Self-Driving Cars Have Twice The Accidents, But Only Because Humans Aren't Used To Vehicles Following The Rules
from the I'm-sorry-you-hit-me,-Dave dept
When Google discusses its latest self-driving car statistics (provided monthly at the company's website), the company is quick to highlight that with two million miles of autonomous and manual driving combined, the company's self-driving cars have only been involved in 17 minor accidents, none of them technically the fault of Google. Or, more specifically, these accidents almost always involve Google's cars being rear ended by human drivers. But what Google's updates usually don't discuss is the fact that quite often, self-driving cars are being rear ended because they're being too cautious and not human enough.And that's proven to be one of the key obstacles in programming self-driving cars: getting them to drive more like flawed humans. That is, occasionally aggressive when necessary, and sometimes flexible when it comes to the rules. That's at least been the finding of the General Motors-Carnegie Mellon Autonomous Driving Collaborative Research Lab, which says getting self-driving cars onto the highway can still be a challenge:
"Last year, Rajkumar offered test drives to members of Congress in his lab’s self-driving Cadillac SRX sport utility vehicle. The Caddy performed perfectly, except when it had to merge onto I-395 South and swing across three lanes of traffic in 150 yards (137 meters) to head toward the Pentagon. The car’s cameras and laser sensors detected traffic in a 360-degree view but didn’t know how to trust that drivers would make room in the ceaseless flow, so the human minder had to take control to complete the maneuver."And while Google may crow that none of the accidents their cars get into are technically Google's fault, accident rates for self-driving cars are still twice that of traditional vehicles, thanks in part to humans not being used to a vehicle that fully adheres to the rules:
"Turns out, though, their accident rates are twice as high as for regular cars, according to a study by the University of Michigan’s Transportation Research Institute in Ann Arbor, Michigan. Driverless vehicles have never been at fault, the study found: They’re usually hit from behind in slow-speed crashes by inattentive or aggressive humans unaccustomed to machine motorists that always follow the rules and proceed with caution."But with a sometimes-technophpobic public quick to cry foul over the slightest self-driving car mishap, car programmers are proceeding cautiously when it comes to programming in an extra dose of rush-hour aggression. And regulators are being even more cautious still. California last week proposed new regulations that would require that all self-driving cars have full working human controls and a driver in the driver's seat at all times, ready to take control (which should ultimately do a wonderful job of -- pushing the self-driving car industry to other states like Texas).
The self-driving car future is coming up quickly whether car AI or self-driving auto philosophical dilemmas (should a car be programmed to kill the driver if it will save a dozen school children?) are settled or not. Google and Ford will announce a new joint venture at CES that may accelerate self-driving vehicle construction. And with 33,000 annual fatalities caused by highway-bound humans each year, it still seems likely that, overly-cautious rear enders aside, an automated auto industry will still likely save significant lives over the long haul.
Thank you for reading this Techdirt post. With so many things competing for everyone’s attention these days, we really appreciate you giving us your time. We work hard every day to put quality content out there for our community.
Techdirt is one of the few remaining truly independent media outlets. We do not have a giant corporation behind us, and we rely heavily on our community to support us, in an age when advertisers are increasingly uninterested in sponsoring small, independent sites — especially a site like ours that is unwilling to pull punches in its reporting and analysis.
While other websites have resorted to paywalls, registration requirements, and increasingly annoying/intrusive advertising, we have always kept Techdirt open and available to anyone. But in order to continue doing so, we need your support. We offer a variety of ways for our readers to support us, from direct donations to special subscriptions and cool merchandise — and every little bit helps. Thank you.
–The Techdirt Team
Filed Under: accidents, autonomous vehicles, cars, driving, rules, self driving cars
Companies: google
Reader Comments
Subscribe: RSS
View by: Time | Thread
Sign me up when KITT is ready
[ link to this | view in thread ]
Zoned Out
[ link to this | view in thread ]
[ link to this | view in thread ]
[ link to this | view in thread ]
Re: Sign me up when KITT is ready
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Figures.
Let's simply say that under ideal conditions, with weather, terrain, traffic, mechanical failures, and serious law-breaking not involved, highly paid engineers have managed to achieve an accident rate only twice the baseline set by "typical" drivers with their "typical" education, training, distractions, and vehicle maintenance. Spiffy.
[ link to this | view in thread ]
Re:
On the other hand, the linked article notes that a Google car was warned by police for going 24 in a busy 35 zone. Now, crashes in that sort of situation are not going to be listed as Google's fault... but at some point, you're increasing rather than decreasing the risk of an accident by driving so far below the posted limit for no discernible reason.
[ link to this | view in thread ]
Re: Re:
I agree, from an insurance standpoint, that the driver hitting the rule-adherent would be at fault. But you're describing something other than what I was describing. I pay attention and don't rear end people and I do expect them to stop at stop signs. I don't hit them because they follow the rules. I don't hit them at all because I don't get into accidents. I'm an alert driver and I know when to drive cautiously.
But when you have 5 cars lined up behind you all wanting to go 10 mph more than the speed limit in a zone that is poorly rated for speed by the local government planners, the rule-adherent becomes the annoying anomaly that may in fact cause an accident due to their deviation from the average, albeit rule-bending behavior of most drivers.
[ link to this | view in thread ]
[ link to this | view in thread ]
a knock at the door was georgette, arriving slightly late. as she stepped in the gang all welcomed her, and she responded to the men, 'oh, don't get up.' whereupon, of course, all the guys dutifully arose.
in her plaintive voice: 'why do they always do that?'
when machines understand why that was amusing, they'll be ready to share the road with us.
[ link to this | view in thread ]
Drive(n)
[ link to this | view in thread ]
[ link to this | view in thread ]
Yeah, it's real "social" to go crashing into other drivers, pedestrians, trees and what not... to the tune of tens of thousands of deaths each year. It's real "social" for butt-holes who drive around with their radios/stereos blasting loud enough for everyone within a mile or more in any direction to hear.
Some people really shouldn't be allowed into "society"... at all.
[ link to this | view in thread ]
Cars or laws, something needs to be tweaked
[ link to this | view in thread ]
where did i say i saw it?
[ link to this | view in thread ]
I could probably get hit from the rear every day
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Re: Figures.
[ link to this | view in thread ]
Re: Sign me up when KITT is ready
[ link to this | view in thread ]
Re: Re:
[ link to this | view in thread ]
As I see it right now, there are three major blocks to the widespread adoption of self driving vehicles.
1. The question of who is legally responsible for the safe operation of the vehicle is currently up in the air. This represents a huge potential liability, particularly when you throw in aftermarket parts, repairs, and modifications. This will require law (preferably legislation, less preferable regulation or litigation) to determine whom is legally responsible for a self driving vehicle.
2. Self driving vehicles must be interoperable by design. The major companies working on them are well known for interoperating inside their own domain, but do not tend to play well with others (I'm looking at you, apple and google). This will also likely require a lead integrator outside of the companies developing the vehicles, either in the form of an industry alliance or a government entity.
3. The technology will require approximately 10-20 years to penetrate to the used market after self driving vehicles become widespread in new build cars. Until there is a healthy used market, the adoption rate will lag what is necessary to pull the trigger on full autonomous driving on the nations highways.
This is why I think that the most likely technology to be fielded is some sort of semi-automatic driving aid. This would be similar to an autopilot in a aircraft. it would be up to the driver to make lane changes, turns, and merges, but the vehicle could keep in the lane and maintain distance. I also think that the more likely field for adoption is on the highway/interstate, because it is a simpler problem than trying to automate in-city driving. Human drivers can already effectively do in city driving, and the majority of the benefit in preventing fatalities would occur with high speed driving on the highways anyway.
[ link to this | view in thread ]
Re: Figures.
I'm curious as to what you base this on.
"highly paid engineers have managed to achieve an accident rate only twice the baseline set by "typical" drivers with their "typical" education, training, distractions, and vehicle maintenance. Spiffy."
On a brand new technology that's still very much into its testing phase and not claimed to ready for mass market. Yet, the biggest problem you have seems to be that the cars don't act like human beings.
Not bad. This far into the early life of the motorcar, people were still arguing about steam vs. petrol and whether or not to have men with flags warning people about them, or even ban them completely to not risk scaring the horses. It wasn't long until those arguments looked dated and the technology improved exponentially. I look forward to future developments from all the companies working on this.
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
Re:
I agree that the legal situation will be messy to begin with, but that's true of any new technology. Standards, liability, etc. are part of the acceptance process of anything not previously encountered by current law.
"Human drivers can already effectively do in city driving"
You must have visited different cities to the ones I've been to.
[ link to this | view in thread ]
bad road design
In the cities I regularly drive there are a few sections of road like that - only visitors tend to attempt the move across many lanes in a few yards manouveres, the locals use different routes around the road network so they avoid those dangerous areas of road "design" and have far safer junction transits involving less lane changes or a far longer distance to change lanes
BTW all the (I'm guessing US comments) about often exceeding speed limits, think yourselves lucky you are not in the UK where there are huge numbers of vehicle speed cameras & ignoring speed limits can rapidly mean loss of driving licence via gatso.
[ link to this | view in thread ]
Need better testing.
Drivers Ed -
Day 1. This is a video of a human head under a tire.
Day 2. This is a video of two kids burning alive in a ditch.
Day 3. This is a video of drunk driver under a semi.
Day 4. This is a video of a torso thrown from the window.
Day 5. Congrats, you finished watching faces of death!!
Here's your drivers license, bro! Now get the fuck out, you reek of sawdust and vomit.
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Figures.
[ link to this | view in thread ]
Re: Re: Re: Figures.
[ link to this | view in thread ]
Control
Are we now going to go beyond reporting one's location to giving the police the authority to determine if we can go or not by giving the police the authority to determine if the car will work and if it does the rout it will take?
[ link to this | view in thread ]
Question
What happens when lane markings haven't been repainted for ten years and they're almost invisible?
What happens when a stretch of road has been resurfaced but the lines haven't been painted on yet?
How well do these cars cope with motorway contraflows?
Can they do roundabouts?
How good are they at detecting potholes and speed bumps?
What do they do when traffic lights are defective?
What happens if there's been flooding and the road surface is under, say, 6 inches of water? It would be passable by a driver in a normal car, how about these?
Snow? Can it do snow?
How does it deal with cyclists? Especially filtering / lane splitting, running red lights, etc?
Sorry if this sounds cynical, but all of these (and more) are real-world problems that autonomous vehicles will need to be able to cope with if they are going to become standard over here on Surveillance Island.
[ link to this | view in thread ]
Re:
Well, we already put inanimate property on trial and give corporations artificial person-hood. So, why not make the car itself responsible? The precedents have already been set.
[ link to this | view in thread ]
Re: Question
What happens when...
As I understand it, Google cars only self-drive on specially pre-mapped roadways. Kind of like some kind of amusement ride and not really useful for general transportation.
[ link to this | view in thread ]
So she started taking turns much faster, speeding up on yellow lights, doing rolling stops, and so on, and suddenly the problem went away.
Also reminds me of the day we got a metro rail above ground. 1st day, like 3 hits, all side hits from people who didnt look where they were going.
[ link to this | view in thread ]
Too damn easy to get a drivers license
[ link to this | view in thread ]
Re: Re: Sign me up when KITT is ready
[ link to this | view in thread ]
Re: Question
Also, the road markings are probably irrelevant. If you search for documentation, what's available certainly shows that they have multiple systems working on positioning, etc., While public roads in California, et al., may be somewhat straighter than in the UK in many areas, they're certainly not better maintained in my experience.
In fact, the company seem to have been over-cautious with other tests (such as setting the speed to 25MPH), so they may not even be testing on motorways. I certainly don't see them being asked to navigate the M25, Spaghetti Junction or the Magic Roundabout in a heavy rainstorm just yet, but I will guarantee they've considered roundabouts and driving on the left in designing their tests.
[ link to this | view in thread ]
Re: Zoned Out
There are two different self-driving car programs by Google (Alphabet). One of them is the one you describe, limited to 25 mph, just around Mountain View (the "bubble" cars). The other is the regular cars, which go well above 25mph. Most of the data comes from the faster cars, not the slow ones.
[ link to this | view in thread ]
Getting rear ended
[ link to this | view in thread ]
Re: bad road design
If traffic was actually moving at a reasonable speed and I only had 150 yards to get across 3 lanes and there was heavy traffic, I'd probably just shrug and slowly start getting over so that I can catch the exit after the one I've obviously already missed.
[ link to this | view in thread ]
Re: Re:
[ link to this | view in thread ]
Re: Re: Question
So, what, we can no longer do construction with a detour because the Google cars keep driving over the pylons and the humans have no controls to stop them?
[ link to this | view in thread ]
Re: Too damn easy to get a drivers license
[ link to this | view in thread ]
Re:
[ link to this | view in thread ]
when did people become perfect???
the software people write is not perfect
the machines (built by imperfect people) that utilize the software (built by imperfect people) aren't perfect
Since people, nor our machines, nor our software are perfect, there still will be accidents. Then come the hackers...
[ link to this | view in thread ]
Re: Question
When the Google car can't determine what to do, it hands things over to a human driver.
Then the programmers go in and study the situation and program a way so that it doesn't happen again - if that's possible. This is how they've been developing the car for the last 10 years - trial and error. From what I've seen, Google cars can do fine with most of the situations you've described.
[ link to this | view in thread ]
Re: when did people become perfect???
[ link to this | view in thread ]
Re: Re: Re: Question
And then there are the real cars that are navigated by software but still require a human driver. They are not restricted to pre-mapped roadways. Most car manufacturers are currently working on these types of cars.
[ link to this | view in thread ]
Re: Getting rear ended
[ link to this | view in thread ]
Re: Re: Question
When I first read California's and Nevada's regulations for self driving vehicles (Nevada is supposedly allowing testing for self driving commercial trucks) I noted for both states that a licensed driver was required to be in the vehicle. Based on some of the comments here and elsewhere I'm wondering if the ultimate goal of these self driving car experiments might be self driving cars that won't have humans in them. If that's the case how would the car react with no humans? And would there be a 'safety shutdown' that the car would stop exactly where it is, even if it's still on a highway?
[ link to this | view in thread ]
Re: Re: Sign me up when KITT is ready
[ link to this | view in thread ]
Re: Re:
-You must be the guy who empties out all the swearing jars
[ link to this | view in thread ]
It will evolve
[ link to this | view in thread ]
Re: Re: Re: Re:
If people are driving 10mph above the posted speed limit because it's the flow of traffic then when the flow of traffic drops to the speed limit, then those same people will be driving the speed limit.
[ link to this | view in thread ]
Re: Re: Re: Question
Figuring out what the likely situations are and accounting for them is exactly what Google and others have been doing for the past several million miles.
[ link to this | view in thread ]
Re: Re: Re: Question
In the near future, you'll be sitting at the wheel ready to take over.
[ link to this | view in thread ]
Re: Re: Re: Figures.
[ link to this | view in thread ]
Re: Zoned Out
I think I need to convince my local police department that she's in a self driving car...
[ link to this | view in thread ]
Re: Too damn easy to get a drivers license
That's because the alternatives to driving are so poor in the US that driving is a practical necessity in many places. So it is made so that almost everyone can get a license.
[ link to this | view in thread ]
Re: Getting rear ended
True. Speed variation, as it's called, is dangerous. Google cars don't care.
[ link to this | view in thread ]
Re: Re: Re: Re: Question
That's not exactly "self-driving", now is it?
[ link to this | view in thread ]
Re: Re: Re: Re: Figures.
[ link to this | view in thread ]
Re: Re: Getting rear ended
[ link to this | view in thread ]
Re:
If Google makes their cars follow the rules uniformly it's more dangerous due to everyone else breaking the rules. Any accidents would then be the fault of those breaking the rules but accidents will occur more frequently. If Google makes their cars break the rules occasionally it's safer but any accidents caused during which Google broke a rule to increase safety would be at least partly Google's fault by law because they broke a rule.
A partial solution is for Google cars to determine how many people are breaking the rules at any given time (though the problem is that Google can't predict whether or not someone will break a rule in the near future). If no one is breaking the rules Google should then follow the rules because that would be safer for everyone. If there are other vehicles breaking the rules Google should then determine if breaking the rules would increase safety. If it would then Google should, for the sake of safety, break the rules. The problem with that is now if there is an accident Google will now at least be partially to blame, legally, for breaking the rules. Google will be legally partially at fault if both Google and the other driver broke the rules. Google will be entirely at fault legally if Google broke the rules and the driver that Google got into an accident with didn't even if, technically speaking, Google was just trying to avoid accidents with other vehicles that are breaking the rules.
I believe the rules are partially to blame here. If so many people are breaking them perhaps it's because they are unreasonably restrictive. The rules should be such that those that break them aren't just breaking an arbitrarily restrictive rule but that they are actually doing something dangerous. Just like humans aren't perfect the humans that design the rules aren't perfect either.
Another interesting question that often pertains to going on a freeway on-ramp (and other situations) are hand gestures. In all sorts of driving situations people often communicate with hand gestures. There could be a situation where a driver is communicating with another driver. There could be a case where a passenger of one vehicle is communicating with the driver of another. There could be a case where a driver of one vehicle is discussing something with the passenger of the same vehicle and one is gesturing to the other (or both are gesturing to each other) and the gestures are unrelated to what's going on on the road. Especially when approaching an on-ramp a common gesture a driver of one vehicle could give to another vehicle is that the driver of the person on the freeway may be directing the driver trying to get onto the freeway to enter the freeway in front of them. Gestures are also useful for situations where a driver legally initially has the right of way but would like to pass that right of way onto another driver first for whatever reason (ie: at an intersection. Perhaps the driver with the initial right of way wants to make a U-turn at the intersection but would like the other driver to pass first). Also humans honk their horns at each other to warn them of potential danger. Usually other humans on the road are good at determining whether a hand gesture or car horn honk is being directed at them. Are Google cars good at this?
[ link to this | view in thread ]
Re: Re:
Usually humans are good at reading hand gestures, gesturing back, and understanding each other. Humans understand each other much better than computers. Take, for instance, speech recognition. Talking into a phone or a computer is much more difficult than talking to another person. If you are talking into your phone you must speak more slowly, distinctly, clearly, and even then you often have to make corrections. Compared with talking to another person the difference is light and day. As humans we generally get each other but computers suck at understanding humans in all sorts of situations.
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re:
[ link to this | view in thread ]
Re: Re: Re: Re: Re: Figures.
[ link to this | view in thread ]
There are victims when others can't follow the rules
http://seahorsecorral.org/lambert2.html
[ link to this | view in thread ]
[ link to this | view in thread ]
Re: Re:
I will admit, what I am thinking of is more of an 80% solution: something that we c an start building the technology base while allowing for the fully automated car to take the time it needs to fully gestate, and to allow time for the law and society to adjust to a self-driving automobile. I don't think we should go from canvas and wire to 777 (to analogize from autopilot technology on airplanes).
As for people driving drunk and general incompetence... seriously? At the worst case, we're no worse off than we are now. The only difference is that we would have a mechanism that would reduce a great deal of driving stress associated with traffic jams and highway driving, which is where the majority of traffic fatalities occur (according to the insurance institute for highway safety).
Were I dictator for a day, I would not choose to invest money in self driving cars. I would take the 80% solution and use the rest of teh money to improve the infrastructure (e.g. widening roads, adding more lanes). I would also look into algorithmically controlled traffic lights, which I think can be made to reduce in-city congestion (at least in my town, where all the damned lights seem to be perfectly out of phase with each other).
[ link to this | view in thread ]
So Install a Warning Sign
Imagine a scenario in which a drunk becomes convinced that a telephone pole is insulting him, and he breaks his hands trying to beat up the telephone pole. That is not an argument for removing telephone poles so that the drunk can be sure of only hitting people.
Safe speed is a function of traffic density. I have sometimes seen fender-benders in traffic so jammed that, going down the sidewalk, I was out-walking the cars. A few weeks ago, I saw a five-way pile-up at speeds which cannot have been much more than ten miles per hour. I heard collisions at approximately one-second intervals, and, looking up, saw a row of wrecks, twenty or thirty feet apart. My impression was that the traffic light had just turned green, and the drivers had all accelerated with such vigor as not to be able to stop which it appeared that there was nowhere to go. The fire department carried one man away to the hospital. At a certain traffic density, it may be the case that the safe speed is two miles per hour. If one were to borrow from the science of hydraulics, one might introduce the concept of Reynolds' Number as the boundary between orderly flow and chaotic flow.
The speed limit is not just made up out of someone's hat. It is based on calculations of things like stopping distances, and it is premised on the observation of the two-second rule. You can find the details in a textbook of highway engineering. Braking distance increases as the square of speed, that is elementary physics. People who insist that they can cope with traffic by going faster are really banging their heads against telephone poles. The textbook solution, of course, is to build more lanes, and build lane-dividers (for different directions, and local-long distance), overpasses, cloverleafs (sometimes stacked four or six or eight layers high), etc. However... in an urban area, one tends to run out of land. I understand that Reason Magazine has put forward a proposal to dig multiple layers of tunnels under Los Angeles, like the Boston "Big Dig," only much bigger and more elaborate, which are likely to cost something like a hundred thousand dollars per capita. Some of the tunnels would be the longest road tunnels in the world.
[ link to this | view in thread ]
Re: Question
I think what I can take from this is that in the short term we'll get normal cars with a sort of intelligent cruise control, and that truly autonomous cars are still a long way off and may even require dedicated roadways.
[ link to this | view in thread ]
The Future Is Teleportation!
[ link to this | view in thread ]
Re:
If police want a safe and sane driving environment, they have to make themselves less of a risk than the other cars.
[ link to this | view in thread ]
Re: Re: Figures.
To the point where we have 100 deaths a day
[ link to this | view in thread ]
Re: Re: Re: Figures.
But, yeah, the huge number of cars on the road aren't the reason for the large number of deaths, nor is the fact that the poor design of many American cities make owning at least one mandatory for a great many people. It's the technology, which means that we should abandon development of a new technology that will improve safety further. You go with that.
[ link to this | view in thread ]
The problem is the human drivers
Some drivers will always drive faster than the law allows, irrespective of whether the limit is appropriate or not. In order to deter those drivers from driving at dangerously fast or inefficient speeds all the time, the limits are kept lower than they need to be. Peer pressure does the rest and results in the entire traffic flow moving close to or above the prescribed limit. The more humans drive outside the laws, the more automated vehicles will be at a disadvantage.
The solution is for more humans to reliably follow the prescribed laws. At which point the prescriptions can be changed with the result of reducing the gap between humans and machines.
[ link to this | view in thread ]
Re: Getting rear ended
Fail.
[ link to this | view in thread ]
Re: Re: Getting rear ended
But you being removed from the driving situation would allow the other cars to reach their destination faster.
[ link to this | view in thread ]