Imagine that for a moment: Self driving car companies could promise to pay all accident expenses provided it was on auto drive at the time of accident.
Not only is that a marketing win, potentially reducing accidents and saving costs for owners... if you jump forward many years of Google (or other) defending owners in court you would reach a point even judges would be biased to believe that the auto-driven car could not have caused an accident, further disadvantaging non-auto cars in the market.
Keep in mind that every autocar will have an extremely accurate recording of the accident that can be analyzed and played back during a hearing. Determining fault will be trivial in almost all cases.
Accidents, when non fatal, are such a waste of anyone's resources. And there are so many way to be cheated. I'm stunned no manufacturer included a logging device. Nowadays you have dash cams though.
Many new cars have black boxes that log what the vehicle was doing, and they can be pulled in case of an accident. It's one of the benefits of having everything be electronically controlled these days.
Happy to hear that. I don't know why it's not advertised more, it's reassuring and enforcing at the same time. Maybe people would flock toward log-free models out of a reflex to drive carelesly.
That's probably why it's not advertised. Because your car will tattle on you. I've often wondered about the black box in my car, if I were in an accident and I was going 71mph in a 70mph and they pulled by black box, would I be given a ticket even though I was not at fault?
Although I'm pretty sure only the manufacturer has access to this data (as part of their copyright on the car), so it would take a warrant for law enforcement to get it. I don't have proof of that, but I would guess that's why you don't hear about it more. It's only used when absolutely necessary.
Precisely. Widespread knowledge would lead to tampering.
I was reading a recent insurance case where the driver staged and accident, but the black box indicated what really happened, and the driver withdrew the claim when presented with the evidence.
Widespread data logging including dash cams is going to improve driver behaviour. Even some cyclists are wearing 'dashcams' and this will no doubt improve treatment of all cyclists, even if overall penetration is low.
Agreed. The issue when collecting data is that it's a rabbithole, you need to have a lot of data to avoid shallow judgement, and careful jurors too. Going at 71 (74,75) instead of 70 might turn against you, but if you have enough evidence that the other vehicle would have it you even at 70mph then the system will work.
Yeah. Austria has them completely banned, and in Germany you can use them,but if you have an accident then whatever is recorded is inadmissible in court. Also if you post the recording to youtube you have to blur out faces and registration numbers to not get in trouble.
I think that will be necessary. The selling point is that these cars drive better than people and get into far fewer accidents. As the manufacturer, you have to show you believe in that.
If the safety records meet goals, it's functionally free money for the insurance company regardless of insurance price. So one could predict a race to the bottom for insuring these machines.
How do you hear that? It's basically a guarantee, and companies generally don't warrant against things they know will happen. The reason you get a 1 year warranty on things is because they are engineered to not break down within one year. After that, the incidence of failure goes up.
"We will pay for any accidents caused" means "this will not cause accidents". Because no company would do that unless they were sure it wasn't going to happen. Just like when Google gives $100,000 for bugs in Chrome. They wouldn't say that unless they were sure there were no or relatively few bugs.
Another interpretation of the Google bug bounty is that fixing bugs of a certain kind is actually worth $100,000 to them (but you're right that it would be hard to sustain if they expected to make thousands of payouts at that level).
It's already common for automobile manufacturers to explicitly market the collision safety features of their cars. Many car commercials have collision test footage.
This is why we’ve programmed our cars to pause briefly after a light turns green before proceeding into the intersection
I programmed my brain to do this after learning to drive in Miami. I always look both ways before proceeding, just as I would before crossing as a pedestrian. When approaching an intersection, especially at night, I also try to look for what's coming from the side.
Intersections can be scary places
As a runner I find intersections terrifying and will typically cross instead a few hundred yards away. I'd rather jaywalk and have to deal with cars from only two directions. In my experience crosswalks are often the least safe place to cross.
It doesn't work, or so I've heard. Drivers in those cities learn to expect the safety margin of the delay, and statistically they become more willing to run a red light right after it changes.
Back in the day they tried the overlapping red thing in my city. Once everyone got used to the new way of doing things the time when both light were red became a sort of anarchy zone. Everyone assumed that everyone else would strictly observe the red so they didn't have to.
I find myself reminded of an article about a German town that stripped out their lights and signs and just had everyone drive by basic priority rules. Apparently this sharply reduced incidents.
Some time ago I drove through that town without knowing anything about it. I didn't realize that they had such a concept, it just felt really stressful, because your normal expectations and intuitions don't apply.
This basic principle also explains why roundabouts (traffic circles) are so much safer than traffic lights. They feel unsafe to drivers, so they slow down and pay attention, instead of blindly trusting the lights.
In the UK, the lights go [red] → [red+amber] → [green], so that drivers are ready to move as soon as the light turns green. People don't run red lights.
What completely infuriates me is pedestrian crossings that don't automatically allow pedestrians to cross when the lights are in a state that would allow it.
Are you asking why peds are required to press a button at some intersections for a walk signal?
If so, consider that some intersections may require a long time (~45 seconds) for a ped to cross. When no one is crossing, this may be sub-optimal and a shorter time for that specific light combination should be used.
Nobody should have built an intersection anywhere that big, then. But such behaviors are totally infuriating in places where there are many pedestrians, for example all of the stations on the SF Muni's T-Third line require passengers to push the button to cross the street after disembarking at center platforms. These crossings are not far (only half of a moderately wide street) but they don't have a pedestrian walk phase by default. If you haven't pushed the button you get to just stand there for another cycle. Example:
As you can see, the street-view car has captured a brilliant scene. Traffic is jammed. Many pedestrians are waiting to cross. The guy in the Mini is either going to run the light or stop in the crosswalk. The person in the white minivan has somehow managed to drive up the street car tracks. And for the topping, someone has blocked the bike lane with a portable sign.
I'm asking why, if the lights will cycle in such a way that pedestrians COULD cross, why they should not automatically allow pedestrians to cross. It's infuriating to miss a cycle as a pedestrian, and whom does it benefit? I'm not suggesting that there not be buttons which can force a cycle or accelerate a cycle. How would motorists feel if traffic lights defaulted to red?
My friend was put into a coma for 6 weeks, suffered a traumatic brain injury and spent a year and a half in rehabilitation because she drove through an intersection on a green light as soon as it went green. The worst part is that the driver that hit her got off scott-free because she suffered memory loss so her testimony was void, and apparently the truck driver that witnessed it wasn't enough.
Every time I hear about the amazingly low accident rates in current generation self-driving cars I'm reminded of this joke (unfortunately I can't remember the source):
Grandpa, what did you do before self-driving cars?
Well, we just drove them ourselves. We just needed a little practice and to take a test.
I've been thinking a bit about good places to adopt this early.
Lānaʻi might be a good candidate: absolute isolation from other road systems, a small population (3000 people), and a single owner (Larry Ellison) who's just the sort of filthy rich technophile who might be willing to replace an entire island's cars.
Singapore could be the first major city to make the move, due to its authoritarian government (hey, I don't like it, but it might make this easier), strict rules to keep the stock of vehicles relatively new, and relative isolation from other road systems (only two bridges to Malaysia).
I'm not entirely sure just how much Larry can do - it's really the 2% of the island that's state owned that includes the public roads.
Hawaii in general seems like a great place to start a lot of "infrastructure 2.0" (if I may carelessly make up a term) projects geographically, but the logistics are tough because so much of the population is desperately poor, there's already a lot of class/ethnic tension and to top it all off, local government is a mess.
While I don't doubt things will move in this direction, dates like this seem extremely optimistic. Just speaking of the US, there are something like 250 million cars on the roads, and we buy cars at the rate of about 12-16 million a year. While I can see fully autonomous cars becoming available in the next decade, it will be some time before all manufacturers have refined enough systems, and legalization is in place, let alone mandating. Then you have the slow process of manual cars cycling out and getting replaced by autonomous vehicles. Another question is for the low income people that subsist on sub $1000 used cars, does a junker autonomous car continue to function, or are the poor now having to devote a significantly larger portion of their income to transportation? Then what happens to motorcyclists? I can't see them functioning autonomously. In some countries that would be a huge hit given the ubiquity of small displacement motorcycles.
To be entirely honest, I don't expect to see the elimination of occupant driven motor vehicles in my lifetime.
I think the existing auto fleet can be overcome fairly easily, albeit by a company that enters the market in a position that's initially losing.
The model I see that will remove individually-owned, occupant driven vehicles will some variation of:
My company will purchase your auto and credit you some amount of haulage on my autocar network, and perhaps some cash. For the low income people using sub-$1,000 cars a token amount of credit will be sufficient. The real value for them will be the convenience and reliability of transport they get from my system. Imagine not losing your hard-won job because your "beater" car failed at an inopportune time. Imagine not having unexpected car expenses that make you choose between, say, food and transportation.
The higher-end market will be a tougher sell, but there are likely productivity, personal time, and environmental impact arguments that can be made there.
I'll sell your car (likely at a loss) in some area when I haven't achieved market penetration yet (or for salvage). Once you're bought-in to my network you might leave me for a competitor, but ideally I'll use regulation, contractual lock-in, and other tools to attempt to keep Customers tied to my product until I've recouped my initial investment.
I think it could work. I don't particularly like it, but I think it has at least a non-zero possibility of happening.
Where do parents fit in this utopia of yours? I've got 2 car seats and a pram and some toys in my car in the driveway. Personally owned transport makes more sense for people who have to move things other than themselves and a laptop (tools, babies, etc)
Utopia? Did you read the last sentence of my comment? I think a self-driving car future is a future with disempowered individuals in a neo-feudal transportation world. It sounds terrible to me, but I think it's likely to happen (at least in the US). There are practical and economic advantages in that future and some large existing economic players who can profit from it.
I'd tend agree with your comment in the short term. Moving my daughter's child safety car seat would be a pain, for sure. Longer-term I would expect to see modular standards emerge for child safety car seats. You lock your child's seat into the standardized base that's hiding under a seat cushion in the back of the car. There are already stroller / pram systems that incorporate the child safety car seat, so I could expect that market to boom, too.
Moving things besides yourself is something a lot of people in urban areas already do with public transit. With a private autocar service it's more likely that you'd get more space than on a subway or bus, so I'd think it would be even easier to move more than just yourself. I would expect a rental option to let you keep the car with you, as opposed to it returning to the fleet when you reach your destination.
In a world where autonomous car are effectively autonomous on ALL roads, that could work. But we are really far from here. How does autonomous car fare when the infrastructure sucks ?
I imagine this going one of two ways: either human-operated vehicles might be outlawed, or auto companies will apply market pressure to favor people not manually driving (steep, steep discounts in insurance for self-driving vehicles, since many states still require insurance but if the self-driving cars reach the safety goals they're aiming at, it's basically free money for insurers to insure them at any price).
If we can require all self-driving cars to take orders from a central authority, we can do some pretty amazing things:
- Most traffic problems are a result of human (over)caution and poor merging technique. A centrally coordinated system of self-driving cars are perfectly aware of each other's intentions and could run much smaller safety margins, as well as make lane/exit access perfectly equitable. The effective carrying capacity of highways would increase immensely, and traffic jams would be a thing of the past.
- 100% of cars on the same coordination system would also allow us to do away with most stoplights, as it could simply modulate speed so that intersections are always running in both directions, but cars never collide. (This could be overridden for pedestrian crossings, and wouldn't be much of a benefit in the city, but in places that are already pedestrian-hostile - i.e. most of Real America - it wouldn't make things much worse and pedestrians would be pretty rare.)
Designing such a system to use a centralized authority, as opposed to having a distributed architecture, sounds like an invitation for all kinds of chaos. Given how seductively "easy" centralized solutions seem to be I don't doubt things will go this way, with results varying from hilarious to frustrating to tragic. (I'm not saying that distributed architecture is a panacea, but it seems like a centralized system is both wholly disempowering to the individual and much more vulnerable to all sorts of corruption.)
Model it on Air Traffic Control perhaps? So that local traffic is handled by a local system, but longer distance traffic is handed off to a higher tier one?
A distributed system would let people behave in ways advantageous to themselves, to everyone else's detriment, by causing their safety systems to yield to the dangerous driver (i.e. what happens currently). If there's nothing stopping people from creating an "everyone needs to wait for me" situation, they will.
That's a problem that has to be worked around in a system with a centralized control architecture, too. Presumably there will be some safety systems acting autonomously in the car (latency and network reliability being what they are) and those systems will be gamed.
To boil my position down: I want to see an Internet-style architecture rather than a "Bell System" style architecture. I'd rather have "smarter cars" with "dumber roads" and as little mandated centralized control as possible.
In a decentralized system attestation of identity (likely tied to one or more existing civil methods) and attribution of actions observed to individual vehicles would definitely have to be a component. Any identity system cannot approach 100% reliability, but if every car is able to report to law enforcement (or private driver reputation-tracking networks) the observations it makes of the behavior of other drivers I would expect that "cheaters" would be visible as statistical outliers.
"The car to my left doesn't answer as a known transponder challenge/response in the vehicle reputation database I am querying. I am going to assign a lower priority to any non-emergency events it is reporting to me."
"Hey-- that car just started broadcasting a different transponder response. Something is really, really funny where. I'm going to alert cars in my proximity that something is up so that they can physically identify the car with their LIDAR / imaging / etc. and treat it cautiously. If enough funny-business occurs eventually law enforcement will be alerted and physical interdiction can occur."
"According to the vehicle database I'm subscribed to this car always traverses statistically similar routes 15% more quickly than other cars. Other cars in its proximity on these routes have their traversal slowed as compared to their traversal of the same routes when this care is not present. It sounds like funny business is at-play. I'm going to cast a vote for this car to be more likely to be subject to an inspection."
etc.
I'd prefer such a world to not involve the ability for totalitarian (or-- ahem-- certain republics who purport to be democratic in nature and are filled with power-mad bureaucrats conducting dragnet surveillance) to use these tools for public surveillance, but I'm resigned to that fate. Self-driving cars themselves aren't a threat to freedom, but the infrastructures of surveillance and control that will grow up around them almost certainly will be.
Freedom aside, whatever happens is going to end up being an information security nightmare. Too many disciplines who have never had any exposure to information security are going to intersect with this technology, and each of those intersections is going to be ripe with all kinds of interesting threats and failure modes. People will be inconvenienced. People will be robbed, physically assaulted, and violated. People will die. I have no illusions that criminals are unsophisticated and that these systems won't be attacked in every possible way.
I think it is extremely likely that hailing and renting a self driving will be much, much cheaper than owning, maintaining and insuring your own vehicle.
I have many friends who buy cars from the $500-$1000 range and then drive them for a few years, fixing everything themselves using replacement parts. The cars are usually 15-20 years old. The insurance is $100-$200 a year. If they drive daily,do you think anyone will rent them a brand new, technology packed car for what is effectively $20 a month? HN lives in this bubble when everyone can afford a new car, or at least a lease on a car, but if you move slightly out of the 1st world countries(and by that I mean good countries with free healthcare, free education, but still with developing economies), you will find that for a lot of people $200 is half of their monthly salary. And yet those people drive their cars too. They don't live in poverty by any means, it's just that the standard of living is different than in the West.
>do you think anyone will rent them a brand new, technology packed car for what is effectively $20 a month?
Self driving cars won't always be new though. The price of the technology will drop rapidly as they become mass produced. And there will always an economic advantage to renting vs owning. The cost of the car is effectively divided over several people instead of one.
Except it will never be cheaper than public transportation. If public transportation is available, they will take public transportation. If public transportation is not available, neither will be driverless taxis.
That could be true... assuming that the government is able to adopt innovations and cost saving measures across all of their public transit systems fast enough to keep up with price drops from private industry.
I suspect that will not be the case. While public transportation may catch up on price (if the political will exists to push it there), I do think there will be a period of time where hailing a self driving car will be cheaper than some or most public transit options.
Will there be a difference between a self driving 1-5 person machine and a self driving 5-1000 person machine?
A robotaxi can take you to the local robobus hub, which drives you to the local robotrain point, whereupon a Just In Time (un)scheduled train will leave, as all journeys can be coordinated.
I can't imagine that there won't be strong efforts made by any entrants into the individual autocar market not to hinder the penetration of new public transit systems, and to hobble existing systems.
Why not? If driverless taxis are cheap enough, even a small town or rural area could have some. Wheras public transportation is available only in very large cities.
The issues with public transports are the same as with biking in the US: It works better when cities are really dense. Suburbs destroy the efficiency of public transport, biking, walking, inline skating, and any other non-car transportation.
I'd be interested to see statistics from more rural areas where there isn't any robust (or, in some cases, any) public transit. In urban areas, where living without a car is a possibility, I don't doubt that the low income population doesn't drive. In more rural areas I suspect it's the opposite.
Public transportation is not cheap and/or available in many places in the world. Look at all the cheap beater cars (even in the US), those aren't going driverless anytime soon.
Your can downvote me for days but that won't make your statements correct. In the 2000 Bay Area Travel Survey, driving was a minority share of trips for people in the bottom income quintile. Walking, biking, transit, riding as passengers in cars, and "other" are 50% more popular than driving.
Jokes on you. My account cannot downvote people yet.
See, that's the thing with Hacker News. A lot of people here think Silicon Valley is the center of the world and everywhere is just like us, when the fact, we are living in a little bubble and a lot of the US operates very differently.
It doesn't matter how big the empty part of the United States is. The Bay Area is as populous as Sweden and therefore big enough to do things its own way. It may in fact be true that the poor person in Pecos County, Texas, drives around in his own car, it is also an irrelevant fact on a population-weighted basis.
This is getting downvoted because it sounds pretty flippant from a US point of view, but it's actually perfectly accurate for most of the world. In much of Africa, you're middle class if you have a bicycle, and only the rich own cars. Motorcycles are wildly popular in much of Asia for the same reason: they're a lot more affordable than cars, not just to own but to operate as well.
This will only be the case when all cars are self-driving.
Many of these articles do not give a comparison...so,11 per 1.7m miles sounds awesome, but how do you know what the current rate is?
The data I found had a passenger vehicle fatal crash per 100m miles as between 1.0 and 1.33, and a property damage only crash rate per 100m miles of 145, which relate to 0.01 fatalities per 1m miles, and 1.45 property only crashes per 1m miles.
The article is pretty clear that most of the light-damage, no-injury accidents don't get reported, and that all 11 accidents fell in that category. Half the point of the article is that this data is not well known.
Granted, taking those 1.45 and 55% figures, we might compare the 11 accidents to approximately 3.2 accident (reported and unreported) per million miles in passenger vehicles. Emphasis on "approximately".
That 11 accidents is per 1.7 mil miles so the number you would want to compare is 6.5 not 11.
That said, I don't think you can compare those numbers because it really isn't clear what that "55%" number is supposed to mean.
Doing my own back of the envelope:
~3 trillion miles driven in 2010
drivers involed in PDO (property damage only) crashes:
18.5 million
People injured in crashes in 2010:
4 million
(from http://www-nrd.nhtsa.dot.gov/pubs/812013.pdf page 143)
These numbers include estimates for unreported accidents
Expected crashes for self driving cars based on the above number
6.2 crashes per million miles
1.3 injuries per million miles
Actual number of crashes per million miles (11/1.7)
6.5 crashes per million miles
0 injuries per million miles
Of course, we have no idea if this comparison is at all valid. Given that the incidence of accidents is highly dependent on the ratio of highway miles to street miles, I doubt the comparison is at all accurate. I suspect that since the greatest challenge to self driving cars in on streets, self driving cars have traveled a higher ratio of street to highway miles than normal.
So it would seem purely based on these numbers that the Google cars have had about 4.5 times the mean rate of property-damage-only crashes. However, these statistics don't show us what the shape of the curve looks like, so we have no idea what percentile the Google cars are in.
If we take Google at their word that not one of the 11 crashes was caused by the self-driving system, then these numbers aren't evidence against the self-driving car being safer even with human drivers sharing the road. It's possible the car was driving unsafely in some way that increased its risk of accident without ever directly causing them, but we can't even guess at that without knowing the probability of their crash rate being this distance from the mean by pure chance.
So, do you really think the roads will only be safer when all cars are self-driving--you don't believe there will either be an incremental benefit in overall safety as more self-driving cars get on the road, or a specific safety benefit to owners of self-driving cars?
I want to hear the argument about people who want to drive.
I enjoy driving. I like driving cars. I still drive a manual and will not buy an automatic transmission. I participate in car clubs and I compete in autocross events. I enjoy automotive photography and I do my own maintenance on my vehicles.
What happens to people like me who actually enjoy driving and the act of owning a car? Do my loves get phased out because the majority of people just think of cars as appliances? Will this give birth to "classic-driving shows" just like we have classic-car shows and cruises?
Not saying I don't like innovation, I'd absolutely love to own a Tesla Roadster, but some things need to be preserved.
I want to hear the argument about people who want to ride.
I enjoy riding. I like riding horses. I still ride a horse and will not buy an automobile. I participate in riding clubs and I compete in horse racing events. I enjoy horse photography and I do my own hoof maintenance on my steeds.
What happens to people like me who actually enjoy riding and the act of owning a horse? Do my loves get phased out because the majority of people just think of transportation as appliances? Will this give birth to "classic-riding shows"?
Not saying I don't like innovation, I'd absolutely love to own a model-T, but some things need to be preserved.
Didn't think of the horse perspective. Thanks for that.
Still feels bad that some I love doing every day will be regulated to somewhere far removed from where I live and done only a few times a month at best.
I doubt it will be outlawed. You can still ride horses on public streets in most places (probably not freeways I imagine), but most people of course don't. In the same way you probably will be able to drive cars manually indefinitely, even while most will choose to use autonomous vehicles.
It won't be outlawed; it'll just be priced out of existence. When autos are good enough, insurance costs for manual driving will skyrocket. Only the very richest will be able to afford driving on roads.
That doesn't seem to follow. Insuring a manual driver wouldn't carry any more risk than it does today (on the contrary, there may even be fewer payouts due to other drivers being replaced with safer autonomous counterparts).
The wildcard is lawsuits. In theory, being one of the (say) .1% drivers on the road that is human ought to be safer than driving today, but if you do hit anybody or anything and cause any sort of significant distress you're going to be looking at a negligence lawsuit that could be wildly larger than anything we'd see today "because they shouldn't have been driving"/"because they should have been in a self-driving car".
But other wildcards could cancel that out. A law could be passed shielding humans from excessive payouts, in which case insurance could stay low. If you were demonstrated to be a safe driver by the automotive panopticon, you might still be able to get cheap insurance, although that one raises its own questions. And of course the ultimate wildcard is that they get sued out of existence entirely somehow, despite being provably safer at some point.
It's possible they don't really skyrocket. However, right now insurance rates are not at all cheap. It's not at all uncommon to have to pay $100+/month on insurance, per driver. For most families, there aren't many things they're willing to spend $1,000+/year on. When it's no longer a necessity, I would suspect that even if prices don't go up, because the marginal cost of driving yourself will have increased (cost to drive - cost of next best alternative, i.e. autonomous driving), most people will choose not to self-insure. Just like in some cities, the marginal cost of car ownership is so large (even if it's only a few thousand dollars more than in a suburb) that it makes more sense to rely on public transit.
This brings to mind the central premise of "Humans Need Not Apply", which I'm happy to bring up as worthwhile viewing whenever the discussion trends towards dramatic lifestyle changes from automation: https://www.youtube.com/watch?v=7Pq-S557XQU
Yes, but they are more or less considered a novelty than an actual efficient means of transportation. Most of the time, horses are not allowed on roads.
People still ride horses and we've had cars for over 100 years. I don't see driving going away, it just might be constrained in when and where you are allowed to drive(maybe like Autobahn in Chicago and Motorsports Ranch in Houston).
I think it's perfectly acceptable to expect that the enthusiasts will be the ones to preserve the things they love that fall out of favor. I enjoy driving on a nice empty road as well, but I don't think it's something that "need[s] to be preserved." It's something we have to do that we've learned to enjoy and do more than we need to. If it stops being something we have to do it's up to us to make time to do it for pleasure.
People love to fly but not everyone should be a pilot. Same goes with driving. The driving test could get more vigorous and you might not be able to drive on all roads but you will still be able to drive.
You mentioned cars as appliances, but you forgot the part where humans driving cars has cause of millions of deaths and many more injuries. I expect it will still be possible to drive cars manually, but in dedicated or isolated areas.
From a sentimental perspective, I hear you. I've also enjoyed driving, taking care of my car, etc. That said, the idea that some things need to be preserved isn't going to keep human-driven cars in the mainstream if the world wants to go in a different direction.
Take a look at newspapers -- I have no fewer than 5 relatives bemoaning the collapse of physical newspapers. Because nostalgia. It's decidedly more efficient to get your news electronically. How long before The Times discontinues home delivery?
Even if (when?) self-driving cars become the majority, I think you'll still see hobbyist avenues for enjoying cars -- with, my guess is, higher expense to you and a lot more regulation on use (e.g. not allowed to drive in cities?).
I have an issue with the efficiency argument. Far too often when things driven by efficiency reach a certain 'point', the density of effort put into it plummets and so does quality (food, music, ...). I'm biased toward old, non automated way of doing things, because people do care about the craft/art and that's where the value is. I'd run a mile to buy a newspaper if the journalists are worth it. Sure, these industry don't thrive like they used to when the underlying tech was the best or only one, but that's a hard fact of nature.
Most likely driving tests will be tightened and be made more difficult to pass. The main barrier to doin this right now is that so many people need to drive.
It doesn't need to be spelled out that crashes are not evenly distributed between the driving population, and taking the 20% worst drivers off the road is going to reduce crashes by 80%.
My insurance rates are low because of a very good driving history despite owning high powered cars. I don't see any reasons why that would change.
Manual driven cars will not disappear, what will happen is Insurance will go through the roof as more self-driving cars are added to the roads. There will be some roads declared for self-driven cars only and some that will allow both.
I see this being the most plausible thing to happen. Separate insurance classes for manual and automatic cars, the costs of each being supported by their perspective groups.
However that begs the question: by sheer numbers would that make insurance on a automatic car comparable to a manually driven one? If there are significantly less manual cars, there are less accidents to pay out. An accident in a manual car is not a guarantee just like no accidents in automatic cars is not a guarantee.
I think if manual insurance was more expensive it would be an artificial premium based on the fear of something happening.
> However that begs the question: by sheer numbers would that make insurance on a automatic car comparable to a manually driven one? If there are significantly less manual cars, there are less accidents to pay out.
But less to spread the cost over. Moreover, insurance rates aren't set at just enough to cover the average expected cost (which would result in, over any given time period, a 50% chance of the insurance company losing money) but enough above that that the insurance company is fairly certain that it will not lose money -- because of the law of large numbers, this will (for any given minimum probability of profitability) be closer to the expected cost when there are large numbers to spread the risk over, and farther (above) that expected cost when there are smaller numbers.
So, smaller numbers of manual cars will -- all other things being equal -- drive up the costs of insuring manual cars if they are a separate risk pool from automatic cars.
"""
While it may seem harsh that the 31st-century equivalent of "Driving Under the Influence" carries with it the death penalty, this is due to an inherent inequivalency between MOUI and DUI.
With DUI, you need only climb into your vehicle while under the influence of alchohol or drugs and attempt to drive it home.
With MOUI you must disable a number of safety systems designed to prevent idiots like you from manually operating their vehicles while inebriated, overtired, wasted, decaffeinated, angry, emotionally distraught, or suffering from hormonal disorders like PMS or testosterone poisoning (the latter having been positively identified as a leading cause of stupidity among males between the ages of puberty and death). After disabling the safety systems (which task almost certainly requires ice-cold sobriety), you must decide to switch the vehicle to a manual mode of operation. In some cases, this requires installing a manual mode of operation.
After this demonstration of superior technical skills and poor judgement, you must then deliberately impair your defective judgement further by quaffing, inhaling, injecting, or otherwise consuming something mildly toxic (or, in some precedent-setting cases, breaking up with your girlfriend.)
At this point you might argue that you are no longer responsible for your actions, which actions include climbing into your vehicle and attempting to pilot it home. Technically, this is true. The prosecution will counter by arguing that it isn't Manual Operation Under the Influence that you are going to be humanely (if rather publicly) executed for. It's the fact that you deliberately made such an irresponsible act POSSIBLE by putzing around with your ride. You know those signs that say "don't putz around with this system -- serious injury or death could result?" Well, they were talking about YOUR death, and it is now resulting.
"""
(http://www.schlockmercenary.com/2002-08-29) ;)
Driving isn't going to go anywhere, but I can see driving being "penalized" via exorbitant insurance rates - so much so that the it probably might offset the cost of buying a manually driven car.
It'll probably end up being a hobby, albeit an expensive one.
I agree in that I hope that ability for people to drive cars will go not away completely. Driving can be a pleasure and a rush.
I do hope that we would start holding drivers legally responsible for the deaths and injuries they cause. If you choose to drive a car yourself and you cause an accident that kills someone, then you should face a manslaughter charge. If you hit or injure someone, that should be an assault charge. I think it is ridiculous that this currently is not the case.
Your only argument seems to be that people should be allowed to do/have things that they like. I don't think this a very strong argument. We can come up a long list of things that people like(d) are no longer allowed because they are considered too harmful.
I wonder if there will come a day when you can still drive your car, but you'll have to pay exorbitant insurance rates. That seems like a reasonable midpoint.
Parking is one of the easiest and most valuable things to automate; it's finicky and humans basically suck at it. Self-driving cars may have manual controls but it probably won't be driven by parking.
Navigating places like parking lots, drive-thrus, garages, driveways, off-road, etc., gets pretty arbitrary. There's not much a self-driving car could even do for you, so manual controls will always be necessary.
I suspect passengers will be dropped off and the car will take itself to an automated parking lot. Likely pull onto a platform and get parked in a dense vertical 'parking farm'. Alternatively it could be put in 'rental mode' for x hours.
However if it is a taxi, it will likely go on to its next fare.
I'll admit that every time I see one of the self driving cars, I'm tempted to do something crazy nearby to see what it does, like cut it off or swerve towards it.
So far I've been a responsible adult and not done that But it's so tempting. Until I think of all the lives I will put in danger. Maybe they can invite me to a test track so I can be an a-hole to the self driving car in a safe and controlled environment. :)
So who's working on the AI that will punish anyone for creating Roko's basilisk, removing the motivation for creating Roko's basilisk in the first place?
You just described my ultimate "marketing" video for a self-driving car. Google should take one to a test track and get other drivers to mess with it and show us the video of what it does :)
Google says their self-driving cars were not at fault in all 11 cases and perhaps technically they were not but I don't think you can discount the possibility that the self-driving cars may have behaved in an unexpected manner that may have contributed to the accidents. A machine might make choices that are logical but which throw off other drivers. Strict adherence to the rules could cause problems. For example, there are cases where changing lanes in an intersection, although illegal, could avoid an accident. If a machine controlled car insisted on following the rules an accident might be unavoidable or the action could force another driver in to an unavoidable accident.
I don't look forward to the day when I'm no longer allowed to drive my own car but it would be awesome if robot controlled cars coordinated among themselves to avoid traffic jams and to operate safely at much higher speeds. I could see cars of the future running normally at 200MPH+ were typical human drivers would find it difficult to operate safely. Hey and no more traffic tickets. One possible way to move in this direction would be to introduce robot only lanes.
Given the information available on how people actually drive (besides that mentioned directly in this article), how would one distinguish the automated car behaving "strangely" from regular human driver weirdness?
I bet if you quiz twenty of your driving friends, you'll find more than three opinions on the right way to clear a four-way stop, for example.
When my 4 year old asks me when she will be allowed to drive, I tell her she won't have to, and that she'll just tell her watch to call a car for her, the same way she sees me talking to my smartwatch now.
The driver is not the problem of cars.
It is the fact that in average 1,5 t and 10 m2 are used to transport 100 kg and 1 m2.
Driverlessness does not change the bad physics of car transportation decisively.
Even worse than that inefficiency is that for most cars, for 22+ hours per day, they are a 1.5T 10m2 beast thats doing absolutely nothing but taking up space. On a big slab of asphalt that we've poured just so it can sit there.
Self driving cars might not be more compact, but if they can keep themselves occupied for a higher percentage of the day they can be a big win for efficiency.
The trumpet can be played one-handed, actually. Gets a little heavy after a while, but you can rest the bell on the dashboard. Also you won't be able to use the first and third valve slides to correct the lower notes, but who's listening?
I was thinking the same thing. If Google start logging license plate numbers (and, hey, it's Google), they could build a database of known dangerous vehicles and automatically give them a wide berth.
At that point they'd necessarily have some way to quantify dangerousity. There will be a ranking. Somewhere deep in the bowels of google, there will be a list, Top Ten most dangerous drivers. Google is about to get into the car insurance business.
Sure. But that requires atleast a subpoena. So there is atleast some expectation of due process. Which means Google or any other company can push back. Failing that, something like Snowden can happen.
Contrast that with police collecting that data and not requiring any, however made-up, oversight.
Can you give me some examples of what google might reasonably do with my data that can negatively affect my life to the same magnitude as what the police/government might reasonably do?
Ideally, maybe no one should track my every move. But to talk about police and google in the same sentence in terms of ramifications is so far off I can't be anything but incredulous :)
Police and Government should be two different entities, but if you're living in a police state then you're probably right.
Otherwise, what Google could do is, say, record your dangerous driving (maybe that one time you were rushing a friend to hospital and ignored some speed limits). And then, every time a potential employer googled your name, they would see this "dangerous driver" notification. And you have no way of removing this, there's no statute of limitations, there's no way to correct it. For the rest of your life, that's the first result anyone sees when looking up who you are.
again, my point wasn't so you could think of what Google "could do". But, rather contrast what Google might reasonably do and contrast with what the police might reasonably do.
But hey, feel free to over-react on google, but not about lack of gov. oversight.
Well I thought that was pretty reasonable. But since you disagree, lets look at what is actually happening now.
The youngest MP for a few hundred years was elected the other day. Her name is "Mhairi Black". Google her name, and see what you find on the first page.
Now, what have Police Scotland been up to....The last thing I remember seeing was a complaint about their ridiculous policy of sending armed officers to events that clearly didn't require them. They apologised and gave assurances that armed units would only be sent out in response to specific incidents.
Or do you mean personal interaction with the police? Last time I saw our local policeman, it was because his young son was scared of dogs as a big dog had knocked him down while playing. We took our puppy round so he could pet him and get over his fear.
I'm not sure what lack of government oversight you think there is. Holyrood is under constant scrutiny, not least by those who'd like to see it abolished!
farfetched and unrealistic at the moment, but totally plausible: compile a list of destinations you visit on some sort of schedule, commit some physical crime along that path during a time you can be shown to be in the area, and frame you for it, potentially by including things like relevant search data, falsified emails... lots of possibilities here.
I'm not sure how this varies from country to country, but in Germany honking is considered a warning signal (similar to hazard lights, but more direct) and intended to be used as such.
Of course most drivers are still primarily using it to vent their aggressions, but doing that is abusive and can result in a fine.
What would complicate automated honking, however, is that the specifics are likely different from legislation to legislation.
When the car is too good and fast to respond, it will get hit from behind it seems. I guess they could change their algorithm to be slow-reacting. dumbing down machines so that petty humans can catch up.
You'd want to get smarter, not dumber: taking into account the fact that the unpredictable (to the car behind) braking could in itself cause an accident, and calculating a best compromise to avoid both collisions, with priority to the more dangerous of the two (the fender-bender could be a great option if you avoid another car doing 50). I'd be surprised if their software doesn't support this kind of thing already.
I'd guess that most from-behind collisions are due to excessive speed for conditions, or driver inattention. Not much that the leading car can do about either of those.
The leading car can do several things in those scenarios, ranging from increased stopping time (which it should already likely be doing to maximize fuel efficiency), brake light usage, or in extreme cases changing lanes to get out of the way.
I hope they already did that, they say they can track all moving agents around, and roboticists know physics so they know they have to dissipate energy as smoothly as possible in the system including all neighbors.
About 33,000 people die on America’s roads every year.
There will eventually be a fatality caused by a self-driving car. I wonder how many of these we will be willing to accept. Say self-driving cars reduce fatalities by 10x. That is, by switching 100% to self-driving cars there are 3300 deaths a year, but they are all caused by self-driving cars.
Would America accept that?
I personally doubt we would. I think the number will have to be in the hundreds at most, closer to what we accept from plane crashes.
How safe do self-driving cars have to be before we start using them?
(Disclaimer: I work at Google but not on self driving cars. My opinion, not my employer's, follows)
What about the reduced individual and societal costs (in insurance, medical, lost productivity, psychological trauma) that come with a 10x reduction in driving fatalities?
People might see the value of a 10x (or even smaller) reduction if it is expressed in those terms.
Common misconception amongst technophiles like us: humans aren't rational. We accept flaws in humans, not so much in machines. Because we can sympathise with human error but a machine error is always a defect.
I think the problem is, people really don't care about autonomous objects.
There is so many unknowns with this, it's really hard to know. Like if the self-driving car gets in an accident (that is a result of the self-driving car), who is responsible? Google? Me? Other driver? Car maker? Car seller? Insurance company?
Humans have kind of accepted that humans make mistakes, you're trained from day 1 to when you die, humans make mistakes. It's a bit different when it's a program with a giant state machine.
> Like if the self-driving car gets in an accident (that is a result of the self-driving car), who is responsible? Google? Me? Other driver? Car maker? Car seller? Insurance company?
The insurance company is not going to be responsible in the first instance, though it may be insuring someone who is responsible.
For the rest, its hardly as if the responsibility for damages that aren't attributable to improper operation but are instead result from the design/manufacture of the vehicle and/or the failure of the owner to properly maintain the vehicle in a safely operable condition have not been extensively litigated already; so, while "self-driving cars" might be a new thing, the legal system isn't devoid of guidance on how liability could apply to them.
> Humans have kind of accepted that humans make mistakes
Self-driving car errors are all human mistakes, even if the mistake becomes visible distantly (potentially in both time and space) from the error itself.
currently there are only two sides, but with self-driving cars both sides could try to blame the other OR any other related entity (even from their own side) --what if both sides agrees to go after a 3rd one?
I think one of the problems is that when a self-driving car do cause an accident ,it will probably do so in an obviously incompetent way. Like running into a tree that was not mapped correctly, in a way a human would never do.
The cars may not be wrong often, but I think we would have a hard time accepting accidents that we cannot understand why they happened.
I don't think it's so different than people being killed by a mechanical failure in a plane. Sure you can put the failure back on the mechanic or engineer who designed the plane but you can do that with software too.
Come to think of it. Imagine driving a car yourself on a road full of self-driving cars. You would be able to drive as fast as you like and cut off anybody without ever being afriad of retaliation. Self-driving cars will always let you in because they do not want to cause a collision.
I imagine that reckless driving laws will still be in effect in that scenario, so that will likely deter the behavior, unless it is sociopathic in motivation.
I'm pretty excited for the prospect of a self driving car. My commute is about 40 miles round trip not too traffic ridden but I wouldn't mind relaxing a little. I would rather telework more often than two days a week, but it looks like he self driving car is more imminent.
Considering that the number of vehicle-miles driven annually is in the trillions, that is still a pretty small sample size. Do we know whether it's statistically significant?
The number that people drive isn't important, what's important is the accident rate: do humans get into enough minor accidents over 1.7m vehicle-miles that 14 is clearly bigger or smaller? Your answer is going to be driven almost entirely by what human-rate you select.
For example, if I use https://www.dot.ny.gov/divisions/operating/osss/highway/acci... , then I see that most of the quoted accident-rates are 1.8-3.29 per million miles. If I use a rate of 2.2/1m and use the simple binomial approach of x events per million miles, then the self-driving cars are statistically-significantly higher accident rates (since 14>3.74, and `binom.test(14, 1700000, p=(2.2/1000000)) ~> _p_=3.778543e-05`).
But what kinds of accidents are those? If I instead look at https://www.dot.ny.gov/divisions/operating/osss/highway/acci... and look at "Property Damage Only" accident-rates, it seems the human rate per million mile is more like 65-85. Now, unsurprisingly, the robots look superior (14<110, `binom.test(14, 1700000, p=(65/1000000)) ~> _p_=2.2204e-16`).
And if you try to argue apples and apples and use fatal accidents from the same source, noting a range of 0-1.33 fatal accidents per million miles, and 0 fatalities due to the robots' 1.7m so far then the difference is unclear (0<0.85 but `binom.test(0, 1700000, p=(0.5/1000000)) ~> _p_=1`) because the usual tests don't like 0-counts and the robots have not driven enough to clearly distinguish themselves from normal human rates.
What will be interesting will be how human behaviour will change once all cars are self driving cars. Some cyclist / pedestrians might blatantly cross a red light intersection because he knows the self driving cars will detect and stop for him... something to think about.
One thing I'm looking forward to with self-driving cars is to be able to step into a street anywhere and have all the traffic stop automatically. Every inch of street now becomes a crosswalk. Even highways.
Not to mention that with the self-driving cars working as a functional panopticon, doing this and getting away with it for lengths of time is unlikely.
Even ignoring the privacy-scary scenarios of the car auto-emailing the cops to say "Dear local PD, this guy just jaywalked and halted traffic on Oak Ln, here's my [sensor gestalt] to prove it and a [photo of the offender]", it'd be pretty trivial for anyone owning a fleet of these machines who was tired of jaywalkers slowing down their commute to dump data on jaywalk-stops in the aggregate, giving cops all the numbers they need to target enforcement.
Also belgium and germany. Getting too close to the street/sidewalk interface will often cause cars to stop even if you have no intension whatsoever of crossing.
Automatically != instantaneously. You'd have to make sure the car is far away enough that speed(car)^2/distance(you, car) < friction(rubber, pavement)*g.
You can't play blindfolded frogger.
EDIT: "friction" is a better function name than "mu", even though we all feel nostalgic for high school physics.
It's more complicated, since cars can steer, but it's still approx. function of speed^2/distance, but a lot more permissive. And then there's emergency coordination among vehicles to prevent the crash among the vehicles themselves too.
There is a table saw that stops the blade the instant it senses the capacitance of a finger. It completely ruins the saw but saves the finger. Something like that could work here if the car could see the pedestrian far enough in advance to stay within parameters. Ruins the road and the car not the pedestrian.
Those jam an aluminum block into the blade. I'm not sure what the equivalent would be for a car. Perhaps you could line the bottom of the car with rubber and then blow the wheels off. But there's plenty of problems with varying road surface and conditions and having a system that's never used.
But a decent brake system and tires can do 1G, which is pretty good. I suppose you could go to maybe 5G for a true emergency brake system (which would be pretty rough on the occupants, eyeballs-out), but it hardly seems worth the cost (in money, weight, maintenance, accidental triggering, etc.) compared to good sensors and the existing braking systems.
Yeah, and don't forget that the car has people in it too :-)
While you can instantaneously stop a saw blade to save a finger, stopping a car 'quickly' will likely move the potential problem to one of ensuring the safety of the passengers...
I think the equivalent of jamming a block of solid aluminium into a spinning blade would involve some kind of anchoring.
It would ruin the car and likely damage the road but as long as the passengers can be protected from the rapid deceleration, it just might save a life.
OTOH you could just wrap the entire car in airbags and protect pedestrians the same way you protect the passengers. Not sure how practical or useful that would be, though.
Wow, that's super neat, i'd never heard of it. Seems like something that should be absolutely mandatory for all factories equipped with such devices! Go technology!
Think about the physics. There's a _whole_ lot more energy in a moving car than in a saw blade. The car is also being operated in physical circumstances that vary wildly (wet road surfaces, high wind, being struck by other cars) that don't happen in the comfortable little micro-universe of a spinning saw blade.
You are thinking of http://sawstop.com/why-sawstop/the-technology. The problem with your comparison is that cars already have brakes. Driverless cars would improve reaction time, but can't do anything to affect stopping time. You would need to activate some additional braking system.
<snark>I also don't see how a car could dive below the roadway instantly. Roads don't have car-sized holes in them. Usually.</snark>
Srsly, it amazes me when ppl don't get out of the way for emergency vehicles.
IMHO one random day out of the year, you should be equipped with heavy machine guns and allowed to blast anyone in your way. A simple solution for a simple problem.
When visiting Vietnam, I was told to look straight ahead, and move at a consistent and predictable pace. The drivers are used to moving around pedestrians. Its a wild sensation.
This could be a big deal actually. Will pedestrian and human driver's behavior change if there is the expectation that the self-driving cars are going react incredibly quickly and reliably? It's hard to imagine everything just stays the same everywhere and as self-driving cars become more and more common.
I imagine there will be pissed passengers in the car who are sitting stopped for far too long because some jackass things he's above traffic laws. I suspect the self-driving car will also usher the age of people getting beat up for fucking with self-driving cars.
There's a million social elements here we haven't explored. Imagine driving through a ghetto and lots of people rush over to your car to rob you (or worse). Your car will happily brake for them, when a human would have recognized the situation and accelerated and veered off to avoid them. Or weird edges cases like garbage on the road or somesuch and the car not being able to outmaneuver it because it may be illegal, but a human with a more lax acceptance of traffic laws will veer into the right lane close to another car to avoid it.
Or just the daily grind of speed limits. On the expressways I drive the posted speed is 65-70 yet most everyone does 80-85. Doing a steady 60-65mph because of how laws were written to get speeding ticket revenue might be annoying. Especially if you're the only non-human driven car and everyone is speeding past you and honking because you're not going the speed of traffic.
You should forward this post to your psychoanalyst. It's a rich, concentrated dose of mental pathology for his consideration, or maybe it's just the ramblings of an old fart looking back on a sixth of his life spent commuting. The cool kids aren't staring out the window searching for bums to beat up. (besides how would they stop the car or get the door open?) The cool kids are playing a game on the shared table in the middle of the car (the front seats face to the rear, you know) or they're working on their phablet or they're videoconferencing with their coworkers in other robocars.
And if there were no self-important driver to get pissed off, would "jaywalking" even be a thing? If the first automobile drivers had been poor folks with little political power (haha yeah right), "jaywalking" would never have been made a crime in the first place. Some streets will have heavy pedestrian traffic; the rational robot routes around them to lower-traffic streets.
I find your post somewhat detuned from reality. For instance, jaywalking is rarely a crime. Also, jaywalking becoming a thing was precisely related to cars becoming federated to the "poor folks with little political power" because in many communities, there is heavy car traffic that was prioritized over walking due to damn near everyone having at least one car.
I'm looking forward to when all of the self-driving cars will be so good at avoiding accidents that you can just take a manually-driven car and blast down the center of the road at 10-20 mph over the speed limit and watch all of the self-driven cars just move out of your way.
... until the smart-grid notices the disturbance in regular traffic flow and auto-notifies the authorities that there's something weird going on on Oak Ln. every Monday morning during rush hour... ;)
In all seriousness though, what would happen in this case? Do self-driving cars have emergency stop procedures? And how safe are they? This is still a serious concern about self-driving cars for me. Especially since digital image recognition still isn't that great. What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode?
> Do self-driving cars have emergency stop procedures?
Yes, of course! Just staying on a road is a piece of cake. Most of the research for self-driving cars is dealing with unexpected cars and pedestrians.
> And how safe are they?
According to the article, 1.7m miles driven without at at-fault accident. Self-driving cars drive very conservatively.
> What if the car sees a tumbleweed that it mistakes for a boulder and puts it into emergency stop mode?
Probably the same thing that happens when humans make visual mistakes: it would slow down and stop.
Of course, the self-driving car is looking at the tumbleweed in the visible spectrum, using LIDAR, maybe IR and probably some other sensors. We just have our squishy eyeballs.
I guess what I'm asking is are the emergency stop procedures safe? If you're going 75mph, how much distance does it need to stop without rolling or harming the passengers?
I read over it, but there are a couple of problems with that. Firstly, I never said that computers understand the images, just that the recognition is better than human. When they get it wrong, they do tend to get it wrong in ways that look ludicrous to people. Secondly, Stanford didn't win the 2014 ImageNet challenge: GoogLeNet did.
Page 31 states: "Annotator A1 evaluated a total of 1500
test set images. The GoogLeNet classification error on
this sample was estimated to be 6.8% (recall that the
error on full test set of 100,000 images is 6.7%, as shown
in Table 7). The human error was estimated to be
5.1%."
There are some bits there where GoogLeNet did better than human, but it was usually on specific classifications (what breed of dog is that, for example). Generally, a human is slightly better.
It's worth noting, however, that the error rate was halved from 2013's winners. I reckon the 2015 error rate will beat human, and in following years there will just be no contest. Just like chess programs have gone far beyond human capabilities.
And, of course, with cars we're talking about having multiple images plus depth perception, which makes it a lot easier for the machine.
For balance: In the past eight months, four of the 48 self-driving cars licensed in California were involved in accidents. Three were Lexus SUVs from Google. http://www.startribune.com/ap-exclusive-self-driving-cars-ge... This does not assign blame. It is an observation.
According to the companies involved, none of the accidents were the fault of the self driving car.
The Delphi car was stopped at a light, in manual driving mode, when another vehicle lost control, crossed a median, and hit the delphi car.
I haven't seen a statement from Google other than that the accidents were minor and the fault of the other vehicles.
I think today's news article flurry might better be titled: Humans run into self driving cars about as often as human driven ones, meanwhile self driving cars do not run into other vehicles.
> If you spend enough time on the road, accidents will happen whether you’re in a car or a self-driving car. Over the 6 years since we started the project, we’ve been involved in 11 minor accidents (light damage, no injuries) during those 1.7 million miles of autonomous and manual driving with our safety drivers behind the wheel, and not once was the self-driving car the cause of the accident.
I'm pretty sure I've seen Google's car being tested up in Tahoe, so the answer is some. (The areas around Tahoe's ski resorts get quite a lot of snow and ice too, so it's certainly a good place for this sort of testing.)
To not be legally "at fault" is important, particularly for a company with liability and public image at stake.
I genuinely look forward to future stats that will tell us self-driven cars have a much lower likelihood of even being involved in accidents, independent of fault. To me, that level of defensiveness and net safety is also a measure of progress.
As an example: Let's say a particular self-driven car has a low threshold to "stomp on the brakes, hard" at the slightest sign of danger. Will this keep it from accumulating "at fault" accidents? Probably. But can that behaviour effectively contribute to causing accidents that it's not "at fault" for (being rear-ended)? Probably. Safer roads? Difficult to say.
"we’ve been hit from behind seven times, mainly at traffic lights but also on the freeway"
If you are stopped at a red light with a car in front of you, sitting there and taking a collision from behind can be the best course of action, depending on the trailing car's speed.
Now, many PDO incidents go unreported, but, medium does not offer a number for comparison, so, when you see the headline "11 accidents in 1.7m miles" you have no reference to whether this is high or low.
Good find. The text of the DOT website is a little vague, but I think you're right that the 1.45 per 1M number is only from police reports.
Data for GES [General Estimates System] come from a nationally representative
sample of police reported motor vehicle crashes of all types, from minor to fatal.
... Although various sources suggest that about half the motor vehicle crashes in
the country are not reported to the police, the majority of these unreported crashes
involve only minor property damage and no significant personal injury.
I guess that's saying the actual number should be about double, so ~3 per 1M.
I'm also wondering if any private operators of fleets publish stats. For example, does FedEx or UPS publish accident rates? I know those numbers would also be skewed, larger vehicles, more frequent stops, etc. But it could be another data point.
For more balance, they have been driven a ridiculous number of miles more than the average car so saying "4 of 48" got ran into (since none were the fault of the self-driving car!) is intentionally misleading.
Not only is that a marketing win, potentially reducing accidents and saving costs for owners... if you jump forward many years of Google (or other) defending owners in court you would reach a point even judges would be biased to believe that the auto-driven car could not have caused an accident, further disadvantaging non-auto cars in the market.