Slashdot Mirror
Google's Self-Driving Cars: 300,000 Miles Logged, Not a Single Accident
An anonymous reader writes "The automated cars are slowly building a driving record that's better than that of your average American. From the article: 'Ever since Google began designing its self-driving cars, they've wanted to build cars that go beyond the capabilities of human-piloted vehicles, cars that are much, much safer. When Sebastian Thrun announced the project in 2010, he wrote, "According to the World Health Organization, more than 1.2 million lives are lost every year in road traffic accidents. We believe our technology has the potential to cut that number, perhaps by as much as half."
New data indicate that Google's on the right path. Earlier this week the company announced that the self-driving cars have now logged some 300,000 miles and "there hasn't been a single accident under computer control." (The New York Times did note in a 2010 article that a self-driving car was rear-ended while stopped at a traffic light, so Google must not be counting the incidents that were the fault of flawed humans.)'"
It's hard to imagine being found at-fault when you are stopped and rear-ended.
There's no shame in being involved in an accident if it's not your fault.
We trust others all around us every day to avoid smashing into us. Even the best drivers get hit.
It is indeed an impressive statistic about the number of accidents by the self-driving car of Google. This does prove that their decision making algorithms are good.
However, comparison to humans is probably not fair. Human mind is more prone to giving in to temptation. Exceeding speed limits, violating lane changing rules once in a while to get ahead, talking while driving, texting while driving, getting distracted by the hot chick/dude in the car in the next lane are all errors that humans would routinely make. Some of them would lead to accidents where the erring driver suffers an accident. Some lead to an innocent driver suffering due to the errors of others. It is the latter condition where the Self-Driving car's algorithms appear good --- handling exceptions generated by human drivers, pedestrians and traffic.
We don't know if the Google car could have avoided it. I was in a similar situation one time, and happened to catch a glance of the idiot in my rear view. I cut out ahead of the adjacent lane into the empty crosswalk. The idiot screeched to a stop in what was previously my lane.
There isn't always empty space; but if there is then the Google program should recognize it as available for evasive maneuvers. The Google car will not have a heart to go pitter-patter like mine did; nor a father who turned to me and said, "you're a good driver".
I also read that the automatic drive wasn't able to cope with simple situations such as another car coming from the opposite direction in a narrow street, requiring manual intervention. So alongside the triumphant tones, they should also explain how much these cars are really self driving; most car accidents don't occur in straight motorways.
You are creating a straw man there, 99% of similar situations with human drivers would either have not noticed the exit or not reacted in time.
Additionally you likely broke the law doing what you did and if you caused an accident or ran over a pedestrian because of it you would have been 100% at fault, whereas being shunted by the guy behind you lands 100% of the resposibility on him (unless you stopped too close to a car in front of you).
I would put money on your driving record being way worse than 300k miles accident free. The actual pouplation-wide average is a LOT higher than that, and you are asking for us to give up reducing that number because we can't reduce it to 0.
That's like people saying "Don't build gas power plants to replace coal plants because they still emit CO2", sure it's not perfect but at least it BETTER.
The moment even one accident does occur, no matter how mild the consequences or much more unlikely the circumstances compared to a human driver, hordes upon hordes of American luddites will man the lines to do their civic duty to shit upon the idea of cars that drive themselves.
Mind you, this is being said by an American who owns a US made car.
Support the EFF and Creative Commons. The war is coming, and they're supporting you...
I don't think he's saying we should give up on reducing that number. I think he's saying it may technically have been avoidable in that case. It's not meant to be dismissive. It's meant to be food for thought.
But if everyone went around in automated cars, the point is there would be no "idiot" in your rear view because he also would have been in an automated car; one which would have stopped in time.
Seven puppies were harmed during the making of this post.
Pretty much this.
Google deliberately avoids the more challenging situations, and a LOT of those miles are highway.
There's a reason insurance rates for someone living in a small town in the country are lower. Right now, google is pretty much "that guy".
That's not to disparage what google has accomplished, but its premature to compare it to the safety record of a downtown urban commuter; driving through rush hour traffic to and from work in a major city daily.
I'm more wondering what it would be like for the driver who actually rear ended a robotic vehicle
I imagine that you exchange details with the human in charge, with the full knowledge that there will be a complete 360 degree video of the accident with measurements of speed of both vehicles.
with the full knowledge that there will be a complete 360 degree video of the accident with measurements of speed of both vehicles.
Only if YOU caused the accident. It's a pretty safe bet that if a glitch in their programming caused the accident, there'll be a tragic loss of data... :-)
He put his boots up on the table and made a face. "The sig," he smirked. "You can waste your life in search of the sig."
How many times has the Google-mobile pulled into and out of parking spaces at busy malls? Frankly, that's where I've had my accidents.
But we will. That's the entire point... Computer driven cars are better than humans, on average, and in an infrastructure re-modelled to suit such cars they are close to infallible compared to human drivers. We are struggling right now to get a computer to navigate the human infrastructure, but once this sort of machine has saturated the market the infrastructure WILL change.
You have to remember you don't live in the world of yesterday, you live in the world of now. The world of now has a very special aspect to it... what we choose to do, changes the world of tomorrow.
Don't arbitrarily limit tomorrow based on the world we had yesterday.
What about cars with no passengers though? Say there is no parking near your work so you send the car home, then tell it to come get you at 5PM. On the way it has an accident. There will have to be some kind of system in place for notifying the owner and allowing the other person involved in the collision to speak with them. Even if it is just a notice placed somewhere on the car it will have to be standardized.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
Okay, unexpected things that doesn't include hitting or being hit by things... Hmm.
At one point I was driving down the road, when the road disappeared.
Sure, it didn't literally disappear, it just so happened that it re-located about 30 feet downwards in an instant due to an unfortunate passionate meeting between a boat and a support pillar holding the bridge I was about to cross up. I guess they really liked each other, or whatever, but all the same I was definitely surprised, there was nothing for me to hit but rock bottom and by the time any computer would have seen that it would have been too late, and nothing was about to hit me.
It's possible their car could handle this situation, but that depends largely on how perfectly they can detect every inch of road. From what I've heard they go mostly by road markings and such, which means that in such a situation they'd have to switch to manual control. Hopefully in time to avoid a nosedive into the amorous steel behemoths below.
Children running into the road is childs play - a clear collision avoidance. How about children jumping over you? There I was driving down a narrow "alleyway" between two walls/brick fences that were just barely high enough that I couldn't see over them, and all of a sudden a bunch of teens on bikes fly over the road over and infront of me. I slowed down, and that was pretty lucky because the last guy didn't make it. Granted that's collision avoiding, but would google's car have been slowing down for things flying above it? Can it tell that this is kids on bikes that might crash, or a ball that might be followed (if it's above you and thus not in your direct path), rather than a dove or flying squirrel with a death wish and a meth addiction?
I could go on, but the point I'm trying to make is simple - being a good driver is not just about not bashing into things and avoiding things bashing into you, it's about avoiding putting yourself in a position where you can't stop. It's about avoiding the scene of the potential accident completely, not about behaving competently once you are already there.
I think you're missing the point, it's not about objective skill, but rather the fact that 80% think they are better than average. In reality 50% is better than average, 50% is worse than average. That means 30% of people at the very least, overestimate their skills. It's called illusory superiority, and you can check it out on wikipedia, it's a basic cognitivie bias. It even quoted the studies I referred to.
"For driving skill, 93% of the US sample and 69% of the Swedish sample put themselves in the top 50% (above the median). For safety, 88% of the US group and 77% of the Swedish sample put themselves in the top 50%"
-- http://en.wikipedia.org/wiki/Illusory_superiority#Driving_ability
You can also read the article on the Dunning-Kruger effect.
"The Dunning–Kruger effect is a cognitive bias in which unskilled individuals suffer from illusory superiority, mistakenly rating their ability much higher than average. This bias is attributed to a metacognitive inability of the unskilled to recognize their mistakes."
-- http://en.wikipedia.org/wiki/Dunning%E2%80%93Kruger_effect
My definition of a good driver is one that doesn't get into accidents, and that can handle an unexpected situation quickly and correctly. Whether they obey traffic laws is more about whether they are a good citizen or not, and that's a different argument, however staying within the traffic laws (i.e. lowering speed around schools) prevent accidents, and is as such a trait of a good driver.
Whether you are indeed a good driver is irrelevant to the fact that a lot of people judge themselves as to be better than they are. It's after all not the individual that matters when we're talking about the group as a whole.
I would really question how these cars function in rush hour in a big city. Driving there is sketchy at best and in order to merge into another lane you sometimes literally need to start heading into the other lane even with traffic that isn't helping you merge. How would a car like this function bumper to bumper?
In the future a bunch of these could eliminate traffic jams, but that isn't going to be a case for a long time.
What if you're in the middle of nowhere and it is simply some debris that rolls over in your path is detected as obstacle. Now you're on the highway doing 65 and your car suddenly slows down/stops for no reason and you get rear-ended.
That's the fault of the idiot tailgating you. If you're driving close enough to the car in front of you that you can't come to a full stop if he does, you're too damned close!
Also, where can I find this awesome magical computer vision & depth perception equipment that can constantly scan 300 or so feet (average stopping distance at highway speeds) and accurately (well lets say >90% confidence) identify any random object in the real world.
The hardware is available from most robot suppliers (being just a few LIDAR units and high-res cameras) and the software can be found within Google's well-secured vaults, being most likely an edge-detection algorithm applied to the pictures, then those shapes projected onto the 3D map from the LIDAR to identify objects, then a pattern-recognition engine to identify the object and its risk. If the object can't be identified with high confidence, assume it's dangerous and slow down until a more accurate assessment can be made.
You do not have a moral or legal right to do absolutely anything you want.