The View from the Front Seat of the Google Self-Driving Car, Chapter 2
medium.com
medium.com
It's easy to play out the use cases and see why Uber is investing in this area and why it feels threatened by this technology advance.
[1] - https://www.yahoo.com/autos/s/uber-ceo-tesla-sell-half-milli...
http://techcrunch.com/2013/08/22/google-ventures-puts-258m-i...
Note how easy it'll be for Google to flip the "Uber" provider in Google Maps to their own service.
Longer term, a majority of people have self-driving cars. If I live in the suburbs, I either keep it all to myself, or loan it out when I'm not using it. If I live in the city, then I likely don't own a car - but instead call on a car as I need to. These probably come from both individual lending their car, and from groups/companies that are setup as a service.
Either way, these cars are likely to be network agnostic - meaning they'll lend themselves out to Uber, Lyft - whoever. So while costs will go down, it could lead to a big margin compression. Meanwhile, the arms seller (Google, whoever) makes money on each car being sold.
Makes sense, therefore, for Uber to continue to focus on capturing the market - their installed consumer base (and their brand) is their strongest asset going forward.
Many of us like the idea, I also talk about it to some people, surprisingly they have no knowledge of the SDV projects (Google or other), even driving instructors. So far it adds to the trust and legal issues, public awareness seem quite slow for such a thing.
How is the car going to park itself? The grandparent's point was about how annoying trying to find a parking space is.
Can self-driving cars parallel park yet?
Can they detect a strip where parking is only allowed between 9-4 and it's 4:30? Can they detect fire lanes?
EDIT:
Also, are these automated taxis really going to be more affordable than owning your own car? Taxis are incredibly expensive today. Sure you save a lot of money when you don't have to pay a driver, but is it really enough to make car ownership undesirable?
Non-completely-self-driving cars can do that:
https://www.youtube.com/watch?v=u-rxJkVzUxI
:)
> Can they detect a strip where parking is only allowed between 9-4 and it's 4:30? Can they detect fire lanes?
I don't have specific examples, but I know that it's possible to read signs easily, (as in you can write a program to do it yourself with no great techno-wizardry,) and I don't see any reason they couldn't use that data.
The logistics of parking go far beyond getting into the space. How does it pay the parking meter? How does it know if a parking lot is public or not, how does it pay for that? How does it know not to go into the unsafe neighborhood where wheels get stolen?
But realistically they are problems that won't be had, because in the near future (and probably in the far future as well) most people aren't going to buy their own personal self driving car because not only are they expensive, but it is pointless to have a personal car when it is way more convenient to offload responsibility for the car ownership to someone else.
Instead they will use an app to summon an automated taxi car owned by a corporate entity (Google, Uber, etc), and when they are at their destination that car isn't going to park, it is going to be off to pick up another passenger.
The car owner will have designated secure parking lots for overflow cars that aren't needed out on the roads during non peak hours. The cars would return to that designated lot when not needed.
Almost any initial solution is going to be imperfect, and is going to involve mistakes if you try to solve for the entire problem space. Usually, the way to address this is to solve a small subset of the problem. If you don't know how to pay for the parking in a particular place, don't park there. Things like that. You limit the problems that you have to address, and the lessons you learn while applying that technology and the infrastructure that develops around its use alters the difficulty of refining it to address other cases.
I can easily imagine the way that I would solve the parking problem, at least initially: Set it up so that the car can pay wirelessly and make it so that people who own parking spaces can add their carpark to a database as enabled for wireless payment. There are even systems which already exist that do something similar to this; toll roads, toll bridges, et cetera: you put a little device on the front of your car and that records your usage of the relevant infrastructure, which is then charged to your account.
Put a bunch of image recognition for parking signs in the car - good parking space usually follow fairly straightforward formats; I'm not sure whether the signage is defined in statute or regulation, but generally they're much alike. Do some simple correlations on whether the person actually indicates they want to park there to refine the algorithm....
Crime figures are publicly available, so you can hook into that database for avoiding bad areas...
And have a rule that says not to park somewhere if you're unsure.
#
In the case of taxis, it's even easier: there are designated taxi ranks where they have to wait. And, considering there commercial enterprise, (there are two types of taxi regulation here, it's a bit weird,) you could rely on the taxi company to fill out a database themselves of other places that the car could wait - it's in their financial interest to do so.
This seems to be a point that is often missed. In denser cities--where utilization can be relatively high--drivers make maybe $15 per hour. (Maybe a bit higher when actually carrying passengers which is probably the relevant metric.) So it's not clear to me how these hypothetical vehicles are going to so revolutionize the way people get around by cutting costs by maybe $15/hour (ignoring other labor costs associated with cleaning or any incremental cost of the vehicle).
I don't know that the traffic would be much worse, because if you assume that a significant number of cars are self driving, then you eliminate many of the problems in terms of traffic flow that come from a human driver: maintaining a constant speed and distance from the car in front, aggressive driving, counter-productive driving habits, tragedy of the commons issues. It is possible to use the roads far more efficiently than we currently are capable of as humans.
There's also the point that it doesn't really matter how bad the traffic on the roads is if the only majority of the things suffering that bad traffic don't have any people in them.
Gas and polution:
If you can tell your car to go somewhere else when you are not in it, then electric cars and out of city parking lots - things like that - become a lot more feasible. You no longer need the infrastructure to be right next to your office to make effective use of it.
Until his town succumbed to the taxi lobby and banned ridesharing, he was able to run errands, get groceries, and take care of himself. Now, he is counting down the days in captivity, waiting until he can once again be self-sufficient thanks to autonomous vehicles. Keep up the great work, Google et al, and let's hope that day comes soon :)
I absolutely hate driving - a complete waste of 30+ minutes of my day, every day, and stressful too dealing with all the idiots on the road.
OK, podcasts make live bearable, but I would love to do way more during this time.
I imagine having a 4G network and being able to use this time for working remotely.
What a nice thought - being home 1 hour early and quite relaxed...
tl;dr: most accidents with Google cars were caused by distracted human drivers. Drive safely! Watch the road!
In the video, there appears to be a shoulder with plenty of escape room.
Is there any attempt made to un-distract the driver? Flashing brake lights? Horn?
The June report from Google (http://www.google.com/selfdrivingcar/reports/) contains some interesting language:
"The Google AV had been stopped for about 11 seconds at the time of the impact"
The cynic in me thinks they are intentionally avoiding technology to address these types of accidents so that they can unequivocally prove the the AV had no fault (i.e., had been stopped for x seconds).
This sounds too similar to victim blaming. In my opinion any action taken by someone who is about to be hit from behind in a low speed crash is likely to make the problem worse. I don't think that is the time to have a quick jerky motion into the shoulder of the road. Even honking the horn could distract or panic another driver that then causes a crash.
To me the solution is to eliminate distracted driving.
It knows the distance to the car in front and has a good idea of the rear car's speed. The only real variable is the weight of the approaching car but a pessimistic estimate could leave more margin to avoid touching the car in front.
When stopped in a place other drivers might not expect, just flashing the brake lights does wonders. Probably because human eyes and brains notice change much more readily than a solid light.
You should expect drivers to stop at intersections. You should normally leave enough space between you and the car in front to avoid rear-end collision, even if that car stops suddenly. And there's already a signal the car in front has slowed - their brake lights come on. (Maybe that's not a US thing though?)
The second point is completely off-topic, as the car in question had plenty of space between them and the Google car's rear-end. It just failed to notice it was not moving.
The last point is already addressed in my previous comment: a solid brake light is much less noticeable than a brake light coming on. In some countries in Europe it's relatively common to have blinking brake lights, but I don't think that's the case in the US.
These points are just basic human behaviour, and they aren't going away, no matter how contemptuous you are and how much you look down on other people.
Watch the video. The self-driving car braked because the two cars in front of it did.
The car behind the self-driving car didn't stop though, which lead to the incident. Presumably because it didn't pay attention to the fact that the cars in front were not moving despite the green light.
I'm not sure I understand where this thread is leading.
https://www.gov.uk/government/uploads/system/uploads/attachm...
> GREEN means you may go on if the way is clear. Take special care if you intend to turn left or right and give way to pedestrians who are crossing"
You seem to be saying that people are shitty drivers and that we need to either just put up with it, or fit our cars to stop some careless inattentive idiot from driving into us.
Fuck that.
Bad drivers should learn to drive.
Lots of non-shitty drivers do not follow every road rule to the letter under every circumstance and, indeed, it's not always possible to especially without pissing off everyone around you.
I actually agree with the basic point that it's the following driver who has 99% of the responsibility not to run into the car ahead. That said, there are plenty of behaviors like coming to unnecessarily abrupt stops in unexpected places (which I'm not saying is the case here) that will absolutely lead to people running into you.
Roads signs are like compiler warnings, it's all cute and nice, but it's just characters on the screen anyone can physically ignore. Let's have roads with nasty speed bumps before all intersections, which are all annoying, impossible to cross fast, roundabouts. #haskell
This is bad driving. You use the brake lights (and if you haven't seen the break lights working, you use the visual speed til you confirm the break lights are working) to determine of a car is going to stop. You should never put yourself in a position where any other factors are involved in if the car in front of you will stop or not.
> a green light is a strong signal that makes people assume the other drivers are moving or are going to.
And such drivers are to be held responsible for crashes and should be educated when found. Those who do not change should lose the privilege of driving.
er:
> The lane to the left of the Google AV was a left-turn-only lane. The vehicle waiting immediately behind the Google AV in the straight-only lane began to move forward when the green arrow left turn signal appeared (despite the signal for the straight-only lane remaining red) and collided with the rear bumper of the Google AV.
Are you talking about a different incident?
But that doesn't mean that Google is somehow off the hook. Google could invent a car that never causes crashes, but yet is still more dangerous than a human driver based on the inability to avoid crashes that humans easily avoid.
And, really it isn't hard to envision such a car being built. If it randomly hits the breaks to avoid phantom sensor readings, then the cars might be stopping in unexpected places an order of magnitude worse than a normal driver.
I'm sure driverless cars will eventually be safer than human, but I'm not convinced they will be.
I'd like to see the video (they have cameras) that goes with the LIDAR reconstruction.
The CMU/Cadillac/Uber self-driving effort doesn't have a LIDAR on a tower, so they don't have as much information. They haven't been reporting accident data, which would be interesting.
This made me curious; wouldn't it make more sense to get that kind of data from crashes reported to insurance companies? At least here in Italy, most minor-but-with-small-damages crashes are reported to insurance companies, but not to the police.
Note: I'm not commenting on the legality of such an approach, which varies by local law and the severity of the crash.
It means you are currently underinsured and maybe committing fraud.
It's part of almost every insurer's terms and conditions.
They use your driving history to set your premiums. Thus they want to know every accident you've been in because that affects how much they charge you - it affects how risky you are.
If you know you've had an accident, and they ask you to report all accidents, and you do not report an accident with the intent to keep your premiums down you've possibly committed fraud.
I'm not sure if it's the same way in America, but in South Africa insurance companies will at least ping the police for information if you make a claim - so drivers generally always get them involved if they plan to claim.
The reason why that wouldn't work here is because there are a few hidden costs if you make a claim: people frequently fix small bumps out of their own pockets. This means that not even insurance companies have decent stats on the type of accidents that Google is talking about.
Could be similar in America.
this is where self driving cars are limited and shoudlnt decide for us.
Deer vs little girl does not really matter, that is just an emotional ploy.
I guess once these cars begin being sold people can start testing situations using dummies and such.
That said, there are certainly circumstances where swerving is the right decision and a vehicle with 360 degree sensors is potentially in a better position to make an optimal decision than a human making a snap judgement based on mostly what they see in front of them. Probably the closest I ever came to being in a really serious accident happened on a highway when someone slammed on their brakes to avoid rear-ending the car in front, ended up drifting into the next lane where another car had to swerve to avoid them. I ended up swerving into the breakdown lane. Amazingly with cars swerving and braking all over the road at 70 mph, no one collided but it was a very near thing.
I strongly suspect that, however many decades away true self-driving is, good assistive driving and collision avoidance systems especially on highways are going to be a pretty big near-term win.
Don't let the perfect be the enemy of the good.
These hypotheticals don't really help the conversation. I don't know how the car would react but it would be nice to know if the car is going to try and work with neighboring cars to try and not hurt anyone. Not to mention that the car will already see the girl crossing the highway before she ever does get out there because of how sophisticated the sensors are.
Unless the AI behind these cars are so advanced that we cannot know ahead of time what it would do and it cannot be put in a simulation to test it, there is a vast difference between a self driving car and a human. Namely, we don't (and with the current level of knowledge can't) know what a human will do til after the situation occurs. And in general, in snap judgment cases like this, we do not punish a person for making a bad choice unless it is outright malicious.
Contrary to this, a self driving car is in a system who we can attempt to simulate (though not at perfect fidelity) and one where we know the underlying algorithms enough to answer the question and even change the algorithms to change the answer before the situation ever occurs.
>Not to mention that the car will already see the girl crossing the highway before she ever does get out there because of how sophisticated the sensors are.
This is attacking the situation instead of answering it. All I would have to do is modify the situation slightly and the question would remain just as valid and thus the response does nothing but postpone.
I managed to reduce the speed to about 40 mph before hitting it by which time it didn't do too much damage.
Your peripheral vision can see stuff coming from the side quite well.
It wouldn't take much — an unnecessary hesitation, a slow pull away or (on the other end of the spectrum) super-human reactions to cause me to rear-end someone.
Most of my near-collision experiences have been at a junction where I've judged a situation to be safe to pull into the road for both myself and the car in front, and I start to manoeuvre while doing my one last check into the oncoming traffic to verify its clear. My assumption is that the car in front will do the same but occasionally they don't and I have to stop suddenly to avoid collision. Every time I feel surprised and embarrassed. Why didn't I check? Truth is the other 400 times we all behaved as expected.
My point being, if the cars are still being limited in their speeds and are (as sometimes reported) a little ponderous in the middle of a manoeuvre maybe it is their behaviour that is causing some of these 'driver' errors.
I see autonomous cars as the solution to this problem.
If you rear end someone it is almost always[1] your fault, and in the examples you give it's definitely your fault.
I'm a little bit surprised that you describe this as the fault of the Google cars.
[1] ALMOST
I don't know if Google's cars do these things or not - but those are the kinds of avoidance techniques I would hope for, even if an accident would not have been their fault.