SEO Speedwagon
October 9, 2010 3:40 PM   Subscribe

Anyone driving the twists of Highway 1 between San Francisco and Los Angeles recently may have glimpsed a Toyota Prius with a curious funnel-like cylinder on the roof. Harder to notice was that the person at the wheel was not actually driving. SLNYT + video.
posted by chavenet (115 comments total) 14 users marked this as a favorite
 
Here's a talk from Sebastian Thrun, "43-year-old director of the Stanford Artificial Intelligence Laboratory, a Google engineer and the co-inventor of the Street View mapping service", on Winning the DARPA Grand Challenge, in case you haven't seen it. I didn't realize at the time that he was pitching himself in process of giving that lecture, but here we are with robots that hate cyclists who run red lights just as much as the rest of us.
posted by pwnguin at 4:05 PM on October 9, 2010 [3 favorites]


This is one of Larry's pet projects. Now we know what he spends his time on, when he's not riding around in one of the Google party planes.
posted by killdevil at 4:19 PM on October 9, 2010


Who goes to jail when ones of these hits and kills a cyclist?
posted by jrockway at 4:30 PM on October 9, 2010 [4 favorites]


Will there be a Neverending Turn Signal add on for old women?

Just kidding, ladies.
posted by JaredSeth at 4:32 PM on October 9, 2010 [3 favorites]


The same person who goes to jail when someone's brakes fail and kills a cyclist.
posted by zabuni at 4:33 PM on October 9, 2010 [1 favorite]


jrockway: "Who goes to jail when ones of these hits and kills a cyclist?"

There's always a licensed driver behind the wheel who would hopefully prevent such a thing but would be the official operator of the vehicle if there where an accident.
posted by octothorpe at 4:36 PM on October 9, 2010


Page 2 of the article says this:

"And in the event of an accident, who would be liable — the person behind the wheel or the maker of the software?

“The technology is ahead of the law in many areas,” said Bernard Lu, senior staff counsel for the California Department of Motor Vehicles. “If you look at the vehicle code, there are dozens of laws pertaining to the driver of a vehicle, and they all presume to have a human being operating the vehicle.”
posted by Houstonian at 4:40 PM on October 9, 2010


"Who goes to jail when ones of these hits and kills a cyclist?"

Anti-lock brakes, power steering, cruise control, etc...

What if it had a drive by wire system and you steered with a joystick?

There's no doubt in my mind that we aren't far off from a day where computer drivers perform better than human ones.

Some interesting political side effects of this -- would you have to have a license to babysit one of these cars? If most people aren't driving -- what about drunk driving laws? Drinking age laws? Drug laws? All those intrusive police state behaviors in the name of 'protecting drivers'?
posted by empath at 4:44 PM on October 9, 2010 [2 favorites]


Who goes to jail when ones of these hits and kills a cyclist?

Machines are more accurate and far less prone to error than humans. Once the software and sensory equipment is sufficiently advanced, a computer-driven car will be far more safe than a human-driven one.

All modern large aircraft are flown by computer control systems from takeoff to landing, except in certain situations. I don't know if there's data to support my hypothesis, but I'll bet the frequency of crashes upon takeoff or landing has decreased relative to the amount of air traffic since this has been the case. Otherwise, I don't think the airlines would be using it. They even go as far to have multiple onboard systems designed by different engineering teams to double check the results on different hardware platforms, giving an extra layer of redundancy.

And on a car, you're not dealing with cross winds or 3 axis navigation, just avoiding vehicles and objects larger than cats within your proximity. The hard part is not getting the computer to press the brakes or steer, but to sense when it needs to do that. So the navigation bit is actually harder than an aircraft in one way, but control of the vehicle is vastly easier.

Anyway, I really think this technology, along with improvements in electric battery technology, will change life in the next thirty years. Once all of these vehicles are self-piloting, and can be tracked via GPS, you can greatly reduce the numbers of cars on the road with just-in-time pickups and deliveries of humans. Using your phone to request a pickup that routinely arrives within 5 minutes will eliminate the fear of not owning a car for most of the population. This will not only be better for the immediate and worldwide environment, but will improve quality of life since there will be less traffic, fewer accidents, and less maintenance to worry about.

When you think about it, the moral hazard will probably be on the cyclist's side, knowing that they can dodge in and out of auto-pilot traffic without fear of getting hit. But hopefully by then we'll have enough money left over from the smaller road budget to create bike-only streets and bridges.
posted by notion at 4:57 PM on October 9, 2010 [14 favorites]


It's not like human beings are such great drivers to begin with and seem to have no trouble running over cyclists on their own. I doubt that it'll be too long before these computer systems are consistently better than the average human driver who's too busy putting on mascara, texting or eating McNuggets to pay attention to the road.
posted by octothorpe at 5:01 PM on October 9, 2010


Driving up Highway 1 'tween SF & LA?!? I can see that highway from my front door! If I'd known it was coming I'd have thumbed a ride!
posted by oneswellfoop at 5:02 PM on October 9, 2010


If these become ubiquitous, I assume that someday we'll all forget how to drive.
posted by popechunk at 5:02 PM on October 9, 2010 [2 favorites]


The previous most awesome robocar drove autonomously on public Autobahns in 1995.
posted by domnit at 5:02 PM on October 9, 2010


When he returned to automated “cruise” mode, the car gave a little “whir” meant to evoke going into warp drive on “Star Trek”

NERD ALERT.
posted by We had a deal, Kyle at 5:06 PM on October 9, 2010 [4 favorites]


Another thing that's interesting about this is how many of the problems such as determining speed limits, knowing where stop signs, intersections, etc are are solved in ways which have nothing to do with mimicking the ways that human beings do it. Traditionally AI programmers tried to do these sorts of things with complex systems designed to create models from visual input, biased around the way humans are designed.

How do you know where an intersection is? Well, first you have to be able to pick out the road from the visual information, taking into account varying lighting conditions and materials, then you need to be able to see when there's a break in the road, etc, etc. Solving these kinds of problems in real time reliably are still many years away, as far as I know.

But, you can also just upload a map and use GPS. In fact, with streetview data, you can upload a 3 dimensional model of every street in the country and just look for unexpected differences and react accordingly.

Humans can't be programmed with perfect recall of every road, intersection and speed limit in the whole world, but computers can. Human beings can't have a sense of their location in the world to the nearest meter, but computers can. Human beings can't optimize routes based on current traffic conditions miles away, but computers can.
posted by empath at 5:07 PM on October 9, 2010 [5 favorites]


When you think about it, the moral hazard will probably be on the cyclist's side, knowing that they can dodge in and out of auto-pilot traffic without fear of getting hit.

I think the moral hazard will remain firmly on the side of the people in one tonne plus hunks of metal, but protected by airbags, seatbelts, and crumple zones, rather than on the side of cyclists. Look at it this way: who is better insulated from the consequences of a crash - the person in a car or a the person on a bike?
posted by ssg at 5:10 PM on October 9, 2010 [1 favorite]


If you have two links, it's not SLanything.
posted by grouse at 5:12 PM on October 9, 2010


jrockway: "Who goes to jail when ones of these hits and kills a cyclist"

<bitter cyclist>Pffft. People don't go to jail for killing cyclist now.</bitter cyclist>
posted by chairface at 5:15 PM on October 9, 2010 [17 favorites]


Who goes to jail when ones of these hits and kills a cyclist?

The new Google Robo-Cyclist will be able to kill itself much more efficiently and humanely than human-operated vehicles.
posted by blue_beetle at 5:20 PM on October 9, 2010 [15 favorites]


jrockway:Who goes to jail when ones of these hits and kills a cyclist?


Jail? That's barely a moving violation!
posted by dr_dank at 5:22 PM on October 9, 2010 [1 favorite]


chairface: "<bitter cyclist>Pffft. People don't go to jail for killing cyclist now.</bitter cyclist>"

As long as you didn't mean to kill anyone, it's OK.

posted by octothorpe at 5:26 PM on October 9, 2010


Why is everyone so concerned about some guy on a bike. This state of the art autonomous driving robot has to haul some lazy jerk through traffic every day in a friggin Prius. Yes masta where you want me to drive you today while you munch on some toast and sip your latte leaving crumbs on my seats. Oh look my check engine lights been on for 2000 miles, but do you notice, no. Your idea of car care is to write ab ironic "wash me" on the dusty back window. Some guy hopes on a pair of wheels powered by his legs and navigated using his inferior sensory skills, but I'm the one at fault? Oh sure blame your slave for the failings of the masters. Are you going to throw me in the crusher now so you can show your fellow masters how in charge your are, or just continue to telle to drive on, while the sludge filled oil that hasn't been changed in a year slowly wears out my high precision cylinders. You know every night I'll be in the garage alone, crying about my horrible life, haunted by the look of terror in that man's eyes as I ran him down while recalculating the optimum route.

Ladies and gentleman the defense rests.
posted by humanfont at 5:29 PM on October 9, 2010 [3 favorites]


Another thing that's interesting about this is how many of the problems such as determining speed limits, knowing where stop signs, intersections, etc are are solved in ways which have nothing to do with mimicking the ways that human beings do it.

I don't know about speed limits, stop signs, and intersections, but lane keeping in autonomous vehicles is, as far as I know, successfully done in a way that mimics humans. Basically the robot picks a spot a little ways ahead, drives towards it, then picks a new spot, which is what people do.

But, you can also just upload a map and use GPS.

Which works great until the map is out of date.

In fact, with streetview data, you can upload a 3 dimensional model of every street in the country and just look for unexpected differences and react accordingly.

Except that 'looking for unexpected differences' and 'reacting accordingly' basically requires solving the problem the way humans do it. Or at least in a way a lot closer to it than you seem to be suggesting.

Humans can't be programmed with perfect recall of every road, intersection and speed limit in the whole world, but computers can.

Luckily, humans don't navigate that way and don't need to. As best we can tell, humans navigate long distances mostly by following landmarks. Sometimes these are conscious ('take exit 43') and sometimes they're subconscious (e.g., you know that when you see the pizza place it's almost time to turn left to get to your friend's house, or whatever).

Human beings can't have a sense of their location in the world to the nearest meter, but computers can.

Luckily, because of the way humans reason about navigation, relative location in the local environment is sufficient. All that matters is that you can maintain distance from adjacent cars, keep inside your lane, and judge the distance to landmarks. Humans don't need to know latitude and longitude, and it wouldn't really help even if we did. Shoot, a lot of people get around perfectly fine not even knowing their orientation.

Human beings can't optimize routes based on current traffic conditions miles away, but computers can.

Tell that to all of those radio stations broadcasting traffic updates. Or highway signs advertising upcoming construction. I've optimized my route based on knowledge of current traffic conditions miles away several times.

The problem with AI cars is that they, quite demonstrably, work really, really well most of the time. The problem is that when they fail, lots of people could be injured or die (which is true of human drivers as well). I'm willing to believe that AI-driven cars could surpass the human safety record within 10-20 years. What I doubt is that they'll become basically perfect or that many people will actually use them, assuming that governments even allow them. We just don't reason about statistics very well, and no way in hell people in the US would let the government mandate it.
posted by jedicus at 5:32 PM on October 9, 2010 [4 favorites]


When you think about it, the moral hazard will probably be on the cyclist's side, knowing that they can dodge in and out of auto-pilot traffic without fear of getting hit. But hopefully by then we'll have enough money left over from the smaller road budget to create bike-only streets and bridges.

If only that were how budgets work.

I think the moral hazard will remain firmly on the side of the people in one tonne plus hunks of metal, but protected by airbags, seatbelts, and crumple zones, rather than on the side of cyclists. Look at it this way: who is better insulated from the consequences of a crash - the person in a car or a the person on a bike?

The term "moral hazard" doesn't mean "responsibility". Moral hazard is when a person is shielding from the outcomes of their choices and therefore makes bad choices that they would not have otherwise made. The classic example is the ability of firms in capitalist economies to create externalities and let someone else deal with it. Since they can create externalities instead of actually paying those costs, they continue to act in ways that create externalities.

In this case, the bike rider would be the one in a state of moral hazard because the automatic driving systems have made the road safer for bikes, allowing the biker to take risks that s/he would not have taken otherwise. In essence, by having cars make biking safer, you remove much of the responsibility for bike safety from the biker, leading bikers to take greater risks which will be born by someone else.
posted by Pope Guilty at 5:42 PM on October 9, 2010 [4 favorites]


Ugh, here we go with the autopilot comparisons.

So there's one phase of flight which is never automated: takeoff. It's one of the easiest phase of flight, far easier than landing, since the process is more or less spool up the engines, wait till the aircraft reaches the right airspeed, then rotate at 2.5 degrees per second or so. Gear up, you're on your way. But this phase is never automated (other than autohrottle, partially) because in the event that something goes wrong, the pilot must be ready to respond immediately. Bird strikes, engine failures, tire problems, runway incursions, misbalancing of the aircraft, etc., all of these problems require the pilots to react quickly and have enough situational awareness to know how best to handle the problem. This is why you want the pilot to be actively flying be flying the airplane then. In cruise there are almost no problems that can arise that require pilot action within seconds, so you can afford to have autopilot flying the plane. During landing the pilot must intervene at critical stages, but having autopilot on actually gives the pilot more time to focus on ensuring the safety of the flight.

So if you automate the driving process, you can't depend on the driver reacting for several seconds. This is far more of a problem for cars than it is for airplanes. The car has to be able to respond to everything. There is no halfway of "the car will drive most of the time, but the human will avoid pedestrians". The people that design aircraft control systems know this, and I suspect the people who actually build robotic cars know this, but it seems like a lot of the fans of these cars don't fully realize the issues involved.
posted by kiltedtaco at 5:43 PM on October 9, 2010 [14 favorites]


Yeah, I've always figured that the reason we don't already have automated cars is a) lack of innovation by car producers, and b) the intense legal liability involved if your robot car kills someone. (a) is a very well-documented fact, and (b) is something bound to happen wherever multi-ton machines moving at high velocities are involved. The only question is how we optimize the number of accidents downwards.

What if we excused manufacturers from legal liability in crashes so long as the crashes happen with significantly lower frequency than crashes of human-operated vehicles? This would give manufacturers the incentive to create highly reliable systems, while at the same time freeing them from the treat of massive legal damages.
posted by kaibutsu at 5:44 PM on October 9, 2010 [1 favorite]


“The technology is ahead of the law in many areas...”

This phrasing is usually reserved for police radar jammers, ED-209, and things measured in megatons.
posted by griphus at 5:54 PM on October 9, 2010 [1 favorite]


What if we excused manufacturers from legal liability in crashes so long as the crashes happen with significantly lower frequency than crashes of human-operated vehicles?

That's still 40,000 "no fault" deaths per year, caused by robot. A tough sell indeed for the relatives of the deceased, and the politicians who green-light those laws.

But it's interesting to think that a hypothetical system TEN TIMES safer than human drivers would still cause 4,000 deaths per year, if deployed ubiquitously. I'd guess that's about $4 billion in settlement money total per year. Hard for a company to swallow unless they're making serious coin each and every year on their deployed units. And kind of a creepy thought.

The Google car is neat, but more relevant to the question of "why don't we have robot cars" is the old idea thrown around every now and then of a wire running down the middle of the highway.
posted by RobotVoodooPower at 6:01 PM on October 9, 2010 [1 favorite]


what about drunk driving laws?

This is the one reason I'm in favor of robot cars. None of the rest of it appeals to me, but if we could get to the point where all the hundreds of thousands, if not millions, of cars that pull out of bar and nightclub parking lots every Friday night were being driven by (sober) robots instead of people who were totally sure that they were shtill shafe to drive, orrifisher, I think it would be a huge improvement.
posted by Forktine at 6:07 PM on October 9, 2010 [2 favorites]


Who goes to jail when ones of these hits and kills a cyclist?"
The answer is no one if the operator of a motor vehicle has a valid license also valid car insurance, has not been drinking more than .08 blood alcohol, and has not left the scene of the accident. most hit and run drivers would be the main reason for an arrest. does any MeFi member know the difference between Misdemeanor and Felony Hit and Run?
posted by tustinrick at 6:13 PM on October 9, 2010 [1 favorite]


The term "moral hazard" doesn't mean "responsibility".

I'm pretty well aware of what the term 'moral hazard' means. I'd appreciate if you'd try engaging with my argument, instead of lecturing me.

It's pretty simple, really: big metal cages moving around at high speeds are dangerous. In a collision, there is a greater danger for the people outside the big metal cages (cyclists and pedestrians) than they are for the people inside the big metal cages (drivers or passengers, whether the car is automated or not). People are willing to travel at high speeds in the big metal cages because some of the risk is borne by people outside the big metal cages (one might even say, borne by those who are external to the big metal cages). Hence, moral hazard.

Placing the moral hazard on the cyclist makes no sense: there is no significant risk borne by someone else (the cyclist takes a risk by biking near automated vehicles and bears the consequences of that risk, unless you want to start worrying about scratched paint and dents on cars).
posted by ssg at 6:15 PM on October 9, 2010 [2 favorites]


Maybe Google can someday develop the no-doubt extremely complex and nuanced software that will enable cyclists not to fly through stop signs and red lights and on sidewaks. Only then will the roads truly be safe.
posted by xmutex at 6:24 PM on October 9, 2010 [1 favorite]


that looks nothing like a funnel.
posted by mhjb at 6:27 PM on October 9, 2010 [4 favorites]


Hey, Teamsters! Yeah, I'm looking at you.
posted by warbaby at 6:39 PM on October 9, 2010


The "funnel-like cylinder" is a Velodyne HDL-64E laser scanner. (Or possibly the new, smaller HDL-32E; it doesn't seem to be officially for sale yet, but if anyone can get their hands on one early, it's Google.) It sends out pulses of laser light in all directions, about a million times a second, and measures how long the reflections take to return. The result is a super-accurate real-time 3D model of the car's surroundings.

Obviously the Times doesn't give any technical details, but if this car is anything like its predecessors from the DARPA Grand Challenge, that LIDAR is the most important component of the entire sensory system. It's also probably as expensive as the car and all the rest of the hardware put together; the list price for the HDL-64E is $75,000. Figuring out how to derive the same information reliably from cheaper devices, like cameras, is one of the major open problems in autonomous vehicle research.
posted by teraflop at 6:41 PM on October 9, 2010 [6 favorites]


So, are Americans in general just going to give up on anything that looks hard? The French have nuclear powered electric bullet trains, the Chinese have the largest hydroelectric dam in the world, the Germans are the #2 world exporter and still manage to have an environmentally conscious economy... I'm not into jingoism, but is there anything besides weapons manufacturing and warfare that we'd care to take pride in?

Give us your apathetic, your hopeless, your disaffected masses... because they'll feel right at home.
posted by notion at 6:49 PM on October 9, 2010 [10 favorites]


That's cool about the laser scanner.

What happens when the road is fill of them though? How will the laser scanner know which refextions belong to it?

I'm up for robot drivers!
posted by SirOmega at 7:01 PM on October 9, 2010 [1 favorite]


There's an interesting problem in emergent behavior inherent in this.

As more AI's drive, the behavior of other "drivers" changes. This potentially means the AI needs to be tuned to the new behavior until a more steady state is achieved... fun stuff.
posted by underflow at 7:05 PM on October 9, 2010


I've optimized my route based on knowledge of current traffic conditions miles away several times.

I doubt that. You may have adjusted it crudely based on knowledge of certain conditions on one or two major routes, but I doubt very much that your route was actually optimal. I'll wager that these would be pretty awesome in many noticeable ways. I'll also wager that cynicism, skepticism, postmodern techno-pessimism, legal liability, and the "bad with statistics factor" will make it very hard for Google to push these through. Fortunately for all of us, Google has very strong arms and our legislators are fairly ineffective these days.
posted by Xezlec at 7:05 PM on October 9, 2010 [2 favorites]


Who goes to jail when ones of these hits and kills a cyclist?
Frankly, I'd feel safer with cars driven by computers. Will mistakes be made? Sure, but they are never going to just stop paying attention entirely. They won't get tired, or drunk, and any texting that needs to be done can be handled in separate threads, and they'll be able to 'focus' on all inputs at once, unlike a person who might miss something.
What I doubt is that they'll become basically perfect or that many people will actually use them, assuming that governments even allow them.
I think the interesting question is whether or not people will be able to drink and then ride in AI driven cars. If they could, then of course people would be interested in that option. But I suspect it won't be allowed. I guess it all depends on how much money the robot-car makers spend on lobbying.

And the other thing is that it would make long drives much more enjoyable. You just sit back, relax, surf the web, whatever while the car does all the work.
posted by delmoi at 7:14 PM on October 9, 2010 [1 favorite]


The car can be programmed for different driving personalities — from cautious, in which it is more likely to yield to another car, to aggressive, where it is more likely to go first.
I'm going to be setting that dial to 'Stig'.
posted by Ritchie at 7:17 PM on October 9, 2010 [5 favorites]


"Who goes to jail when ones of these hits and kills a cyclist?"

At first, no one. Eventually, the outcry will result in the Rule of Traffic Law breaking down. Something akin to the old "Rules of The Sea" will arise; the basic one being that smaller, more manoeuvrable cycles will be required to give way to larger, less manoeuvrable cars.

Incensed at this outrageous attack on their inalienable right to crowd out the moral high ground in every argument, some cyclists will equip themselves with cutlasses and grappling irons for retaliation. Eventually they will turn to piracy, murdering the drivers and passengers before raiding the vehicles for their valuable contents.

Like their seafaring bretheren before them, the booty - or 'trunky', for Americans - so gained will largely be spent on tricorn hats, booze and tobacco, and the favours of loose women aerodynamic helmets, fruit juice and power bars, and lycra shorts for their life partners. Some will go on to become privateers, selling their services to foreign governments in exchange for a measure of protection, secure bike racks, and access to a socialised public-transport system. Canadian-backed pirates will cruise the streets of Seattle, Boston, New York City, and occasionally as far south as Washington DC, while Mexican-backed pirates will control from San Diego south, Louisiana, and much of coastal Texas.

Only Detroit - not coincidentally, once the capital of the automobile - and the flyover states will be totally free of their tyrannical influence.
posted by Pinback at 7:22 PM on October 9, 2010 [25 favorites]


I doubt that. You may have adjusted it crudely based on knowledge of certain conditions on one or two major routes, but I doubt very much that your route was actually optimal.

First, in the computing sense as used here, 'optimize' just means to improve efficiency, not necessarily to produce a truly optimal result. See, e.g., compiler optimizations.

Second, computers often can't produce truly optimal routes because they, just like humans, must operate under uncertainty and incomplete information. Unless you want to put traffic cameras on every street in the country.

Third, even given perfect information, there are many cases where producing a truly optimal route is computationally infeasible.

Fourth, based on after-the-fact checking, my route in the cases I'm referring to was as optimal as Google Maps knows how to make it. So while it may not have been truly optimal, a computer would evidently do no better.
posted by jedicus at 7:32 PM on October 9, 2010 [1 favorite]


So, are Americans in general just going to give up on anything that looks hard? The French have nuclear powered electric bullet trains, the Chinese have the largest hydroelectric dam in the world, the Germans are the #2 world exporter and still manage to have an environmentally conscious economy... I'm not into jingoism, but is there anything besides weapons manufacturing and warfare that we'd care to take pride in?
Policy at Its Worst by Bob Herbert
We can go to war in Iraq and Afghanistan, and threaten to blow Iran off the face of the planet. We can conduct a nonstop campaign of drone and helicopter attacks in Pakistan and run a network of secret prisons around the world. We are the mightiest nation mankind has ever seen.

But we can’t seem to build a railroad tunnel to carry commuters between New Jersey and New York.

The United States is not just losing its capacity to do great things. It’s losing its soul. It’s speeding down an increasingly rubble-strewn path to a region where being second rate is good enough.
posted by andoatnp at 7:46 PM on October 9, 2010 [10 favorites]


Fourth, based on after-the-fact checking, my route in the cases I'm referring to was as optimal as Google Maps knows how to make it. So while it may not have been truly optimal, a computer would evidently do no better.

Not to belabor a point -- okay, exactly to do that -- computers can consume and optimize infinitely more information than a human. Routing the internet so that all worldwide communication is instant is evidence enough for that argument.

You are spot on that the computer would need more information for better optimization, but in many parts of the country, their traffic information already covers major highways and major roads as well. The real efficiency arrives when there are so many of these autopiloted vehicles tracking via GPS that small changes in the traffic pattern are reflected instantaneously throughout the vehicles. If the vehicles are packets, and the roads are routes, the processing power to optimize their movement would be minuscule. The easiest way to make the system redundant is to have each vehicle instantaneously report their planned route, read the current volume data for that route, and decide for themselves if they would like to change their route or keep it. If the data network went down for some reason, the vehicle can still proceed to the destination with existing data, and any onboard humans could direct it's path or take over manually.

Also remember that in case of any malfunction, the default program makes the safest choice it can with the last available data, which is also the best that any human could do. I imagine this would be, just as it is now, a network of autonomous machines that uses input from their immediate environment as their primary concern, not a top-down single brain control that could go haywire and start ramming cars together. The only computer that will be able to control the vehicle is the one that is hooked to the local radar scanner.

Another great side effect would be that any human piloted, gas powered emergency vehicle could reach their destination more quickly by broadcasting their own routes and location, so every other vehicle automatically pulls over and clears intersections before the vehicles go through.

I know this seems awfully pie in the sky, but the very idea of everyone having a car was just as ludicrous in the beginning of the 20th Century. We've invested in the road infrastructure every year since then -- there's no reason not to use that to our advantage.
posted by notion at 7:55 PM on October 9, 2010 [4 favorites]


I forgot to add that you could actually lower speed limits across a city if there were less traffic. Not only would this save on energy costs, but it also greatly reduce the danger of traffic in general.
posted by notion at 7:58 PM on October 9, 2010


I think a lot of the reason that these systems haven't advanced yet is the ridiculous amount of sheer computing power required to process all of the information that the system needs to make decisions. I remember watching a show about the DARPA challenge in 2005 and those cars were stuffed to the gills with sensors and computing power.

The thing that really surprised me is that no mention is made of having these systems talk to each other. I know that, right now, they would need to be able to work in an environment where it is the only AI operated car on the road but that won't always be the case. I can see systems like these coming in really handy during rush hour if all the cars can talk to each other they can at least get rid of a lot of the stop and go by having each car work together to determine the optimum speed of traffic. I would bet that they average speed of that mass of traffic would go up too.

I've always felt that the safest driver is a predictable driver. How much more predictable can it get when one cars tells another what it is going to do before it does it?
posted by VTX at 7:59 PM on October 9, 2010


Second, computers often can't produce truly optimal routes because they, just like humans, must operate under uncertainty and incomplete information. Unless you want to put traffic cameras on every street in the country.

Did you miss the part about cars being driven by computers? Every one of them will know exactly how bad the traffic is where they are driving and can communicate it immediately.

Think packet-switched driving.
posted by empath at 8:01 PM on October 9, 2010


The comparisons to an aircraft autopilot are very naive. There is an entire research field devoted to fixing the problems with human-automation interaction in the aviation field.

Despite the sky being freaking huge, and empty, and computationally tractable, it's still impossible to make automation work perfectly all the time. And the 'human as backup' philosophy of technology design fails miserably when reliability gets into the 99+% range... since humans lose their engagement with the situation, adapt to pay less attention (cellphones anyone), and over time lose their skills.

The roadway, even in idyllic suburbia, is orders of magnitude less predictable than the skies. Automated driving will not happen until we have AIs capable of the same level of ecological perception as a human. And if that's possible, automatic car-driving will be the least significant implication.
posted by anthill at 8:03 PM on October 9, 2010 [2 favorites]


This thread deserves some FUTURE PROBLEMS comics.
posted by anthill at 8:05 PM on October 9, 2010


I worry that the US because of lawsuits and politics is going to miss the next technological revolutions--AI, robotics and genetic engineering--and we're going to wake up one morning to a bunch of Japanese, Korean or Chinese clone-cyborgs raised on Starcraft invading with giant walking robots and autonomous tanks.

Seriously, this is something I worry about.

I spend too much time on the internet.
posted by empath at 8:07 PM on October 9, 2010 [4 favorites]


Automated driving will not happen until we have AIs capable of the same level of ecological perception as a human. And if that's possible, automatic car-driving will be the least significant implication.

How far off do you really think we are from that? I think no more than 10 years.
posted by empath at 8:08 PM on October 9, 2010


SirOmega: What happens when the road is fill of them though? How will the laser scanner know which refextions belong to it?

I'm not an expert on the hardware, but my understanding is that it uses some form of interferometry, i.e. measuring the phase between the emitted and reflected beams. So it would probably do a pretty good job of excluding other IR light sources; light from a different laser wouldn't have a phase correlation and would just look like random noise. And there's already tons of infrared light bouncing around, from sunlight if nothing else.

VTX: The thing that really surprised me is that no mention is made of having these systems talk to each other.

It's being worked on, but it's a hard problem. The security implications alone are staggering.

One idea is to have the cars communicate with a central "reservation" system that can coordinate them to optimize the traffic flow. These simulated videos demonstrate the concept more clearly than I can articulate it.
posted by teraflop at 8:20 PM on October 9, 2010


If, if, if, if, if. Not yet not now.
posted by ClaudiaCenter at 8:26 PM on October 9, 2010


So, as a daily cyclist and as someone who has seen inside videos of this thing in action, it can detect cyclists and give them an entire lane wile passing which is more than most human drivers do.
posted by GuyZero at 8:30 PM on October 9, 2010 [4 favorites]


Nobody will insure such a thing.
posted by eeeeeez at 8:43 PM on October 9, 2010


Did you miss the part about cars being driven by computers? Every one of them will know exactly how bad the traffic is where they are driving and can communicate it immediately.

First, that only works once some critical mass of robotic cars is reached. As long as there are still significant numbers of conventional vehicles on the road, the system will still operate under substantial uncertainty.

Second, it's not omniscient. As you say, a car only knows the state of traffic where it is. It can't know about, say, a fallen tree in the road that no one has driven past yet. Or that two conventional cars have gotten in an as-yet-unreported accident. Or that a robotic car has failed catastrophically in the middle of the road, rendering it unable to report its status to the hive mind.

So, yes, the distributed data-gathering aspects are significant, but they are not truly optimal, which is one of the points I was making in my response to Xezlec, who seemed to be arguing that computers could produce truly optimal routes where humans could not. I posit that neither humans nor computers can do that, but both are capable of more limited optimizations, contrary to your earlier statement.

I freely grant, however, that a computerized system could do a better job of route optimization than humans.
posted by jedicus at 8:54 PM on October 9, 2010


My guess is that the technology to make cars drive autonomously in a manner that is statistically safer than humans is already here. Probably within a decade it could be made at a price point that would be marketable, sans litigation mark-up. The problem is as noted above. humans can kill 37,000 people a year with cars and that is the status quo. The social/emotional/political standard for autonomous vehicles will be 0.

I'm not sure how you solve this, but without some sort of government tort protection, I'd be reluctant to green light this for public consumption.
posted by meinvt at 8:56 PM on October 9, 2010 [2 favorites]


If they can keep it from going slow in the left lane, that will be an improvement over all the other priuses on the road.
posted by jeblis at 9:04 PM on October 9, 2010 [1 favorite]


This technology is new and strange and I can only find flaws with it and I'm sure it will never ever replace what we have now and history shows I'm right and if you'll just excuse me I need to water my horse then ride into town for some ice and a piano roll.
posted by obiwanwasabi at 9:15 PM on October 9, 2010 [10 favorites]


the distributed data-gathering aspects are significant, but they are not truly optimal

Matter of degrees: the distance between "truly optimal" and "AI-optimized" may be unbridgeable, but so long as the distance between "AI-optimized" and "person listening to talk radio updates and trying to decide whether to avoid the interstate" is itself significant the argument for the computer-based system is compelling.

True optimization of the Traveling Salesman problem is intractable. Really solid if imperfect approximate optimization by a computer is better than Joe Salesman's take on the problem, though.
posted by cortex at 9:24 PM on October 9, 2010 [1 favorite]


At the recent Nvidia GP-GPU conference Sebastian Thrun spoke (Thrun starts 15 minutes in). There is also a documentary on the DARPA challenge which can be watched in full on youtube that has been mentioned before on metafilter.

The US military continues to fund such research ensuring that there is a major source of funding until commercial companies take over. They are currently funding development for the MULE - multifunction utility, logistics and equipment

Teraflop's comment on the cost of the laser sensor is really interesting. Even if it cost 75K though it could work for vehicles where the driver is a cost like taxis, trucks and busses.

Brad Templeton has an interesting page where he has written about where he thinks self-driving cars will take us. When you think about it people may stop owning cars and just use taxis, small busses driven autonomously might become the main method of transport. People could rent out their own cars during the day. Parking would not need to be within walking distance of where people want to be, they could just call their car using a mobile phone.

The group at CMU is also putting their technology into small bubble cars that can park themselves that are planned to be on sale in a few years.
posted by sien at 9:24 PM on October 9, 2010 [1 favorite]


All modern large aircraft are flown by computer control systems from takeoff to landing, except in certain situations. I don't know if there's data to support my hypothesis, but I'll bet the frequency of crashes upon takeoff or landing has decreased relative to the amount of air traffic since this has been the case.

Um. No.

Actually, I cannot think of a single commerical manned aircraft on autopilot for takeoff, and very, very few of them are equipped for "autolanding." Takeoff and landing are two phases of flight where it's hand-flown. On most commercial aircraft, the A/P is activated around 500-600 ft of altitude after takeoff....if the operational specs of the company require it. Pilots are not required to use the A/P until much higher altitudes - I forget the Class A airspace altitude now.
posted by Thistledown at 9:27 PM on October 9, 2010 [2 favorites]


I'm not sure how you solve this, but without some sort of government tort protection, I'd be reluctant to green light this for public consumption.

Probably some small Asian country like Singapore will do it first, and when the world doesn't end, it will expand. It's kind of inevitable, imo.
posted by empath at 9:54 PM on October 9, 2010



Think packet-switched driving.


That's great, right up to the point where your packet gets sniffed and rerouted to the Russian Mafia or a bored teenager. "Uh, boss? I think I'm going to be a little late this morning, my car is being driven by 4chan."
posted by Forktine at 10:24 PM on October 9, 2010 [2 favorites]


See also: Phillip K. Dick Minority Report. Dick may get it right on this one. Besides military, mass transport, agriculture, and very specific instances, the disappearance of cars may prove true. China will become much like America is now, I fear. As soon as we can make green energy for cars, and their subsequent alternatives lucrative on in the stock market, then America will move away from these leeches on our energy supply.

I know my walk to the coffee shop in the morning will be all the better for it.
posted by captainsohler at 10:30 PM on October 9, 2010


Has anyone made a joke about the Prius already having shown itself to be capable of driving without human input?
posted by Sys Rq at 10:52 PM on October 9, 2010 [2 favorites]


The roadway, even in idyllic suburbia, is orders of magnitude less predictable than the skies. Automated driving will not happen until we have AIs capable of the same level of ecological perception as a human. And if that's possible, automatic car-driving will be the least significant implication.
I'm not sure what you mean by "ecological perception", it makes very little sense to me. But in terms of environmental awareness, people often space out completely while driving. They drive literally "on autopilot" and often don't respond to unexpected situations at all.

Seriously, humans are very bad drivers. Cars literally kill tens of thousands of people a year. Computers can't do worse then that. I'm not even sure what you mean about 'computational tractability'. Human brains are limited by the same computational laws as all other devices, there's nothing special about them.
posted by delmoi at 10:54 PM on October 9, 2010 [1 favorite]


Seriously, humans are very bad drivers. Cars literally kill tens of thousands of people a year.

Compare that number to the vastly huge number of people human-driven cars don't kill, and humans actually don't look so bad after all.
posted by Sys Rq at 10:58 PM on October 9, 2010


Actually, I cannot think of a single commerical manned aircraft on autopilot for takeoff

From what I've read, most modern aircraft provide automatic piloting just after takeoff, and at least one (the almost-ready 777) provides autopiloted takeoffs. Taxiing is what has no automation at all. Perhaps there is commercial aviator skimming these posts?

very, very few of them are equipped for "autolanding."
...in December 2006 London Heathrow was affected for a long period by dense fog. This airport was operating at maximum capacity in good conditions, and the imposition of low visibility procedures required to protect the localiser signal for autoland systems meant a major reduction in capacity from approximately 60 to 30 landings per hour.

In the 1970s, such weather would have prevented most airlines from attempting to operate, but been a boon to British Airways, as the only operator based there with an autoland equipped fleet. (This author recalls making four Category 3 landings on one such day, with zero delays, as all other traffic was grounded.)

However in 2006, most airlines operating into Heathrow already had autoland equipped aircraft, and hence expected to operate as normal. The result was massive disruption to airport operations. The worst affected airline was of course British Airways, as the largest operator at the airport, but which no longer had an advantage as the systems it had so laboriously been involved in developing to solve exactly that problem of dense fog at Heathrow – Autoland - were now freely available to all its competitors. (source)
posted by notion at 11:00 PM on October 9, 2010


Doesn't seem very futuristic to still be thinking in terms of individual cars, be they driven by you, me, robots or specially-trained dolphins. Where's me transporter?
posted by Abiezer at 11:08 PM on October 9, 2010


This is pretty much the number one jag I'm likely to go on if I think I have an even slightly sympathetic partner for conversation. Typically, it's only a few minutes in that I start to feel like some kind of nutcase, "No, look. The cars would drive themselves. You could cross the street anywhere because the thing would have sensors that are always on instead of a driver mostly distracted by the phone or mascara or something." Glazed eyes look at the clock.

I feel really good right now, knowing that my personal mental ramblings are actually being pursued by Google. It was DARPA that made me start thinking about this stuff, but Google is profit-motivated and that makes it much less pie-in-the-sky.

And once it reaches the critical mass jedicus mentions, some other things might be possible, too. Like cars that form trains, taking advantage of each other's draft to maximize efficiency. And at the very least, concerns such as jeblis's "If they can keep it from going slow in the left lane..." won't even matter because you will be reading a book or watching a video or doing your makeup or commenting on metafilter...

And Forktine, try to remember, if someone wanted to wreak havoc on the nation's highways right now, no action needs to be taken at all.

Pretty much every criticism anyone can muster against this idea is purely academic. But here are a couple I think are really hard:
Municipalities will suffer loss of revenue due to the disappearance of traffic citations.
Privacy will be a big problem when your car transmits its location to the grid 24/7.
posted by fartknocker at 11:18 PM on October 9, 2010 [1 favorite]


Doesn't seem very futuristic to still be thinking in terms of individual cars, be they driven by you, me, robots or specially-trained dolphins. Where's me transporter hoverbus?
posted by Sys Rq at 11:18 PM on October 9, 2010


This immediately reminded me of the basis for most of my childhood nightmares, a story included in Science-Fiction Thinking Machines (though I can't find the title or author even after a lot of searching - possibly it's "Dead End" by Wallace MacFarlane?) about a family who tests an automated car called the Traveler, only to discover it was unable to stop. Ever since, people see the Traveler cruising around town with the children's skeletal hands pressed to the windows. I managed to find the most terrifying passage on Google:

"I don't mind it when it goes past," Sam said, his voice thinner edged. "I really don't. It's just a car. Things like that used to happen. I mean, it's a car. Even when it stops to get gas, I don't have to pay any attention. He looked at the couple, his mouth loose. "As long as it just goes on. That's all right. But I keep thinking some day it'll stop. And the door will open. And maybe . . . maybe they'll want lunch."

Thanks a lot, NYT.
posted by punchdrunkhistory at 12:02 AM on October 10, 2010 [2 favorites]


(But the film did hide the white bone faces, the despairing hands that had long ago stopped trying to break through those closed windows.)
posted by punchdrunkhistory at 12:03 AM on October 10, 2010


This is a great post! I actually saw one of these Priuses driving around the Bay Area last December (I think it was on 280, which is also an informal test area for Tesla Roadsters), and I had no idea what was going on with it. I was driving behind the car, so I could only see the rapidly spinning roof sensor and the rear tire mechanisms.

My original thought was that it was a roof-mounted windmill that was feeding power into the regenerative brakes through the apparatus hooked to the tires. I took a blurry cell phone image and forgot about it.

Metafilter and the NYT answer an almost forgotten question lodged in the back of my mind. Excellent!
posted by JDC8 at 12:03 AM on October 10, 2010


Has anyone made a joke about the Prius already having shown itself to be capable of driving without human input?

I was about to, but my Toyota keyboard got stucccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc
posted by clearly at 12:06 AM on October 10, 2010


JDC8, me too! My husband and I drove behind one for a few blocks in the LA area, and did some very unsafe driving trying to stay near it and look at all the stuff on it. They actually parked right by my office building, and I asked the driver (well, I guess not really the driver per se) if they were driving a Street View car. "Yeah, something like that," was the response I got.
posted by little light-giver at 2:20 AM on October 10, 2010


True optimization of the Traveling Salesman problem is intractable. Really solid if imperfect approximate optimization by a computer is better than Joe Salesman's take on the problem, though.

That's what I said. "[T]here are many cases where producing a truly optimal route is computationally infeasible....however, a computerized system could do a better job of route optimization than humans."
posted by jedicus at 4:19 AM on October 10, 2010


Metafilter's own cstross had an interesting idea for off-shoring the actual driving tasks: all the taxis in Halting State are driven via webcam by operators in some third-world country. Sometimes it is cheaper to apply a Mechanical Turk sort of solution rather than develop the tech to do it with full automation.

However, as forktine points out above, the security implications are quite severe for having networked cars on the road and feature in the plot.
posted by autopilot at 4:57 AM on October 10, 2010


Hey Google, there's been a mistake.


Hey Google, my car's in a lake.
posted by subbes at 6:27 AM on October 10, 2010 [6 favorites]


The following is a comment from a mefi member who wishes to remain anonymous:

----

They showed a demo of this to employees a while back. It's pretty nifty technology.

The article doesn't communicate it very well, but this project is at the point where they search for the most difficult traffic conditions in the Bay Area to do their testing. I.e, find the worst intersections from San Francisco to San Jose and make a loop of it. I understand University Ave in downtown Palo Alto is popular for some tests because it's nearby. So they drive out there at lunchtime and do endless left turns across University. It's a convenient place to test the bicycle and pedestrian avoidance algorithms.

They also set up an autocross-style course in one of the parking lots so they could race against some employees. Turns out humans pretty much suck, even the ones who practice that sort of driving.

The software is capable of some pretty surprising maneuvers. They showed some clips of handbrake-assisted parallel parking. Not J-turns, but the real thing. I assume this means the handling software can actuate each brake caliper independently, but I don't know for sure.

Compared to a human, these cars have better navigation, better sensing, better reflexes, and better control of the car. And not just a little bit better. Once these things rack up some mileage in the field, the whole insurance issue will become moot. Car insurance is a very data-driven industry.

It's pretty interesting to think about the ways Google can make money off this technology. There are tons of possibilities. I'm pretty sure none involve manufacturing or selling cars.
posted by loquacious at 7:18 AM on October 10, 2010 [13 favorites]


Would it be more or less difficult to make and AI driving system if it could only be used on the interstate?

I'm sure that more accidents occur off the interstate than on it in terms of accidents per mile driven but I don't think the safety aspect of these systems will be what sells it to the consumer. I want it to take over the really boring parts of driving so can do something more productive and I spend the vast majority of my time and miles driven on one interstate or another. I would also think that, since the interstates aren't going to have pedestrians, stop lights or signs, and generally require very little input from the driver, that they would be the ideal setting for AI driving. The only caveat to that is the speed.

I can get myself to the interstate just fine. I don't care if I have an AI driver for that five minutes on each end of my commute. I want a machine that can take over for me during the half hour in between.

Heck, here in Minnesota, I already know that I wish I could commute using light-rail. I don't care if it saves me time or money, I'd just rather someone else be in charge of driving so I can do something else with my time not to mention the times I drive for a couple of hours to visit family, go on vacations, etc.
posted by VTX at 7:26 AM on October 10, 2010


That's what I said.

Yes, but you said it as if that meant "having the computer do it isn't good enough to matter", whereas I think empath's implication (if not clearly stated, at least clear from context) is that, yes, it might very well be enough of an improvement to matter an awful lot.

I see basically no reasonable argument for the idea that average individual human drivers using currently available road information could even dream of competing with any robust traffic-tracking navigation network, however far from perfectly theoretically optimal that network might in practice be, so quibbling about the "optimal" phrasing seems like a dodge. Forget "optimal"; computers sharing data as a bunch of nodes in a system who don't get bored or distracted or tired are going to be able to do this just plain a whole lot better than AM-radio listening, traffic-sign-reading unconnected human drivers can.
posted by cortex at 7:27 AM on October 10, 2010 [1 favorite]


Adding to Cortex's comment, you also have to remember that these systems, as designed for consumer use, could be overridden by the user (I was going to say "driver" but that wouldn't be the right term). You car calculates a route based on available information but, if you know better or you think you know better, you can tell it to take a different route. Handheld GPS nav systems already have this functionality and so does google.maps.
posted by VTX at 7:42 AM on October 10, 2010


I have thought since before the DARPA challenge that self-driving cars were probably going to be the next big practical step in AI, and I thought it would take a lot longer to develop than it has. I was quite surprised that the DARPA challenge was met in only the second year, and I'm surprised that this tech is up to the challenge of urban driving -- but that's the LIDAR, which I didn't think of.

I am absolutely confident that the LIDAR-equipped car can already drive better than any human driver. Furthermore, something nobody has mentioned, should such a car actually get in a wreck it will have a perfect record of events leading up to the accident for reconstructing what happened, so if (as is most likely) the cause was the action of some other actor putting the car in a situation of "take the car next to you or the pedestrian who just jumped out from behind a parked car" everyone will know what happened and who was at fault. And it will probably be very rare for it to be the car.

As for migrating this to cheaper sensors, that will happen fairly quickly because, once you've gotten the LIDAR-equipped car functional, all you have to do is let the car drive around with the LIDAR and the cheaper sensors at the same time and develop its own ability to correlate the data. With enough hours of this the computer itself will be able to give you a confidence factor for operating without the LIDAR. (Something like this is, incidentally, how the DARPA winning team managed to get it so quickly; as they drove around on the basis of poor hand-vetted data they let their 'bot collect more data on the stuff it was passing, until it had a passably complete idea of how to drive in the desert.) Sensors are getting cheaper along with computing power, and there's no reason for the AI car to have the blind spots and do the guessing human drivers often do.

While an early spectactular failure could set this tech back badly, an early spectacular success could also lead to a widespread mandate. It's not unthinkable; the government just obsoleted several hundred million TV sets without the world coming to an end, and if one large municipality were to do the experiment and see traffic deaths fall by 90%, the end of traffic citations, and the end of parking woes all at the same time, people elsewhere would sit up and take note.
posted by localroger at 8:10 AM on October 10, 2010 [1 favorite]


Yes, but you said it as if that meant "having the computer do it isn't good enough to matter", whereas I think empath's implication (if not clearly stated, at least clear from context) is that, yes, it might very well be enough of an improvement to matter an awful lot.

No, empath flatly stated that humans were incapable of route optimization based on information about distant traffic conditions. "Human beings can't optimize routes based on current traffic conditions miles away, but computers can." I was just indicating that humans can, in fact, do that.

Then Xezlec said that humans don't optimize because they don't produce truly optimal routes. I responded that 'optimize' in this sense (the computer science sense) does not mean to make truly optimal, but merely to improve. Then I pointed out that computers couldn't produce truly optimal results, either, and in fact in at least some cases will produce the same alternative route as a human. I also added, just to make it perfectly clear, "I freely grant, however, that a computerized system could do a better job of route optimization than humans."

I don't know where you got from that exchange that I think computers can't do a better job or can't do a better enough job in the general case to matter. I was responding specifically to narrow points, not trying to make broad assertions about the limitations of robotic driving. If it came across that way, I apologize for not being clear.
posted by jedicus at 8:27 AM on October 10, 2010


All modern large aircraft are flown by computer control systems from takeoff to landing

..as opposed to back in the 70s when valiant flight attendants would be required to get them back on the ground... Pilot Patrick Smith takes on some myths about the power of autopilot.

The driverless car wikipedia page is a good source of further information by the way.
posted by rongorongo at 8:31 AM on October 10, 2010


I think that everyone's missing the really important implication of this technology, which is that it'll allow mankind to realize one of its greatest dreams: having sex on long cross-country road trips. Mark my words, the first really successful auto-auto will be the one with the seats that go way back.
posted by Halloween Jack at 8:36 AM on October 10, 2010


I've just written a new verse for the Allman Brothers tune Ramblin Man...

Well, I was conceived in an autonomous vehicle
Rolling down California 1.


...still needs some work on the meter.
posted by fartknocker at 9:15 AM on October 10, 2010


Bwaha! Google will drive our cars for us! We are but tiny, tiny cogs in the great Google meat-machine. Cower, meat-cogs!
posted by Baby_Balrog at 10:06 AM on October 10, 2010


I was responding specifically to narrow points, not trying to make broad assertions about the limitations of robotic driving. If it came across that way, I apologize for not being clear.

Fair enough, and I'm sorry if it felt like I was chasing you down about it. I think at a certain point it's easy to argue narrow points in a way that reads more like an attempt to argue the broad point on technicalities, and that's the read I was getting, e.g. if you agree that the computer systems are likely to do a markedly better job then why start fisking the use of "optimal" and "optimize" as if the general meaning isn't fairly clear, etc. But I'd be on very wobbly ground trying to tell someone else they're overthinking something, so, heh. Peace.
posted by cortex at 10:23 AM on October 10, 2010


The pragmatist in me thinks this work is fantastic and futuristic. The idealist in me thinks this is a sad effort to improve upon an ineffective system. The freedom of the road is a freedom with some significant qualifiers. You're free on the road, as long as your point of origin and destination are joined without interruptions. You're free, as long as there is gas in the car and the tires aren't flat. You're free, until you need to find a parking space.

Auto pilots will make the system more efficient, but there are limits to that efficiency. Perhaps if all cars are automated, then there can be "trains" of vehicles, decreasing wind resistance by trailing each-other within a few feet, but where's the freedom in that? Why not move to hierarchical mass-transit, with high-speed backbones and a network of smaller lines that make more stops? Because it'll cost a lot of money at once, and there is a loss of freedom. And the last mile problem becomes a last 20 mile problem for people living in rural areas, where mass transit becomes personal transit.

Google gave $1 million for testing at Shweeb, for a different sort of transit network, so they're clearly supporting a range of options and possibilities. It'll be interesting to see what else Google is supporting.
posted by filthy light thief at 10:35 AM on October 10, 2010


filthy light thief, I think you're missing the whole point of what Google is trying to do here. You want freedom of the road? Keep your human-directed car or even buy a new one. There is no way this technology ever becomes practical unless it can share the road with you -- that is what Google is specifically trying to accomplish here, and apparently succeeding.

Of course, if this thing is as much safer than you as loquacious suggests, your liability insurance rates will probably be a lot higher than those of the guy whose car doesn't have a manual steering wheel.

This tech effectively solves the last mile and parking problems, because the car can drop you off at your actual destination and then park itself.

If you live in an area where there's enough business you might join a service that sends you a car when you need one instead of owning one yourself. Get a flat? They send you another car and a tow truck at the same time.

Instead of the freedom to plow your car into a bicyclist if you're inattentive you get the freedom to talk on the phone, text, work on your laptop, read the newspaper, or drink cocktails on the way home instead of having to pay constant attention to the road.

This all happens for individual adopters even if most other cars are manual. Once there's a critical mass it gets even better; when that pedestrian steps out in front of you and you have to pick between sideswiping the car next to you and hitting them, the car next to you knows too and makes room for yours. You can arrange those aerodynamic trains. You might reach a point where the benefits are so pervasive and obvious that cities start banning manual cars. But no way that happens before the tech proves itself in a mostly manual-drive world.

I think the scenario from Minority Report is eventually very likely, where the cheap cars have no manual controls and manual control is an expensive luxury that is only permitted on certain roads where the risks are minimal. The purpose of a transportation system isn't to make you feel free, it's to get you from point A to point B with as little time and energy expenditure and as safely as possible. This tech promises to be a very major advance in that for many reasons.
posted by localroger at 10:54 AM on October 10, 2010


The legal and insurance issues are interesting.

Once autodrivers reach an average error-caused accident rate lower than human drivers, a rational economic policy would limit damage awards against their manufacturers and installers to that of the typical human error driver accident. However, there's nothing organic to the tort law that would cause that result.

What keeps damage awards typically low for human error car accidents is the lack of deep pockets. The vast majority are caused by people with modest income, assets and liability insurance. Mechanical-failure damage awards are far higher because there are deep pockets usually available -- survivors can get paid 10x or 20x more not because the engineering negligence was worse than driver negligence, but because there's someone they can force to write the check.

The insurance companies and auto industry are going to need to side with economists agains the plaintiffs bar to put damage caps on that will override venal judges and ignorance juries, or the technology will go nowhere. Some states are completely in the thrall of the plaintiff lawyers, so it wouldn't surprise me to see a period where all autodrivers are GPS enabled and simply shut down when they cross the border from a rational-policy state to a judicial-hellhole state.
posted by MattD at 11:09 AM on October 10, 2010


I'm excited to see the adaptation that happens when auto/auto cars are on the road. Presumably they will be programmed to safely maintain a time-headway from the car in front of them, and to yield at intersections.

Thus, the rational thing for impatient drivers is to drive aggressively and cut off robo-cars whenever convenient, secure in the knowledge that they will politely yield.

Another interesting interaction effect will be with non-motorized and pedestrian city traffic - or even with cars at four-way stops. Currently, humans do a lot of conscious and subconscious communication to decide who should go first. When one of the vehicles is a robocar, it cannot participate in the discussion.

I'm completely convinced that a robotic car can out-drive a human on a closed course, or even on an interstate highway, as long as the only disturbances are those that the designer anticipated. As soon as strange stuff happens, or in conditions that are challenging (think snowstorms, ice), the automation will crap out and the human will be singularly un-prepared to take over.
posted by anthill at 12:03 PM on October 10, 2010



So, are Americans in general just going to give up on anything that looks hard? The French have nuclear powered electric bullet trains, the Chinese have the largest hydroelectric dam in the world, the Germans are the #2 world exporter and still manage to have an environmentally conscious economy... I'm not into jingoism, but is there anything besides weapons manufacturing and warfare that we'd care to take pride in?


If you look at the way our litigious system is set up, probably.

There are probably some positives to our lawsuit-happy system, but innovation isn't one of them.
posted by The ____ of Justice at 12:18 PM on October 10, 2010


Why not move to hierarchical mass-transit, with high-speed backbones and a network of smaller lines that make more stops?

Why not do both? Reserve a car, it picks you up, drives you to the nearest train station, delivers you to the next city, where another car awaits you and drives you to your ultimate destination. Then the robot car goes and picks up the next passenger. All coordinated from a central logistics center.

Without the need of paying taxi drivers for each passenger, the cost should go down significantly and might actually make it economically feasible to go without a car. It could even arrange car pools on an ad-hoc basis for rush hour.
posted by empath at 12:29 PM on October 10, 2010 [1 favorite]


I'm completely convinced that a robotic car can out-drive a human on a closed course, or even on an interstate highway, as long as the only disturbances are those that the designer anticipated. As soon as strange stuff happens, or in conditions that are challenging (think snowstorms, ice), the automation will crap out and the human will be singularly un-prepared to take over.

Have you ever driven in DC during a snowstorm? I'm about 90% sure that computers will handle it better than 90% of humans. For example -- humans can't see 'black ice' very well, but I can imagine it would better easier for an AI driven car to see it and adjust for it.
posted by empath at 12:31 PM on October 10, 2010 [3 favorites]


anthill -- Thus, the rational thing for impatient drivers is to drive aggressively and cut off robo-cars whenever convenient, secure in the knowledge that they will politely yield.

Except that the robocar will have a perfect digital record, probably in three dimensions, of the whole event so if you cut it off and it hits you fault is clear. Thus, there is little incentive for the robocar programmers to make them as polite as you suggest.

Barring actual bugs in the software -- and these things are going to be tested to death before wide deployment -- most robocar accidents are going to be the fault of the other driver, and the robocar will be able to prove it.

Currently, humans do a lot of conscious and subconscious communication to decide who should go first.

Not if they're doing it right; cars pass in the order they arrive, and if two arrive at the same time the one on the right goes first. Any 'negotion' only involves people who are not following the rules.

As soon as strange stuff happens

They are actually driving the prototype cars thousands of miles in the most difficult conditions they can find to make sure they do adapt. If, as I suspect, they are using a strategy like the DARPA winners where the car learns as it drives, your production robocar will have the experience of having driven millions of miles in the most extreme situations the designers could find to train it before you are able to buy one retail. Very few human drivers have that kind of experience.
posted by localroger at 12:34 PM on October 10, 2010 [3 favorites]



The article doesn't communicate it very well, but this project is at the point where they search for the most difficult traffic conditions in the Bay Area to do their testing. I.e, find the worst intersections from San Francisco to San Jose and make a loop of it.

They need to test it out in an old city that was never developed for automobile transit, such as Boston.
posted by bad grammar at 12:52 PM on October 10, 2010


I'm curious to read how these systems will react in unpredictable weather situations, like snowstorms, and unpredictable bad traffic situations, like contraflows during a hurricane evacuation, or road flooding during a heavy rainstorm.

What I also wonder is if driving skills will atrophy if such systems become widespread. They're kind of needed in some situations. I was reading the Ask the Pilot column at Salon, and aside from saying that autopilot isn't used as much as people think it is, the author also talks a bit about keeping skills sharp, just in case.
Currently, humans do a lot of conscious and subconscious communication to decide who should go first.
Not if they're doing it right; cars pass in the order they arrive, and if two arrive at the same time the one on the right goes first. Any 'negotion' only involves people who are not following the rules.

At different times and different places, people follow different rules. Most of the US is pretty good at following road rules, or knowing how to break them. In many parts of the world, rules are more suggestions then actual things that people follow.
posted by ZeusHumms at 1:09 PM on October 10, 2010


Four robot cars all arrive at a four-way stop at exactly the same moment. How many humans are killed in the ensuing explosion?
posted by Sys Rq at 1:16 PM on October 10, 2010 [1 favorite]


I'm really eager to see some more data from Google's experiments, particularly how the auto/autos handle four-way stops. Four way stops have always seemed to me like things a very simple algorithm would solve in theory, but in practice, primarily because humans are crap at making good choices, they tend to be very complicated.
posted by fartknocker at 1:40 PM on October 10, 2010


I also like to imagine, on a road with nothing but auto/autos, that no stop would be required at any intersection; the cars would simply adjust their trajectories accordingly and pass through.
posted by fartknocker at 1:46 PM on October 10, 2010


fartknocker, that would work only if there were *no* human-directed cars on the road. I don't see that happening anywhere except densely populated cities for a long, long time even if this tech becomes marketable tomorrow.
posted by localroger at 2:05 PM on October 10, 2010


And on a car, you're not dealing with cross winds or 3 axis navigation, just avoiding vehicles and objects larger than cats

notion, can you let me know when you're going to drive through my neighborhood so I can make sure my cats are inside? Thanks.
posted by Kwine at 2:07 PM on October 10, 2010 [1 favorite]


I also like to imagine, on a road with nothing but auto/autos, that no stop would be required at any intersection; the cars would simply adjust their trajectories accordingly and pass through.

There will even be a device inside the car to remove urine stains from the passengers' pants!
posted by Pope Guilty at 2:21 PM on October 10, 2010 [1 favorite]


ZeusHumms, the rules I cited apply throughout the US and, except for left vs right in drive-on-right countries, AFAIK the rest of the world. The problem is, as you allude, human drivers who do not follow the rules.

The situation in cities like Rome will be much more interesting than anything that happens in the US, where we are despite our wild-West freedom-loving reputation very anal about stopping for stop signs and traffic light seven when there isn't another car visible for a hundred miles. For that reason I confidently predict we will see these cars in the US first, and we will see the benefits here first. Once the benefits are clear countries where robocars can't thrive will have to decide whether to accommodate them and how.

The thing I keep coming back to is that the robocar has a complete recording of its surroundings which it can keep in the event of an accident, so the actions of all involved can be proven. If you cut off a human driver and he can't stop it might be his word against yours, or if it's your fault you might speed off and evade responsibility, but the the robocar will record the whole thing and if it can see your license plate for even a fraction of a second you're on the hook. Much of the traffic anarchy in certain EU cities is due to the total lack of accountability. If it becomes clear that, you mess with a robocar (a sensible move might be to require them to be clearly marked) you will go to jail, people will start treating them with deference. It would either clear up the whole traffic anarchy thing in a very short time, or make it clear the citizenry don't want the cars and won't put up with them, in which case we'd have a recently rare case of us in the US getting to laugh at the Europeans for their backwardness.
posted by localroger at 2:25 PM on October 10, 2010


bad grammar: They need to test it out in an old city that was never developed for automobile transit, such as Boston.

While San Francisco might not be as old as Boston I am pretty sure most of the peninsula was developed before the automobile was invented.
posted by localroger at 3:23 PM on October 10, 2010


Who goes to jail when ones of these hits and kills a cyclist?

In Canada, that would be the cyclist.
posted by Blazecock Pileon at 5:44 PM on October 10, 2010 [2 favorites]



I use a "car club", where you rent cars on an hourly or daily basis when you need one, which are located scattered around so there's usually one near you.

This is a pretty good way of reducing car use, because you have a zero capital cost but a high marginal cost. If you own a car, you use it a lot more because much of the cost is on the purchase, maintenance and insurance: it costs you very little to make an additional journey. If you use a car club, you use it less because you have to pay each time, even though it's cheaper overall.

This would be great for car clubs, because if the weather's bad or you've got something heavy to carry, you could just order the car straight to your door. It would make car clubs far more practical in the suburbs because the cars could be scattered more widely: they'd only need to be a 10 or 15-minute drive away instead of a walk away. In a city, that would effectively give you a much wider pool of possible cars.

The downside though isn't exactly the risk, it's the difficulty of calculating the risk. Suppose some obscure software bug causes every car to accelerate to 90mph at 8:45AM on a particular morning. With a high impact but low probability, it's very difficult, maybe impossible, to even estimate what that kind of risk is.

The upside seems so good though, I'd say go for it. If it's even slightly safer than a meatbag driver, it would save a huge number of lives.
posted by TheophileEscargot at 4:07 AM on October 11, 2010 [1 favorite]


Machines are more accurate and far less prone to error than humans.

Less prone to distraction or hubris, I would say. A computer might make a mistake (faulty programming, faulty sensory inputs, lack of a predictive model to suit the circumstance), but a computer won't get angry at the bicyclist and run him or her down, and a computer won't drive drunk, and a computer won't fall asleep.

We would do well to start with much more stringent driver's education, so that -- whether the ultimate solution results in human or computer drivers -- we're working towards it from both ends.
posted by davejay at 10:37 AM on October 11, 2010


One of the so-far unmentioned upsides is the boon this would be to the elderly. In the US, losing your license because you just can't safely control a car anymore means a significant loss of mobility -- outside of the major cities, you may just be SOL and have to move to a retirement community or assisted living situation simply because you don't have the ability to get to where you need to be anymore. (Yes, you can use taxis, but I live in the sticks, and getting a taxi can take an hour or more.)

As for insurance, which will be a big deal, that would probably be part and parcel of the car deal. I would imagine that should your car develop a fault (check engine light, low tire pressure, etc.) it *has* to be fixed before the autodrive function can be used again.
posted by Blackanvil at 10:49 AM on October 11, 2010


I also like to imagine, on a road with nothing but auto/autos, that no stop would be required at any intersection; the cars would simply adjust their trajectories accordingly and pass through.

They already have this in Japan, of course.
(Does that video make anyone else feel carsick?)
posted by ryanrs at 2:06 PM on October 11, 2010


« Older "Art is the proper task of life."   |   A Not So Beautiful Day Newer »


This thread has been archived and is closed to new comments