Self-driving cars getting hit in California

May 11, 2015 byJustin Pritchard
Self-driving cars getting dinged in California
This May 13, 2014 file photo shows a row of Google self-driving Lexus cars at a Google event outside the Computer History Museum in Mountain View, Calif. Of the nearly 50 self-driving cars rolling around California roads and highways, four have gotten into accidents since September, 2014. That's when the state officially began permitting these cars of the future, which use sensors and computing power to maneuver around traffic. Three accidents involved Lexus SUVs run by Google Inc. The fourth was an Audi retrofitted by the parts supplier Delphi Automotive. Google and Delphi said the accidents were minor and their cars were not at fault.(AP Photo/Eric Risberg, File)

Four of the nearly 50 self-driving cars now rolling around California have gotten into accidents since September, when the state began issuing permits for companies to test them on public roads.

Two accidents happened while the cars were in control; in the other two, the person who still must be behind the wheel was driving, a person familiar with the accident reports told The Associated Press.

Three involved Lexus SUVs that Google Inc. outfitted with sensors and computing power in its aggressive effort to develop "autonomous driving," a goal the tech giant shares with traditional automakers. The parts supplier Delphi Automotive had the other accident with one of its two test vehicles.

Google and Delphi said their cars were not at fault in any accidents, which the companies said were minor.

Since September, any accident must be reported to the state Department of Motor Vehicles. The agency said there have been four, but would not comment about fault or anything else, citing California law that collision reports are confidential.

The person familiar with the accident reports said the cars were in self-driving mode in two of the four accidents, all of which involved speeds of less than 10 mph. The person spoke on condition of anonymity because they were not authorized to discuss the reports publicly.

Five other companies have testing permits. In response to questions from the AP, all said they had no accidents. In all, 48 cars are licensed to test on public roads.

The fact that neither the companies nor the state have revealed the accidents troubles some who say the public should have information to monitor the rollout of technology that its own developers acknowledge is imperfect.

John Simpson, a longtime critic of Google as privacy project director of the nonprofit Consumer Watchdog, pointed out that the company's ultimate goal is a car without a steering wheel or pedals. That would mean a person has no power to intervene if a car lost control, making it "even more important that the details of any accidents be made public—so people know what the heck's going on."

Self-driving cars getting dinged in California
This May 13, 2014 file photo shows a Google self-driving Lexus at a Google event outside the Computer History Museum in Mountain View, Calif. Of the nearly 50 self-driving cars rolling around California roads and highways, four have gotten into accidents since September, 2014. That's when the state officially began permitting these cars of the future, which use sensors and computing power to maneuver around traffic. Three accidents involved souped-up Lexus SUVs run by Google Inc. The fourth was an Audi retrofitted by the parts supplier Delphi Automotive. Google and Delphi said the accidents were minor and their cars were not at fault.(AP Photo/Eric Risberg, File)

A chief selling point for is safety. Their cameras, radar and laser sensors give them a far more detailed understanding of their surroundings than humans have. Their reaction times also should be faster. Cars could be programmed to adjust if they sense a crash coming—move a few feet, tighten the seat belts, honk the horn or flash the lights in hope of alerting a distracted driver.

A higher priority so far is teaching them to avoid causing a serious accident that could set public and political acceptance of the technology back years, said Raj Rajkumar, a pioneer of the technology with Carnegie Mellon University.

In the October accident involving Delphi, the front of its 2014 Audi SQ5 was moderately damaged when, as it waited to make a left turn, another car broadsided it, according to an accident report the company shared with AP. The car was not in self-driving mode, Delphi spokeswoman Kristen Kinley said.

Google, which has 23 Lexus SUVs, would not discuss its three accidents in detail.

The accidents are not Google's first: In a briefing with reporters a year ago, the leader of Google's self-driving car program acknowledged three others between when the company first sent cars onto several years ago—without the state's official permission—and May 2014.

In a written statement, Google said that since September, cars driving on streets near its headquarters in Mountain View had "a handful of minor fender-benders, light damage, no injuries, so far caused by human error and inattention."

Google said that while safety is paramount some accidents can be expected, given that its cars have gone "the equivalent of over 15 years of typical human driving," or approximately 140,000 miles.

The national rate for reported "property-damage-only crashes" is about 0.3 per 100,000 miles driven, according to data from the National Highway Traffic Safety Administration.

Self-driving cars getting dinged in California
This May 13, 2014 file photo shows a row of Google self-driving Lexus cars at a Google event outside the Computer History Museum in Mountain View, Calif. Of the nearly 50 self-driving cars rolling around California roads and highways, four have gotten into accidents since September, 2014. That's when the state officially began permitting these cars of the future, which use sensors and computing power to maneuver around traffic. Three accidents involved souped-up Lexus SUVs run by Google Inc. The fourth was an Audi retrofitted by the parts supplier Delphi Automotive. Google and Delphi said the accidents were minor and their cars were not at fault.(AP Photo/Eric Risberg, File)

In that context, Google's three in about 140,000 miles may seem high. As the company pointed out, however, perhaps 5 million minor accidents are not reported to authorities each year, so it is hard to gauge how typical Google's experience is.

Three other states have passed laws welcoming self-driving cars onto their roads. Regulators in Nevada, Michigan and Florida said they were not aware of any accidents.

As self-driving cars proliferate, others issues will arise that human drivers have dealt with for decades, notably who's liable for an accident. Each test car is required to have $5 million insurance.

Interest in accidents will remain high, especially if the self-driving car is at fault, said Bryant Walker Smith, a law professor at the University of South Carolina who has written extensively on the technology.

"For a lot of reasons," Smith said, "more might be expected of these test vehicles and of the companies that are deploying them and the drivers that are supervising them than we might expect of a 17-year-old driver in a 10-year-old car."

Explore further: Google self-driving cars pass 300,000 mile mark

Related Stories

Google self-driving cars pass 300,000 mile mark

August 8, 2012

(Phys.org) -- Google has just released an update on its blog boasting about how its fleet of self-driving cars which the company has designed and is operating on public roads, have collectively racked up over 300,000 miles ...

Self-driving cars now need a permit in California

September 16, 2014

Computer-driven cars have been testing their skills on California roads for more than four years—but until now, the Department of Motor Vehicles wasn't sure just how many were rolling around.

Recommended for you

Apple issues update after cyber weapon captured

August 26, 2016

Apple iPhone owners on Friday were urged to install a quickly released security update after a sophisticated attack on an Emirati dissident exposed vulnerabilities targeted by cyber arms dealers.

Auto, aerospace industries warm to 3D printing

August 25, 2016

New 3D printing technology unveiled this week sharply increases the size of objects that can be produced, offering new possibilities to remake manufacturing in the auto, aerospace and other major industries.

31 comments

Adjust slider to filter visible comments by rank

Display comments: newest first

Eikka
2.6 / 5 (5) May 11, 2015
Their cameras, radar and laser sensors give them a far more detailed understanding of their surroundings than humans have.


I think "understanding" is not the correct term in this context. That's a great part of the problem.

verkle
May 11, 2015
This comment has been removed by a moderator.
Lex Talonis
1 / 5 (9) May 11, 2015
"the company's ultimate goal is a car without a steering wheel or pedals."

Well this right here is a HUGE fucking problem.....

Even for minor issues, like it breaks down, and you want to push it off the road?

What then?

The people running Google should be punched in the face, just for being fucking idiots.

dogbert
1.5 / 5 (8) May 11, 2015
the company's ultimate goal is a car without a steering wheel or pedals.

So, the system fails while you are traveling the interstate system at 70 mph (say and electrical system short). You have no steering wheel and no brakes. When you slam into the bridge abutment at 70 mph, you will be comforted knowing that:
Their cameras, radar and laser sensors give them a far more detailed understanding of their surroundings than humans have.


Just the moderate amount of automatic response built into many of today's cars (auto braking, for example) is capable of causing serious injuries and death.
zoljah
2.8 / 5 (4) May 11, 2015
If its connected to the Internet or similar networks..
What if some hacker is capable of and decide to modify the code so instead of preventing impacts, the cars start to aim for .. well too much of imagination?
dogbert
1.5 / 5 (4) May 11, 2015
zoliah,
I am sure those cars are as connected as possible -- GPS, WiFi, etc. You point out another vulnerability and one which someone will doubtless exploit.

Consider a terrorist seeking to do damage ...
krundoloss
4.3 / 5 (4) May 11, 2015
The real difficult part is trying to design an automated car to handle an organic world and all its randomness. This would work much better if all cars had these systems, and there were certain areas where all the cars in that area were controlled centrally, by computer. The computers could do a much better job if they controlled all the vehicles and had much less to deal with. Something similar to the scheme in "Minority Report", where in the city cars drove themselves, but once you left that "grid" the car had to be driven by standard means. This setup seems to offer the best of both worlds, with less risk than trying to mix human and computer drivers.

I am uneasy with these new "autobrake" features of cars. Braking is not always the best solution to collision avoidance and reduction. What is a calculated swerve is a better reaction, but the auto brake limits the turning ability and causes you to crash? We should, as drivers, at least have the ability to turn these features off...
dogbert
1.8 / 5 (5) May 11, 2015
krondoloss,
I am also concerned about autobreaking and autosteering cars. Years ago, I was driving on a four lane road in a town. I was in the right lane when a car full of teenage girls turned in front of me. There was no way to stop in time and I could not drive away from an accident. Instead I allowed and guided the car to strike the other car with my front left corner on the area between the front and back seat ofc the other vehicle. No one was hurt. Any other response would have seriously injured or killed someone. If the car had had today's technology, the outcome would have been bad.
bid
5 / 5 (3) May 11, 2015
This would work much better if all cars had these systems, and there were certain areas where all the cars in that area were controlled centrally, by computer.


I think a centralized system is a bad idea, since you introduce a possible single point of failure. It may be a better idea, if every car is able to communicate its "perception" and immediate "plans" to all cars in the vicinity. That way they could work together to find the best possible solution for any given situation. And of course every car would have a LOT more informations about the outside world, when sharing the points of view of other cars around.
bid
4.3 / 5 (6) May 11, 2015
If the car had had today's technology, the outcome would have been bad.


How do you know without doing any tests? Maybe the computer would have come up with another, maybe even a better way.
PhysicsMatter
2.7 / 5 (6) May 11, 2015
Major problem with those cars is that they do not compute problem of sacrifice or self sacrifice. For example. Situation, split second decision. You want to hit fuel truck or get off the bridge or you want to hit kid on crossing or the wall, or you want to hit the tree on driver side or passenger side. When adrenaline rushes there is always time to do something just before crash. In studies of accidents it was discovered that such "moral" decisions are being made. I would like to see source code of such sacrifice program or they just use code from Grand Theft Auto. May be autonomous cars are safer and making less mistakes but I am sure we gonna see lawsuits when victims discover that what happened was written as scenario at some XXXXX line of code. What if computer chooses to run over one man to save three in the car? What victims family would have to say about it. We are far from such situations I guess, but now is time for questions before googlers decide for us.
gkam
3 / 5 (2) May 11, 2015
PM, you are correct, there will be much to design around, (and litigate). But when they start to talk to each other, the cars will be finally safer than they are today, . . . theoretically.

I worry about EM effects, and how to code all the stuff we have going on with all our electronics without interference between systems. Then,what about vandals? We have them here and everywhere, folk with problems they want to take out on others?

When we invent new tools, we sometimes forget others can find different uses for them. The more powerful the tool, the more damage done.
bid
4.7 / 5 (3) May 11, 2015
What if computer chooses to run over one man to save three in the car? What victims family would have to say about it. We are far from such situations I guess, but now is time for questions before googlers decide for us.

I understand what you want to say, but killing the three to save the one doesn't sound like a better choice. It could be a choice a human makes though. A computer should always be able to make the "best" decision based on available data. He doesn't kill three just because his subconscious tells him to hit something in the shape of a box, instead of something in the shape of a human. Computers are actually better at split second decisions, since they don't have to override the neo-cortex to save time, like humans do. (Of course that's only true if there are enough available computing resources - but you can always throw more transistors at a computing problem, while the size of our brains is finite due to the size of our skulls and available energy supply.)
syndicate_51
5 / 5 (3) May 11, 2015
Benefits of Automatic vehicles, massive drop in accident rates, faster reaction times than human, period, by integration of automated vehicles the road system can be used far, far, far, (I can't say far enough times) more efficiently, humans make way more errors than computers, emotions do not get in the way of driving (I would highlight this if I could), computers don't impair their driving with alcohol or other mind altering substances, analyze the driving conditions with far more information available to the automated car than any human driver.

Hacking, there are far better systems to hack existing now that could have a far wider effect than trying to hack a car, specifically if the programming is compartmentalized in each vehicle.

Loss of control. 1st the car could sense and read the road far better than any human. Much less occurrence than loss of control. Also giving that the road can be "seen" it could have a default protocol to just pull over if there is a problem.
dogbert
1 / 5 (4) May 11, 2015
I don't know why people assume a machine can drive a vehicle better than a human being. No machine on earth has the reasoning ability of normal person. People solve problems of mass, acceleration and momentum effortlessly every day.
syndicate_51
5 / 5 (3) May 11, 2015
I don't know why people assume a machine can drive a vehicle better than a human being. No machine on earth has the reasoning ability of normal person.


And machines, with 360 degrees of situation awareness and that can incorporate data that no human can recognize while they drive do that, but far faster.

In driving situations that require split second actions, the process of reasoning con take far too long.
thomowen20
3.8 / 5 (6) May 11, 2015
Spot on Syndicate!
-----------------------------------------------------------------------------------------------------------------------------
As to the damned naysayers, read this twice of you have to:

"Google and Delphi said their cars were not at fault in any accidents, which the companies said were minor."

Now let the people that want to ACTUALLY solve problems and make the world a better and safer place carry on with their work. If you have enough time to decry a technology as impossible take a little more time to see HOW a thing can be done. Who knows, you might find your calling!
syndicate_51
5 / 5 (3) May 11, 2015
"As to the damned naysayers."

1903 first powered flight.

1914, 11 years later we were flying them in combat.

Naysayers, history is not on your side.
Thnder
1 / 5 (2) May 11, 2015
I think a centralized system is a bad idea, since you introduce a possible single point of failure. It may be a better idea, if every car is able to communicate its "perception" and immediate "plans" to all cars in the vicinity.

I would be more open to a combination of a central controller with the big picture, able to step in and the individuals with a localized picture able to to respond to the controller as needed. Seems better suited to high volume situations imop. Air traffic control comes to mind, port a version suited to ground traffic.
syndicate_51
5 / 5 (1) May 11, 2015
"the company's ultimate goal is a car without a steering wheel or pedals."

Well this right here is a HUGE fucking problem.....

Even for minor issues, like it breaks down, and you want to push it off the road?

What then?

The people running Google should be punched in the face, just for being fucking idiots.



Vehicles with electronic transmissions have what is called shift lock release. It's on current vehicles today so there is no reason why it couldn't be applied to the autonomous cars.

Shift Lock Release lets you bypass the transmission feature that prevents you from say, blowing up your transmission by a poor shifting choice or what not. Thus pushing a malfunctioning vehicle out of the way.

In the speak of autonomous entities everywhere, objection nullified.
krundoloss
5 / 5 (2) May 11, 2015
Perhaps an ad-hoc approach might be better than a centralized control approach. If we can get cars to follow simple rules and flow around one another like blood cells, always the perfect distance from one another, always calculating the proper stopping distance and following distance, etc.

I think the "Liability Problem" is a more significant barrier than any of the technical problems.
SciTechdude
3.5 / 5 (2) May 11, 2015
I've had a vision for a long time of driving your car to the main road, the putting it into "train mode" where essentially all the cars communicate with each other and form an automatic queue. You get to the road, you signal to your car where you want it to go, take your hands off the wheel, and it finds you a slot in traffic in a few moments. (Nobody is crowding you out, nobody is in more if a hurry than others, too distracted to let you merge, etc). All the traffic within a distance of X miles knows of your car, and everything adjusts it's speed ever so slightly, your opening appears, your car slots you into traffic, and you take a nap. When you get near your destination, the car chimes or w/e a lot until you wake up and take the wheel, or it slides you off to the side of the road into a parked position until you are capable of continuing onward. Then you drive along the side streets to your destination, guided of course by GPS and radar etc.
bid
not rated yet May 11, 2015
...an ad-hoc approach might be better than a centralized control approach. If we can get cars to follow simple rules and flow around one another ...

That's what I meant (but didn't communicate properly ;) ). In AI / robotics it's known as "swarm behavior" and we have the needed algorithms down. In a flock of birds / swarm of insects no animal knows what the swarm is doing. Every animal only cares about its immediate neighbors and reacts with few and simple reactions to changes in the neighbors behavior. But from a "higher point of view" (the whole swarm), some VERY intricate reactions to outside stimuli emerge. That's because the swarm as a whole acts as an easy version of a neural network. In our brains every neuron does only very simple reactions and communicates only to the direct neighbors. The great feats like intelligence only emerge, when you look at the network as a whole. In that sense, the more cars are working together the more intelligent the whole road becomes. ;)
bid
5 / 5 (1) May 11, 2015
...a combination of a central controller with the big picture, able to step in and the individuals with a localized picture able to to respond to the controller...

Interesting thought. I think the "central controller" shouldn't be able to step in though. I have worked with networks & security for too long to trust any system with a central control instance. ;) But if the controller collects data from all the cars and feeds a simplified overview of the whole vicinity back to each car, it could be great for safety and also take some load from the individual computers. It would be something like traffic news on steroids.

You made me think about the need to switch off runaway cars externally though. How about roads that are able to stop all traffic on it, so the police could just switch of the road and walk up to the car in question. Less security problems there and technically / financially very easy to do (just reuse the coils in the road at traffic lights).
zoljah
not rated yet May 11, 2015
... all the cars communicate with each other and form an automatic queue ...


Then again, the system needs to be 100% secure to any malfunction/hacker based disasters -> not possible without a new type of encryption/computers which i personally rather not like to see to be common by now.. ask Simon Phoenix why ;)
Steve 200mph Cruiz
not rated yet May 12, 2015
Excluding road conditions, all accidents are simply a matter of human error. There are no true accidents, there was a cause and effect that led to the situation were that accident could occur.
I think you could actually go faster of everyone was integrated into a system

But what I think is actually important is what applications this technology could have when combined with drones. Maybe we will finally be able to have our flying cars in the future.
PhysicsMatter
5 / 5 (1) May 12, 2015
I agree that self driving -cars may be safer but we should ask why. In most case because they will restrict our movements when and where we are going and why, one way or another. But if we override them, again we would have created additional risk which may or may not be handled properly by car's software. I am also concerned with people's attitudes that these machines are infallible. I think they may make mistakes as long as they are designed by humans. One day you get to your car and tell it got to GoGo bar as a response you would hear " enough booze and women for this week, its bad for your liver and your pocket" in such case accident would be waiting to happen and not covered by insurance.
antialias_physorg
5 / 5 (1) May 12, 2015
In most case because they will restrict our movements when and where we are going and why, one way or another.

Could you provide such a restriction? Unless you mean "they won't let us go off-road" - but that is not a usual scenario for normal drivers, either. In any case off-road driving would be one of those instances where it would be perfectly fine to override self driving software (as all other issues of road safety/rules go out the window when you do that).
I am also concerned with people's attitudes that these machines are infallible.

You put your trust in a lot of autopilot features in your life. Planes, trains, some subways, ... why should cars be any different?
Eikka
5 / 5 (2) May 13, 2015
How do you know without doing any tests? Maybe the computer would have come up with another, maybe even a better way.


The point is that with -today's- level of technology, it won't.

Today's anti-collision systems would probably have slammed the brakes in a millisecond and caused an immediate loss of steering. While the impact force would have been less, the car would hit the back or front of the other vehicle and send it spinning into other traffic or crush the doors in and injure passengers.

You put your trust in a lot of autopilot features in your life. Planes, trains, some subways, ... why should cars be any different?


Because in an airplane there's a pilot to take over and miles and miles of open sky around. Trains and subways run on tracks and they too have "pilots", whereas cars run on roads and the autopilots aren't nearly smart enough to deal with it.
adam_russell_9615
not rated yet May 16, 2015
Excluding road conditions, all accidents are simply a matter of human error.


I would say excluding equipment failure, medical failure, and trees and things falling on the car. Road conditions are the responsibility of the driver so that would fall under human error as well.
Bob Osaka
not rated yet May 18, 2015
Combine autonomous cars with a smart road system and raised-lane interstate linear motor propulsion and we might have the beginnings of an infrastructure for the twenty second century the whole world would emulate.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.