Google Self-Driving Car Project
google.com
google.com
Will Google’s self-driving cars get into accidents? Have they gotten into accidents before?
> Safety is our top priority. In the 6 years of our project, we’ve been involved in a small number of accidents in more than 1.8 million miles of autonomous and manual driving combined. Our vehicles have not caused any accidents while in self-driving mode. For more information, view our monthly reports.
So no accidents in self-driving mode; a "small number" during "manual driving"...and credit to Google, even though there is only one monthly report, it contains a listing of past accidents:
http://static.googleusercontent.com/media/www.google.com/en/...
I skimmed over the dozen listed incidents. Several of them were during "autonomous" mode but are incidents in which another car is described as hitting the vehicle. Arguably, none of the "manual" mode accidents were egregiously the fault of Google.
However, I think the definition of "caused" will be debate here and for years to come. In one Feb. 2015 incident, the autonomous Google car is struck in the side rear by a car that rolled through a crossbound stop sign. The Google car is described as "Applying the brakes in response to its detection of the other vehicle's speed and trajectory"...since the collision ended up happening anyway, and it hit the rear of the Google car...doesn't that imply that if the Google car hadn't cautiously applied the brakes while going through the intersection, it would have cleared the intersection without getting hit? Also worth noting: the Google human driver tried to take control during the autobrake sequence...so it's possible that his/her reaction and manual braking was what led to the rear collision.
It's worse on county roads (like in my local area, 50mph) without passing lanes. If you drive the speed limit you can almost guarantee to be passed by many cars, sometimes even dangerously, because it annoys people to go below the speed limit they perceive as correct.
I like to go around the speed limit (admittedly, simply to avoid tickets), usually ~3mph over, but it will annoy me to no end if i have to drive 5mph slower because of someone in front of me. On these popular county roads i speak of, i usually go faster than i want (5-10mph over), because i don't like being dangerously and needlessly passed.
Now, i'm not saying the self-driving cars are wrong in any way. I'm also aware that the roads i speak of will be among the last to get self driving cars. Nevertheless, i think humans will have to learn to drive far less aggressively. Though, when self-driving outnumber the normal drivers, it may not matter.
Granted, Google's security is about as good as it gets for most stuff, but what about companies like Audi, BMW [1], Daimler, Ford...Toyota [2]? Many people will still buy cars from them well into the future.
That said, when they sell one of these things to a person, how the heck are they going to handle the insurance?
Once robot taxis are everywhere, I would expect that there will be robot car rentals and robot car long-term rentals. They are also not different from the taxis; people just rent them for longer.
As for real ownership, the question is unclear. Probably, it will be so rare, that a deposit will work too.
Forget Uber. Get a Google-AV subscription, find the nearest "unoccupied" AV (or call for one to come pick you up, preferably scheduled in advance to it's just waiting for you when you get out), hop in, state destination, get off. Car continues on its merry way to its next duty. Repeat when you want to go back home. Much less massive waste of space involved in all those parking lots. Much less opportunity cost waste. Less total cars in existence for the same amount of travelers.
It's like a public transit pass, but the bus stops are exactly where you want them, and the bus passes exactly when you need it, and you don't have to deal with that insane, smelly old guy who dances in the middle of the bus and then shakes peoples' shoulders so they give him pocket change.
This is not the same for human-caused incidents...though we do fixate on pieces of incidental data that might exist: is there a record showing that the driver sent a text right before they crashed? Are there credit car receipts for a bar? Or, in a more extreme case, the medical history of the Germanwings pilot who is believed to have downed his plane.
With autonomous vehicles, two cars of the same model running the same software should be more or less interchangeable. So even leaving aside the number of accidents that occur, the level of uncertainty should be a lot lower.
Insurance companies will hate insuring these cars because the first human fatality is absolutely going to court, and it'll be a complete circus. The media will go apeshit.
There will be a large fluctuation prices as data becomes available but its like insuring any other system ultimately. They calculate the odds, add in their needs, and come up with a figure.
I doubt it'd be significantly different than today since all of these can be switched to manual mode.
http://www.openstreetmap.org/#map=18/37.38487/-122.08136
Without more information about the collision, I think it's hard to say what would have happened if the system had not applied the brakes.
I know this is simply phase one, but I'm hoping google puts some emphasis on individuals in wheelchairs in future prototypes because this technology can be life changing for some with extremely limited mobility - my mother being one of them. Independence is key here and the the majority of us can't conceive what it's like to to be dependent on others for everything. A user with a wheelchair or power chair needs to able to get in and out of the self driving car with zero assistance.
On the other hand I am glad somebody is pouring money into fundamental research. The government has limited funds. Many big companies that formally had big R&D cutback during downsizings.
At the same token it is because they have these vast troves of data and some of the best machine learning scientists in the word, they are suited to do this type of research.
Another thing to think about is that if we are not distracted by the commute we will spend more time using Google services.
Let's say they invest...I dunno $300m in developing this technology (completely made-up number)
- The average commute time in the U.S. is ~30 minutes...or 1 hour per day. I don't know what it is globally, but for fun we'll assume it's the same for everybody on the planet.
- Google made $66billion in 2014 (or about $9.43 per person on the planet, assuming 7billion people)
- Assume most people are awake 16 hours during a day.
- However, if people are stuck driving for 1 hour out of those 16, that's only 15 hours they can build revenue for google.
- This means google is generating about $.63 per available waking hour per person.
- 1 more available hour is $.63*7billion = a possible "new" market of $4.4billion/year assuming every person
This would pay back the R&D costs handsomely + tons of profit in the first year if google can get everybody to switch all at once.
Even if it takes a decade or two to switch humanity to self-driving cars, it would still pay for itself quickly.
I tip my hat to Google, for attempting to stay ahead of the curve.
1. Enable more people to use your service 2. Create devices that can gather more data
The thing is, a ton of device types can fit into one of these schemes. Phones for example, both let people use your service, as well as gather a ton of data. As we increase the technology in phones, they may even be mapping the environment.
Likewise, these self-driving cars are able to exploit much of Google's current offerings, as well as contribute mapping of a near limitless amount of roads/etc.
If Google can pull far ahead in the technology they can become the self-driving service. Even if they don't create the cars but instead lease the technology to normal makers, Google now has constantly updating maps, traffic patterns, routes, building changes, etcetc.
Sure, Google made a car, but i don't think they car in the slightest about the car itself. They need to push to improve the technology and legislation as fast as they are able - everything else, people and car manufactures, will catchup eventually.
Google Now - In car edition: "Since we have plenty of time before your appointment, press 'accept' to stop at the showroom for those soft-furnishings you searched yesterday, and receive a credit towards this ride".
http://adage.com/article/digital/google-profit-misses-estima...
> Google hasn't posted an annual increase in the average cost-per-click since the third quarter of 2011. It hasn't posted a quarterly boost since the second quarter of that year. Both streaks remain intact.
Their per-click margins aren't going up and there is a limit to what they can do in that situation to grow net revenue.
So they are looking for new growth markets since the margins on their core business aren't really going to change substantially with new investment.
http://en.wikipedia.org/wiki/Google_driverless_car
Throwing small amounts of money at things like driverless cars is something they can control and will likely give them the first mover advantage in a new market. They take over and become the "Microsoft of Driverless Cars" and they've got a huuuuuuuuuuuge new profit center.
If not, they blew through a bit of cash. They have plenty of cash.
Sensor technology hasn't improved much. It's still mostly cameras, Velodyne rotating LIDAR units and off the shelf centimeter Doppler RADAR units. Flash LIDAR and millimeter RADAR aren't volume products yet. They will be once auto companies get serious about this.
The DARPA Humanoid Challenge is live today. Watch: http://www.theroboticschallenge.org/
The problem is, they drive the same routes over and over. I always see them in the same places, and nowhere else. They have hyper-mapped a small number of routes (mm resolution). And all that highway driving: MV to SF and back on 280, over and over and over again.
I am sure they are making progress, but how much?
Then again, Smart Cars are equally ugly and they have gained some huge traction, at least with car2go, so what do I know.
Say for instance that two pedestrians, a young child and an old person, suddenly find themselves in front of a self-driving car. There is not time for the car to brake in time so the algorithm has to make a choice: Which pedestrian gets hit?
I think we'll see more questions like these in the coming years as AI progresses and we become more dependent on it.
Let's give it the most realistic scenario of a blind-turn intersection with an obstruction such as a bush make it impossible to see the sidewalk on the right-hand side. The two pedestrians in question are a grandparent and their grandchild crossing without looking both ways for their safety.
Is there oncoming traffic? Is there an empty sidewalk to drive on? Can the car make an attempt to brake as safely and quickly as possible and sound the horn to alert the pedestrians to get out of the way? Is it safe enough to perform a handbrake turn? Is it safe enough to perform a bootleg turn?
Before asking which pedestrian the car chooses to hit, I would exhaust all other available options to prevent such a scenario from happening in the first place.
The scenario is one that requires so many worst-case-scenario events to be happening simultaneously as to say the scenario is likely avoidable altogether in one of numerous ways. So yes, I'm calling your hypothetical scenario a little contrived.
To answer the hypothetical with the answer you're expecting; people have already made this choice in the past for disaster scenarios:
"Women and children first."
The algorithm would choose to hit the elderly person over the child in your scenario. The ethical choice becomes more complex if the pedestrians in question are both children or both women. I don't think humans have made an ethical decision on such a choice.
So yes, I'm calling your hypothetical scenario a little contrived.I was not expecting any particular answer, I simply find the problem interesting. It begs the question: Does this mean there is a line of code somewhere that ultimately has to make that decision?
Without the code a more simple scenario plays out. The car continues along its current trajectory and kills whichever pedestrian was in front of the vehicle, which could be both of them.
Women and children first is not a policy of who is most valuable. It is a policy of who is most vulnerable. In disasters it is presumed that the women and children cannot fend for themselves (and while there is a biological aspect to women being on average weaker than men the presumption of women and children first is still painfully sexist) and that the men coming up behind are more likely to survive than any alternative, where men go first and "the vulnerable" take up the rear.
Even besides that, how do you know if a child is more valuable than an elderly person?
Its not as simple a problem as yesteryears well intentioned sexism and ageism. But the real answer is much simpler:
The car would, in the situation where it could not avoid hitting a person, hit whomever it computes to be most likely to survive.
The choice of children vs elderly rarely comes down to value - but rather life. The elderly person has lived their life while the child, potentially, has many more years ahead of them. Humans don't tend to give a value to life other than life itself - and extending life or saving multiple lives over a single life tends to be the popular choice. It would be impossible to judge a persons' "value" or measure it without having information on them prior to the event.
See: The Trolley Problem
[0] http://healthland.time.com/2011/12/05/would-you-kill-one-per...
There are tens of thousands of people dying every year in very real car crashes, plus probably an order of magnitude more suffering life-changing injuries, and self-driving cars have a huge potential to cut way down on that. Meanwhile, situations like that virtually never happen - maybe like once a year in the entire US. If they can eliminate even 10% of the actual accidents, then they could run down both pedestrians in that imaginary scenario, and still be way ahead.
It may be fun to think about ethical dilemmas like that, but they are fantastically rare compared to the huge numbers of perfectly ordinary crashes that happen every day. Let's fix those, and worry about the rare one-offs after we've reduced the accident rate by 95%.
It will be interesting to see how insurance rates change for autonomous vehicles—a vehicle with no potential for manual control should in theory be much cheaper to insure.
If its because human drivers can better "build a mental model" of whats happening with limited visibility/data - that seems like the same problem that autonomous cars are designed to solve.
Whats stopping autonomous cars from making progress in the bad weather arena? Is it just we haven't done enough miles in that weather and we need to train the ML some more?
We don't yet know how the human brain processes images at a detailed level, and it's too complex of a system to optimize by brute-force. The best ML image processing systems still rely on a lot of pre-encoded assumptions, and require huge amounts of training data/computational resources, and don't perform as well as humans even on simple tasks.
For now, autonomous vehicles "cheat" a lot by using sensor technologies like lidar and radar -- techniques that produce spatial data that can be interpreted more easily than raw imagery. That adds a lot to the hardware costs, but it's currently the only way we can make them perform acceptably well in real-world scenarios. And conditions that interfere with those technologies more severely, like fog or snow, are still crippling.
Mind you, I'm extremely optimistic about the potential successes of autonomous cars in the near- to mid-future (the next 5 to 25 years, say). But I don't want people to get the wrong impression about where the state of the art is today. And in particular, I want to push back against the idea that machine learning is a solved problem, and that autonomous cars can teach themselves to drive as well as a human if we only give them enough miles of practice.
[Because they are.](http://jalopnik.com/this-is-how-bad-self-driving-cars-suck-i...)
Basically, the sensors they tend to equipped with aren't even close to being sufficient in poor whether. Humans are EXTREMELY good at pattern recognition and can manage to make it work, but computers just aren't at that level yet. It'll take a combination of improved sensors and computer vision (ie, pattern recognition) to make it happen.
Arguably, no car should be driving in those conditions unless it is a life or death situation. Because you have a very good chance of dying doing so.
5 years tops. I'd bet on it.
I think we won't need to own cars and will simply press a button like Uber and the car drives us wherever.
I'll bet you $1k that fully autonomous cars will not be: (1) legal to ride on at least 90% of public roads AND (2) available for sale to the general public AND (3) for under $250,000 in 2015 dollars AND (4) in the state of California AND (5) on or before June 5 2020.
Deal? (If I were richer, I'd bet more, in order to up my chances of being happy either way. I want to lose, but don't think I will.)
If you want to hammer out details, my email is in my profile. We'd use LongBets [1] to keep track, and under the condition that the loser has to donate the amount chosen to a charity of the winner's choice.
I imagine something like this Google car could go to market for around $80k today, and thats factoring in huge amounts of per-unit overhead for building factories to assemble them en masse in the first place.
An auto upgrade kit to a traditional car probably won't cost more than 20k if they permit the retrofitting of old vehicles when the ball starts rolling.
What I'm excited about is how we will soon see more and more autonomous 'features' i.e. lane changing, smarter adaptive cruise control and parking lot mode.