It may be safer to share the road with self-driving cars, but it's still scarier to rely on a machine that you don't understand.
The most important barriers to adoption are social, not technological, and that means that even one major catastrophe can cause a wave of backlash, regardless of whether or not it's justified.
I agree that people are not rational and will fear the outliers and that the biggest barriers will be social. I'm hoping that there are very few glitches along the way, since we'll be safer if this happens. Most people feel confident about their own driving, but I think they also remember a lot of the close calls they've had.
I think—and hope—we'll be surprised at how quickly we acclimate to magic. Sure, we'll be completely freaked out by letting go of the steering wheel. Then we'll be irritated at how our car thought the leaves blowing across the road were an obstacle. Finally, we'll wonder how we ever survived our own driving and we'll mourn for those that didn't.
That would be true if we were all rational. But the chicken little behavior of us Americans over the past 11 years really demonstrates that we're not able to correctly evaluate and compare risks of minuscule magnitude.
If all the sensors involved in the self-driving aspects are recording (a la an airplane's black box), re-creating a given automobile crash will probably aid insurance claims more than hinder them. It will allow the parties involved to accurately place fault - be it failure of the AI, unexpected circumstances (tree falls down, non-AI car sideswipes, etc) and so forth.
The sad thing is that most cars have "black boxes" but are usually never used in cases for whatever legal reason. Hopefully that will reverse some over time.
But most of those black boxes only cover actions of the vehicle with said box installed.
With the amount of data required for self-driving cars, the entire situation could be recorded - things external to the car. It would be like the airplane black box, but for all the vehicles in close proximity too.
As the parent of two young boys (one of whom will be 16 in ~10 years) I can't begin to tell you how excited this makes me.
Ford Taurus introduced a 'parent key/child key' concept a few years ago that did things like limit stereo volume and top speed. This takes it one further.
No fretting about stupid accidents! No fretting about drunk driving!
"Sorry son, you can't ride with Johnny - he still has a manually controlled car."
But it leaves me deeply suspicious. Having some insight into what it took the team at Google to get theirs to work I find that unless GM starts with that team, and their patent portfolio, this will be be a longer term thing. I'd put money on a self driving car from someone before 2020 but 2015 seems a bit early.
In spite of Google's fudging the delicate bits (see my comment on the drive thru non-experience) they have done a tremendous amount of work that a car company would be hard pressed to duplicate quickly. That and Lidar or their functional equivalent systems are still darn expensive.
That being said, I think it is AWESOME that we've got a major car company putting this stake firmly in the ground. If nothing else it will motivate a response from the others.
I remember back 15 years ago or so when I was a kid visiting Chrysler's proving grounds testing facility there were some self-driving mini vans that had robotic drivers. They were driving (I think) on predefined paths and not on the open road.
I know nothing more about them but it is evidence that people at car companies have not been totally ignoring the concept.
"Having some insight into what it took the team at Google to get theirs to work I find that unless GM starts with that team, and their patent portfolio, this will be be a longer term thing."
You write that as if Google's team is the only one in the world working on this, or far ahead of all others. Their PR department may be superior, but I do not think they are the only ones; many car companies have been working on this for years, sometimes decades (http://en.wikipedia.org/wiki/Driverless_car#History). I cannot judge whether they are far ahead; please show data proving that if you know more.
http://en.wikipedia.org/wiki/DARPA_Grand_Challenge shows Stanford and Carnegie Mellon switching places between the 2005 and 2007 versions. Also, in both cases, I would say the top three are so close together that chance may have affected the rankings more than 'being ahead'. For example, if the course would have been slightly different (tighter corners, different road signage, whatever), would the teams have finished in the same order? Also, in both challenges, I would say the third-placed team had similar performance as the top two.
So no, I do not think that that shows Google is far ahead of the field. Or did Google hire both teams? Even that would not totally convince me., as it looks as if only US teams took part in those races.
> I'd put money on a self driving car from someone before 2020 but 2015 seems a bit early
2 Weeks ago, I would have agreed with you entirely. It seems unimaginable that GM could even have a working prototype in 3 years.
But last week I found out there's companies working on mining astroids for real.... so now i'm just going to sit back with a healthy skepticism, i have no idea what's possible in terms of years any more.
DARPA challengers have gotten really good at this, even when the road isn't that well marked. I think they can do it. I just hope there's a "sport mode" button to make it launch at its quickest 0-60, drift around corners and such.
They may be onto something. If we get all the bad drivers into these things. Maybe We'll finally have drivers that can drive and follow the rules of the road. Hooray! One concern though if these cars will watch for the lines on the road, how will they handle roads with no lines? I live in a place where apparently lines on the road are a new concept. And are non existent in a lot of places.
Question re. LIDAR (slightly OT to the specific article, but relevant to the general subject). How will LIDAR behave when there are multiple sources transmitting similar but overlapping scanning patterns across the terrain (like, say, a highway full of self-driving vehicles)?
I think the journalist made a mistake because he mentioned radar but not LIDAR. Far as I know no one has a serious self-driving vehicle without LIDAR, and I doubt that Cadillac is really proposing to change that.
As a motorcyclist, I really hope that all the upcoming variations of autonomous vehicles are heavily tested in all conceivable edge cases. Also, as a motorcyclist, I welcome the well-tested, never-tiring, unimpaired, highly predictable, self-driving AIs! :-)
With all manufacturers pursuing their own AI, though, I wonder if we'll have a future that echoes the past. If, say, Volvo has sub-optimal testing, perhaps 'ovlov' (an old Usenet joke - Volvo in your mirrors) will again become known as the bane of motorcyclists. Will Toyota launch ads bragging about their higher AI safety rating than Volvo?
As a bicyclist, I echo your hopes and concerns. I'll also add a hope that the most common car-bike accidents (right- and left-hooking) are tested for extensively, as they require a bit more math and foresight than simply recognizing that I'm there.
What I want is a stick or something that comes out near the bumper to block the path of bikes to protect my passengers. If they ignore it, they get deflected off into the sidewalk.
I try to pull as close as I can to block bikes from getting between me and the curb and still have idiots trying to do 20mph in an impossibly tiny gap.
Filtering is legal in many places, and municipal governments explicitly enable this by placing bike lanes on the curbside of auto traffic. So if you live in one of these municipalities, all you are accomplishing is impeding the flow of bicycle traffic, making it actively dangerous for bicycles to follow the law, and worst of all perpetuating the so-called "war" between bicycle and automobile traffic.
Disclaimer: I avoid the issue by travelling in the center of the auto traffic lane. If you don't like that, tough luck.
It wouldn't have to lock it. A simple warning LED located on the upper part of the door at eye-level that lights up and blinks when an oncoming cyclist from the rear is detected would prevent a lot of accidental injuries.
I think their issue was not having control over when they leave the vehicle. If the locks are computer controlled, there's risk of a false alarm preventing your exit, potentially in an emergency.
I agree that this is better left to a warning indicator.
Yep, that's exactly my issue. Warning light is fine, lockout is not. Too many situations in which a lockout would be dangerous - false positives/failure of the system locking me into the car, carjackers taking advantage of it by parking a bike behind me, etc.
I welcome the well-tested, never-tiring, unimpaired, highly predictable, self-driving AIs! :-)
No more drunk drivers, no more mothers dealing with their kids while driving, city buses whose drivers don't need breaks, rental cars that drive themselves from their lot to your house (they pick you up instead of you picking them up), 100% automated home food/parcel delivery, etc.
The problem, though, is that as soon as a self-driving car crashes and kills someone, people will declare self-driving cars to be evil and unwanted -- even if they're actually safer than human-driven ones.
I'm going to wager that, while not perfect, AI drivers may well be safer for motorcyclists than the average human driver, and probably quite a bit more predictable.
far far safer. Even the best driver sneezes. AI drivers, even if basically not great drivers, are uniform and predictable.
Also, i think they'll be heavily weighted towards "safety" and not "excitement". They might drive slower, and stop more often than strictly required, but not causing lawsuits must be priority 1
On the other hand a pair of bikers could start a new sport of Volvo-sniping.
Assuming the algorithm is something like - steer away from another vehicle that appears in your lane - your could 'herd' lines of Volvos into each other.
Ob biker joke - what's the 2nd most dangerous thing on the road? A Volvo driver in a hat.
What's the most dangerous ? A Volvo driver without a hat - because he is driving around trying to remember where he left it.
That sounds like a severely limited AI.
I wonder if some of these half solutions won't actually be worse for safety. Is someone a great deal more likely to completely shift attention from the road if the car is driving? Maybe even by dozing off? When that happens, how long will it take to regain sufficient focus if the white lines disappear on a stretch of road? How robust is the AI for a car that could be fooled by deceptively-painted lines?
Google's car is much more sophisticated. Given what Google has demonstrated, I'd rather we skip any intervening generations of half-self-driving cars. These pseudo driving cars seem just as likely to give the whole AI vehicle concept a black eye and set the industry back 10 years.