The long, winding road to fully automated cars
September 7, 2017 1:28 PM   Subscribe

Automobile automation has been promoted as a boon to safety since 1939, though for much of this time, the plan for vehicle guidance relied on in-road guides, with public demonstrations of wire-guided roads continuing into the late 1990s. As recently as 2014, Volvo promoted the value of embedded magnets in roadways to help with autonomous vehicle guidance. Meanwhile, "self-sufficient" autonomous vehicles have been discussed, and designed, for decades, but it wasn't until DARPA offered millions in three successive contests, starting in 2004, that the autonomous vehicle industry really took off. With hundreds of companies now involved in vehicle automation, the U.S. House of Representatives are trying to standardize and streamline rules governing self-driving cars.

Wikipedia has a lengthy article on the history of autonomous cars, starting with the first driverless car demonstration on July 27, 1925, when the Houdina Radio Control Company (formed by Francis P. Houdina, no relation to or imitation of Harry Houdini, despite Harry's beliefs) "started its engine, blew its horn and rotated through its various gears." General Motors included a reference to managing cars through radio signals in To New Horizons (YouTube; Archive.org), their 23 minute documentary on their 1939 Worlds Fair Futurama exhibit. But the exhibit designer, Norman Bel Geddes, had another idea.

In his book from 1940, Magic Motorways (Archive.org), he has a chapter titled "Eliminate the Human Factor in Driving," and after describing the advances in vehicular safety, he muses:
But how about the driver? Has he too improved in these thirty years of motor-car experience as the car has improved? Not by any means. He is still, day in, day out, on three million miles of road, the same, as bad a driver as the fellow who drove a Chalmers in 1910. His eyesight is no better, he reacts no faster, he doesn't think any better, he gets drunk just as easily, he is just as absent-minded.
In the next chapter, titled Separated Lanes of Traffic, Geddes offered two technical solutions:
Within the field of science there are many potential devices which could be developed to fit exactly the needs of traffic control as they have been defined here. One of these, for instance, the radio beam, is now being used in a limited form for the guiding of the airplane on its course. The field of electromagnetic emanations, which cover a very wide field of electric-wave impulses, is probably the best adapted to the control of traffic.It is conceivable that a control operating directly — as a radio beam, broadcast from stations located along the highway — could provide the control desired. Or perhaps more simply, an electrical conductor imbedded within the road surface, carrying an electric current producing an electro-magnetic field, might provide direct control.
Roadway-guided cars were the vision of "autonomous vehicles" from the 1939 World's Fair through the 1950s, even into the 1960s in the US and the UK, where the UK's Transport Research Laboratory put magnetic sensors in a Citroën DS19 to test a driverless car on a still-wired (as of 2001) roadway. The push-button driverless car was a common dream depicted in such midcentury utopian artifacts as 1958′s Disneyland TV episode “Magic Highway, U.S.A.,” though it's the last 15 minutes that really get futuristic.

The idea of the road telling the car where to go was attractive, but over the years, cars got "smarter." Self-contained autonomous systems began to exist in the form of function-specific automation, first with cruise control (not a new concept, but the modern iteration was invented by Ralph Teetor, who was completely blind), commercially available in 1958 (in addition to the optional Safeguard Sentinel Lighting that could turn on headlights in low-light situations), and then with Anti-Lock Braking Systems (ABS), available in 1971.

With Shakey the (wheeled) Robot (24 min documentary from 1972; Wikipedia), Stanford researchers developed "the world's first electronic person" for ARPA, renamed Defense Advanced Research Projects Agency in 1972. DARPA funded a significant portion of autonomous land vehicle development, as documented in A Brief History of Unmanned Ground Vehicle (UGV) Development Efforts, published in 1995. With the "failure" of Shakey to achieve significant autonomy, there was a lull in funding and research for a few years.

In 1977, S. Tsugawa and his colleagues at Tsukuba Mechanical Engineering Lab, Japan, used two cameras and analog computer technology to process images to move the vehicle at 30 km/h by tracking white street markers on the road for up to 50 m, but was also aided by an elevated rail. Two years later, a "seeing robot rover," the Stanford AI lab cart, took was able to autonomously travel across a chair-filled room without human intervention, in about five hours. The cart moved in one meter spurts punctuated by ten to fifteen minute pauses for image processing and route planning, as seen in this short video (.MOV - warning: strobing lights).

In 1986, Ernst Dickmanns was able to capitalize on improved computers and drastically improved the sight (and a basic brain) in autonomous cars. Daimler-Benz wanted to celebrate their 100th birthday with an autonomous vehicle. Dickmanns' team's proof of concept was the VaMoRs:
VaMoRs was a 5-tonne Mercedes van equipped with cameras and other sensors, modified so that all necessary controls — steering wheel, braked and throttle. The software largely looked at the white lines on the road, and major colour differences in images.

VaMoRs made its first autonomous drive in 1986, for safety reasons taking place on streets without traffic. Due to the incredibly slow (at least by today's standards) processing speed of the computers the team were using, Dickmanns had to come up with a way to navigate in real time using a computer that processed images in a matter of seconds, rather than nanoseconds or even milliseconds. He and his team called this the '4D approach'.
...
The VaMoRs made its move onto public roads just a year later in 1987, driving autonomously on the Autobahn at speeds of up to 96km/h — the maximum speed the van was capable of.
VaMoRs was succeed by VaMP (Versuchsfahrzeug für autonome Mobilität und Rechnersehen or Experimental vehicle for autonomous mobility and computing), now operating in a modified 500 SEL Mercedes. These fell under the Eureka PROMETHEUS Project (PROgraMme for a European Traffic of Highest Efficiency and Unprecedented Safety), which also included a Jaguar modified by Lucas, seen here being demonstrated in 1994.

Back to DARPA: a multidisciplinary team was brought together in 1986 to develop a ALV (Autonomous Land Vehicle), capable of navigating difficult terrain. Some of those sub-teams had been working on their respective fronts for years, and would continue to do so. Case in point ...

Since 1984, The Carnegie Mellon University Navigation Laboratory, or CMU Navlab, has developed a series of robot cars, vans, SUVs, and buses. In 1995, technology had advanced that fewer, less expensive computers were needed, transitioning to a SparcStation LX used for used for on-road navigation experiments including autonomous lane keeping, lateral roadway departure warning and support, and curve warning. The team's confidence in the system was so great that Navlab 5 is in the Robot Hall of Fame for being the first autonomous vehicle to travel across the U.S., steering for 98 or 99 percent of the 3,000 mile trip from Pittsburgh to San Diego, while the the throttle and brake were still managed by people in the vehicle. Alan Alda took a ride with Navlab 5 for Scientific American Frontiers, where Dean Pomerleau talks about the limitations of the system at the time.

The United States Congress must have felt that the time was right to support automated vehicles in a broad way when they instructed USDOT to "demonstrate an automated vehicle and highway system by 1997" as part of the Intermodal Surface Transportation Efficiency Act of 1991 (ISTEA) Transportation Authorization bill. In 1994, Nita Congress wrote an article for the Federal Highway Administration (FHWA), titled The Automated Highway System: An Idea Whose Time Has Come, in which she wrote:
In an automated highway system, the car will be guided by the road rather than by the driver. Sensors and communication devices will link the road and the vehicle to maximize driving performance. Driver error will be reduced and ultimately, with full implementation, eliminated.
...
Although AHS represents a long-term effort, perhaps the most exciting aspect of it is that the technology is ready now. The technology to automate routine driving functions exists and will be demonstrated in 1997.
ISTEA directly lead to the creation of the National Automated Highway System Research Program (NAHSRP), then the U.S. Department of Transportation launched the National Automated Highway System Consortium (NAHSC) in late 1994. NAHSC exhibited the state of the art in vehicle automation starting in August 1997, with Demo '97: Proving AHS Works (10 minute demo tape). Unfortunately the Congress of 1997 disagreed, cancelling the program. "This decision was the result of both a shortfall in research funds and the shift of U.S. DOT’s priorities to promoting adoption of near-term, safety-oriented technologies." (PDF -- An Overview of Automated Highway Systems (AHS) and the Social and Institutional Challenges They Face, by Sanghyun Cheon)

The US Government funded three military efforts known as Demo I (US Army, 1990–1992), Demo II (DARPA, 1992–1998), and Demo III (US Army, 1998–2002), as documented in detail in the Historical Perspective appendix in Technology Development for Army Unmanned Ground Vehicles, written by the National Research Council and published in 2002. The focus shifted from off-road applications to on-road navigation, but those two different program areas continued with the next iteration of DARPA-funded unmanned ground navigation challenges.

On March 13, 2004, 15 vehicles drove into the desert outside of Barstow, CA, with the goal of autonomously reaching Primm, NV, crossing 142 miles of desert without direct human support, to win $1 million USD. That was the DARPA Grand Challenge (previously), but the farthest any of the contestants got was 7.5 miles.

The second DARPA Grand Challenge followed 18 months after the first (previously). This time, five vehicles out of the 195 teams that entered successfully completed a 132-mile course in southern Nevada.

The third and final iteration took the challenge to urban streets (previously). It was 2007, held at the former George AFB in Victorville, CA, and
this event required teams to build an autonomous vehicle capable of driving in traffic, performing complex maneuvers such as merging, passing, parking and negotiating intersections. This event was truly groundbreaking as the first time fully autonomous vehicles have interacted with both manned and unmanned vehicle traffic in an urban environment.
The first commercially available unmanned / autonomous vehicle entered the workforce in the mining industries, with Komatsu's Autonomous Haulage System (official website with embedded video), with a small fleet of large trucks (PDF) rolling out by Rio Tinto Alcan in December 2008, which expanded to 150 trucks by 2015.

In 2010, four electric, autonomous vans left Parma, Italy, on 20 July 2010, and arriving in China three months later, having successfully navigated 8,000 miles (13,000 kilometers) [mostly] without human intervention, they arrived in time for Expo 2010 in Shanghai.
Two vans travel in line. The first uses maps and GPS to drive itself whenever possible, but a human driver is in control most of the time. The second van uses its cameras and navigation system to follow the first; it visually tracks the lead van, plans a trajectory in real time, and generates controls for steering and accelerating or braking. (If a car gets in between the two vans, the second van guides itself using GPS instructions it receives from the leader.)
...
Even with all the planning, some problems are unpredictable. During a recent stretch of the trip, the convoy found itself in the middle of the notoriously bad traffic of Moscow. Because of the congestion, a two-lane road had three rows of cars. The van's system insisted in staying on its lane, so the researchers had to turn to manual driving.

“It was too dangerous to drive lawfully!” [ Alberto Broggi, VisLab's director] says with a laugh.
This method is often called platooning, technology promoted as a mid-term solution for increasing freight truck efficiency and safety.

But what does autonomous even mean? Different publications in various countries use differing terminologies, and 2025 AD has compiled four for comparison: Society of Automotive Engineers (SAE), US National Highway Traffic Safety Administration (NHTSA) (who now refer to SAE's 0 to 5 scale), German Association of the Automotive Industry (VDA) (who use a 0 to 4 scale), and the Germany Federal Highway Research Institute (BASt) (who use a 0 to 5 scale [PDF, 2012]).

Given the rate of change in these inter-related industries and laws that regulate them, your best resource for a broad overview of current events might be the lengthy Autonomous car Wikipedia article.
posted by filthy light thief (87 comments total) 52 users marked this as a favorite
 
Given the problems inherent in long-haul trucking, the platooning system seems promising. You could have the vehicles switching off in the "lead" role at rest stops to ensure that every driver has adequate rest time.

I'm skeptical of the general-case implementation of self-driving cars. There are just too many edge cases to be for any such system to be truly safe on the roads, at least within the bounds of modern control systems. Maybe there's a huge breakthrough around the corner, but I doubt it.

More practically, I think we might see self-driving in highway-only situations. Basically, the more constrained the driving environment, the more likely that you can have a fully self-driving car that operates safely. Eliminate the edge cases by only engaging the self-driving system on single-mode roadways, rather than mixed-use roadways where the driving algorithms have to account for pedestrians and bicyclists.
posted by tobascodagama at 1:42 PM on September 7, 2017 [5 favorites]


Absolutely amazing post.
posted by Pope Guilty at 1:49 PM on September 7, 2017 [11 favorites]


I'm gonna need an autopilot to read all those links!
posted by notyou at 1:53 PM on September 7, 2017 [4 favorites]


Platooning is pitched as a system that could be integrated into any truck, so truckers could hop on and off of a platoon at any time, with other incentives built into individual platooning systems, as the second truck gains more benefits (higher fuel efficiency, less need to pay attention to the road).

Regarding safety, I'm interested to see if there are reports or projections of the potential for broad safety improvements compared to current crash statistics. It seems most people are focused on autonomous vehicles being 100% perfect, while I think you could easily sell driverless cars and trucks by comparing them to current motor vehicle death rates, where we have around 100 fatalities per day in the US. Cut that number with any significance, and I'm sold.
posted by filthy light thief at 1:53 PM on September 7, 2017 [8 favorites]


i find it hard to believe that autonomous vehicles will end up being worse for people who walk and bike than american drivers are, but i know that we as a culture will give it our best shot
posted by entropicamericana at 1:54 PM on September 7, 2017 [5 favorites]


This post came about because someone casually stated "DARPA's desert and city driving challenges are why autonomous vehicles are a big deal," which lead me down this particular rabbit hole.
posted by filthy light thief at 1:57 PM on September 7, 2017 [9 favorites]


Excellent post! I don't doubt we will see autonomous cars to a limited extent, but I wonder how they will ever handle things like winter driving conditions.
posted by fimbulvetr at 1:57 PM on September 7, 2017 [1 favorite]


I just wonder were another half billion tyres will come from?
posted by clavdivs at 2:06 PM on September 7, 2017


On one hand I think some of the rosiest (or alternately, fearmongeriest) predictions of autonomous vehicles upending the economy in 5 years are much, much too optimistic about the development cycles of complex technology. I am reminded of AI predictions in the 80s.

On the other hand, the idea that the problem space is just "too complex" for machines also strikes me as shortsighted. And also underrates our ability to work on both sides of the equation: if autonomous vehicles can't handle the edge cases, at some point you just remove the edge cases from the environment.

That might sound ridiculous, but consider how much (trillions of dollars, probably) are spent modifying the environment in order to reduce or eliminate edge cases that human drivers and current vehicles are not good at handling. Everything from the road surface itself, to guardrails, to restrictions on vegetation and building construction near roadways, to the way we bank corners and add reflective tape and prisms, to the legal environment (speed restrictions, speed control devices, laws against jaywalking and against non-automobiles) are designed for the facilitation of a very particular implementation of the technology. We tend not to think of how much the built environment depends on the limitations of a particular technology, because we use the technology so often we just accept its limitations as though they're natural laws.

If the economic incentives are great enough, I'd expect dedicated lanes for autonomous vehicles, and perhaps significant modifications over time to roadway environments to make them more navigable by them. Which could eventually squeeze manually-driven vehicles off the roads, just as automobiles in their popularity drove pedestrians, cyclists, horses, and basically everything else off the roads for their own benefit. If the public decides that's what they want to ride around in, very little is sacred.

I've always been a little surprised that the guidewires or magnetic lane departure technology never really took off. Burying a wire in the road would probably not cost much, compared to the now-common reflective prisms, or all the heavy wire and steel guardrails that are basically standard near high-speed roads. The problem is not on the infrastructural side. I think the problem is positioning it as a safety device: the automotive industry is very slow-moving when it comes to rolling out safety equipment, so something that was high-tech in 1947 is probably right on time for serious consideration in our generation — airbags were from about the same period, and ABS was first put on aircraft in the 1920s.

Though at this point, magnetic field sensors and buried guidewires might not be the best / cheapest way to do it; plopping down more of those reflector prisms (optimized for IR rather than visible light, perhaps) in the middle of lanes instead of the edges might be more effective and cheaper, at least in areas without snowplows.
posted by Kadin2048 at 2:08 PM on September 7, 2017 [12 favorites]


The scenario that makes me think this stuff is way farther off than the manufacturers would have us think is this one: an autonomous car is driving down a city street, when two smaller wheeled vehicles suddenly roll out into the road in front of it. One is a shopping cart, the other is a baby carriage. The situation is such that the car can avoid one of them, but in doing so will definitely hit the other. How does it decide?

That's obviously a pretty contrived, extreme case but it illustrates a point. Anyone who drives in a city encounters similar sorts of situations every day. Humans are (generally) really good at handling stuff like this—we have no problem differentiating between a shopping cart and a baby carriage and prioritizing appropriately. However, that kind of thing is super duper context-dependent and I've never seen anything that makes me think an AI would be anything other than awful at dealing with this kind of stuff.

Maybe on highways, as tobascodegama says. But I really think that this is one of those situations where we can get 90% of the way there, but that last 10% is going to be a killer.
posted by Anticipation Of A New Lover's Arrival, The at 2:22 PM on September 7, 2017 [1 favorite]


I have my doubts about self-driving cars - first, because the environment doesn't need "no driver" cars; we need "multiple-passenger" cars to reduce emissions; next, because I am not looking forward to the rounds of legal dodgeball as owners, passengers/not-drivers, designers, and software devs all try to evade responsibility for whatever the first high-profile horrific accident turns out to be.

Mostly I can see them failing when people realize that an auto-driven car will not break the legal speed limit no matter how empty the road is, and won't drive closer than the recommended distance, no matter what that is.

People can lobby the car companies and legislature to change those limits, but once they exist, the cars are going to follow them. No more bumper-to-bumper 80 mph zipping down the freeway. Commuting with a self-drive car will be reliable, but you won't be able to shave 15 minutes off the trip if you got a late start.
posted by ErisLordFreedom at 2:22 PM on September 7, 2017 [6 favorites]


Also, autonomous cars will need to have a system whereby they can recognize a turtle trying to cross the road, stop, and deploy a robotic arm to pick the hapless reptile up and gently deposit it on the side of the road closer to its destination.
posted by Anticipation Of A New Lover's Arrival, The at 2:23 PM on September 7, 2017 [11 favorites]


As someone who literally drives to multiple appointments a day for a living, I would be totally fine with no more "bumper-to-bumper 80mph zipping down the highway." Unless your commute is over two hours you're shaving five minutes at best with those kind of shenanigans, and that kind of craziness is how traffic jams (and pileups) happen. If people would just chill out a bit we would actually all get where we're going much faster and much more safely.
posted by Anticipation Of A New Lover's Arrival, The at 2:26 PM on September 7, 2017 [14 favorites]


In the unlikely event of them becoming technically feasible in the next few decades, once again what problem are we trying to solve?

Road deaths? I suspect the money would be better spent on driver training and law enforcement.
posted by GallonOfAlan at 2:27 PM on September 7, 2017


Road deaths, and needing to spend hours a day doing something that is both super boring and super dangerous. I would love to be able to hang out on Metafilter while driving from job to job, or take a nap while zipping up to the mountains for an early morning hike. And of course in freight, the freight companies would love to not have to pay drivers and to be able to run their trucks 24 hours day instead of the legally-mandated 12 hours that is currently the most a trucker is supposed to be allowed to drive for.
posted by Anticipation Of A New Lover's Arrival, The at 2:31 PM on September 7, 2017 [14 favorites]


Everything I need to know about autonomous vehicles I learnt watching Logan.
posted by signal at 2:33 PM on September 7, 2017


Anticipation Of A New Lover's Arrival, The: " autonomous cars will need to have a system whereby they can recognize a turtle trying to cross the road, stop, and deploy a robotic arm to pick the hapless reptile up and gently deposit it on the side of the road "

The turtle lays on its back, its belly baking in the hot sun, beating its legs trying to turn itself over, but it can't. Not without the car's help. But the car's not helping. Why is that?
posted by signal at 2:36 PM on September 7, 2017 [12 favorites]


SDC's are here. On the road every day in certain cities. SDC trucks have traveled long distances.

What problem, take the trucking industry, say you have a widget to sell them that would shave 5% in cost and speed the load 10%? You'd be our next billionaire. Now look at the percentage cost of wages and that silly requirement that a driver has to sleep. Robots don't sleep.

It's far from clear the ultimate value but if you need to choose a crowded bus for $1.75 or an SDC that picks you up at your door for $2.50 or free if you agree to be dropped at a restaurant of an advertisers choice? Hmm, nudge nudge?

Yes I'm a google-sdc-fanboi, but having driven a metric bunch load of miles, ah, gratis, I would soooo like to spend the time with the kindle ;-)
posted by sammyo at 2:37 PM on September 7, 2017 [1 favorite]


Excellent post.

It really illustrates that we have chosen to take the absolutely hardest route to solve this problem. We could have had quite a bit of this in the 70s with analog systems if we had really wanted to do it.

My desires for autonomous cars are pretty easy: I want to be in control on city streets, and I want to hand off to the ai on the freeway. Combine with an electric car and Bob is my uncle.

The solution to the baby or shopping cart is that the car will see it earlier, react faster, and stop sooner so neither one gets hit. Whenever these threads come up everyone suddenly turns into a professional driver instead of the distracted, flawed creatures we all are.
posted by BeeDo at 2:37 PM on September 7, 2017 [9 favorites]


How do we know Bob isn't your uncle, BeeDo?
posted by Anticipation Of A New Lover's Arrival, The at 2:40 PM on September 7, 2017 [1 favorite]


I would be totally fine with no more "bumper-to-bumper 80mph zipping down the highway."

So would I, but I expect that early adoption will be significantly delayed by people who want both control over their speed and the ability to ignore the road. And I'm not sure that they work as "interesting novelties" the way electric cars do - I think to be safe, they need to be on automated-only roads (or at least lanes). Mixing automation with regular drivers sounds like a fast way to disasters, especially when you consider that some people are going to be looking for ways to "trick" the automated cars into swerving or worse.

There may be corporate adoptions - shipping companies especially - but I don't expect companies like Uber to pick it up; Uber survives by not having to cover parking and maintenance costs for the vehicles.
posted by ErisLordFreedom at 2:41 PM on September 7, 2017 [1 favorite]


I suspect the money would be better spent on driver training and law enforcement.

Sure it would. But it's not, and it's not going to be. Everyone knows this; we are apparently comfortable as a society with our roads being a slow-motion bloodbath, and are not going to fix it simply for safety's sake. If we were going to spend money on driver training and law enforcement we might have done that 60+ years ago when the problems first became really apparent. We didn't then, and I don't think there's been some giant awakening that makes us likely to do so now.

But if the increased safety were a side-effect of something that as a society we apparently do want (like cheap shit and monster suburban tract houses with hourlong commutes), and more importantly the market is actually going to pay for, well, I'll take the safety.
posted by Kadin2048 at 2:44 PM on September 7, 2017 [3 favorites]


Adoption will be sooner than a few decades but there are forces that are slowing adoption other than the "code". Sensors are not at an economies of scale, the best are available for pretty big bucks ($8-15k) with over six month backlog. Serious research into a semiconductor LIDAR device, the current LIDAR sensors are a highly tuned optical devised spinning really fast. And ugly. That is probably the biggest restriction at the moment.
posted by sammyo at 2:44 PM on September 7, 2017 [2 favorites]


I suspect the money would be better spent on driver training and law enforcement.

Also this has been studied and it just does not have an effect, over a long period. People are very good at driving 95% of the time, distractions, substances, tiredness just happens. (the 95 is made up but it's a pretty good guess or we'd all be dead).
posted by sammyo at 2:49 PM on September 7, 2017 [3 favorites]


a) i think removing the edge cases for light rail would be cooler

b) i think I've said this before: you ever see the drop off line for an elementary school? they're not gonna work, for any lower middle class reality definition of work.
posted by j_curiouser at 2:50 PM on September 7, 2017 [3 favorites]


That's obviously a pretty contrived, extreme case but it illustrates a point.

Does it though? If autonomous cars get such a contrived scenario "wrong" (and kill 1 baby) but get the much more common and less contrived scenarios "right", the net would be far fewer auto accident deaths.
posted by paulcole at 2:56 PM on September 7, 2017 [4 favorites]


My Uncle Bobness remains in a state of quantum flux. He probably maybe may or may not want to probably maybe get out of the box that may or may not exist.
posted by BeeDo at 2:57 PM on September 7, 2017 [1 favorite]


That's obviously a pretty contrived, extreme case but it illustrates a point. Anyone who drives in a city encounters similar sorts of situations every day. Humans are (generally) really good at handling stuff like this—we have no problem differentiating between a shopping cart and a baby carriage and prioritizing appropriately. However, that kind of thing is super duper context-dependent and I've never seen anything that makes me think an AI would be anything other than awful at dealing with this kind of stuff.

I have no idea why you assume that humans are good at this sort of task, considering how regularly we fuck it up. Right now, in the world that exists today, every single day 15 pedestrians are killed by drivers. Thousands of people have been put in the "kill a human or swerve to avoid" situation and failed it miserably. It's not clear to me that an AI that is otherwise capable of following the rules of the road could actually be significantly worse than humans at this.
posted by Copronymus at 3:05 PM on September 7, 2017 [7 favorites]


People are very good at driving 95% of the time, distractions, substances, tiredness just happens. (the 95 is made up but it's a pretty good guess or we'd all be dead).

Have you seen the numbers on car crashes in this country? A third of people in the US will experience a serious injury in a car crash in their lifetime. That's 100 million people. These things do happen all the time and we just sort of quietly accept that we must make continual blood sacrifices to the Car God because there is no other way.
posted by Copronymus at 3:10 PM on September 7, 2017 [11 favorites]


The scenario that makes me think this stuff is way farther off than the manufacturers would have us think is this one: an autonomous car is driving down a city street, when two smaller wheeled vehicles suddenly roll out into the road in front of it. One is a shopping cart, the other is a baby carriage. The situation is such that the car can avoid one of them, but in doing so will definitely hit the other. How does it decide?

That's obviously a pretty contrived, extreme case but it illustrates a point. Anyone who drives in a city encounters similar sorts of situations every day. Humans are (generally) really good at handling stuff like this—we have no problem differentiating between a shopping cart and a baby carriage and prioritizing appropriately.


Firstly, I think you overstate the extent to which humans are good at this kind of thing. I see all kinds of irrational, dangerous behavior from human drivers every day. It's possible that a human driver in the situation you describe is looking at their phone picking a new Spotify playlist, doesn't notice either wheeled vehicle, and crushes both the baby and the groceries.

Secondly, I don't think it makes sense to look at individual edge cases like this without considering risks in aggregate. Speeding contributed to 27% of motor vehicle crash deaths in 2015. Let's take a SWAG and assume it was a sufficient condition in half of those cases. That would mean ~4,250 lives saved just by programming cars not to speed, which is maybe the easiest piece of the whole self-driving car problem. Add in other very simple safe practices that irrational humans fail to follow all the time (slowing down in bad weather, not tailgaiting, not trying to beat yellow lights, not driving tired, etc...) and you are going to prevent an enormous number of accidents.

Self-driving cars may mean that 3 or 4 extra strollers get run over instead of shopping carts every year, or maybe the fact that the computer doing the driver has a response time orders of magnitude faster than a human means the car just stops and both groceries and little Timmy are perfectly safe.
posted by Aizkolari at 3:11 PM on September 7, 2017 [9 favorites]


BeeDo: My desires for autonomous cars are pretty easy: I want to be in control on city streets, and I want to hand off to the ai on the freeway. Combine with an electric car and Bob is my uncle.

Interesting. I find city driving to be more harrowing and stressful, and I find highway driving more enjoyable. But then I commute via city streets, so the highways are generally pleasant car trip type drives.

An interesting thought for folks who want to drive, from Sandra Button, chairman of the Pebble Beach Concours d'Elegance, via NPR (Aug. 20, 2017):
And with self-driving and electric cars in our future, I had to ask.

Are Pebble Beach's days numbered? That's the question.

BUTTON: When you think about it - I mean, if you go all the way back to the days of the horse and carriage - right? - people don't need horses in the same way we used to. I mean, everybody used to really need their horse. And there's still horse shows, and there's still places to race them or to hunter jump from or whatever people do. You know, it may turn out that way - that, you know, we're going to have great places to enjoy our cars but not in the everyday way.
posted by filthy light thief at 3:14 PM on September 7, 2017 [1 favorite]


I don't think setting up artificial thought experiments about shopping carts and baby strollers is necessary when we already know that self-driving cars are perfectly content to kill the shit out of bicyclists.
posted by tobascodagama at 3:25 PM on September 7, 2017 [3 favorites]


In the unlikely event of them becoming technically feasible in the next few decades, once again what problem are we trying to solve?

Road deaths? I suspect the money would be better spent on driver training and law enforcement.


Road deaths are a pretty worthwhile problem as they are the number one cause of unintentional injury death in the US. But besides that, I can think of several of problems that they solve:
  • Increased mobility for people who can no longer drive. With many Baby Boomers rapidly entering this demographic, this is going to be increasingly important.
  • Doubly true for people who cannot drive due to disabilities.
  • With cars as a service you have a lot less traffic, so less urban space wasted with parking spaces/garages.
  • Allow drivers to sleep, work, read or whatever.
posted by justkevin at 3:33 PM on September 7, 2017 [9 favorites]


we already know that self-driving cars are perfectly content to kill the shit out of bicyclists

they're much like regular cars in that respect
posted by entropicamericana at 3:35 PM on September 7, 2017 [7 favorites]


I always think of the long nasty stretch of I80 between Paxton and SLC, and how much safer everyone would be if the long haul truckers weren't humans who always seem to be anything they can possibly do other than pay attention to the road/hold onto the goddamn wheel/stay in a goddamn lane.

Then I get pissed off about trains and have to go have a lie down.
posted by aspersioncast at 3:37 PM on September 7, 2017


If we moved to self driving cars, the elementary school dropoff line could become a thing of the past. SDCs, vans and buses could pick up and drop off kids in groups. No more one-child-per-car malarkey! Fewer cars, no parents on the brink of a breakdown as they worry about being late to work. When fully automated, cars become more like other forms of mass transit.

With a self driving car I could go on a 12-hour road trip with my dad without having to deal with his unhinged driving. He could continue to go to work (in a location that requires a car) without endangering himself or others.

There are just so many positive side-effects of automated cars that it is hard to overstate them.
posted by grumpybear69 at 3:51 PM on September 7, 2017 [4 favorites]


'Cars as a service' just seems like a total pipe dream to me, especially in the States. Americans love to live in their cars. Not owning your own vehicle means having to schlep all of your crap in and out of the car every. single. trip. I don't see how that's not a total non-starter for everyone that has a baby, small children, does sports, uses their car for work (construction, real estate, etc etc etc...). The idea that self-driving cars will be anywhere near the end of individual car ownership in this country is a fantasy.
posted by aiglet at 3:52 PM on September 7, 2017 [8 favorites]


they're much like regular cars in that respect

Yes, which is why I'm not too keen to add "Well, it's just the algorithm, what can you do about it?" to the list of reasons why some cyclist totally deserved to die that day.
posted by tobascodagama at 4:00 PM on September 7, 2017


There are just so many positive side-effects of automated cars that it is hard to overstate them.

That's only if they are shared-use, electric autonomous vehicles, but that's not a given at this time.

It's interesting that in this discussion nobody's talking about land-use, emissions, or travel behavior. While it's true that automated vehicles could revolutionize transportation for the better - reduce emissions and congestions, less reliance on personal ownership, giving people more flexible options, etc - there's still a tendency to think of how we can "improve" our current system which is a crash course for a dystopian nightmare.

It's hard to say how things will go since this radical mobility shift hasn't happened yet, but I would be shocked if people didn't go to the 4 hour commute while they napped, encouraging more sprawl and pollution, and an even starker divide between those who can afford private automated vehicles and those who cannot. Most of the focus on the system is abstract on how it will benefit you, the individual, not the mobility for all. We can make it awesome, we just have to keep our eyes on the whole network not our individual nodes.
posted by kendrak at 4:00 PM on September 7, 2017 [4 favorites]


I have a hard time believing that manufacturers of "performance" vehicles won't deliberately juice their autonomous vehicles to be slightly more aggressive in speeding/passing/cutting corners in order to justify their douchbag clientele.

I will only accept autonomous vehicles when there's federal regulations mandating that BMWs, Audis, Lamborghinis, Lexuses, and Mercedes will drive themselves in exactly the same way as Honda Civics.
posted by RonButNotStupid at 4:42 PM on September 7, 2017 [1 favorite]


Almost no one has had a serious try at increasing AI-specific computing density. The bulk of it is still being done on graphics cards. Google's first effort at it was rather impressive though. Now that AI is the next computing frontier, it would be extremely unlikely if we didn't see substantial breakthroughs in AI computing density. I mean, if the normal Moore's law driven 3 orders of magnitude increase that we'll see over the next decade doesn't seem sufficient.
posted by lastobelus at 4:44 PM on September 7, 2017


The next time anyone hears themselves saying they're not sure if computers are powerful enough to do self driving, play any video game released this year. Then remind yourself that 12 years ago there were no computers in existence anywhere that could render that video game (with the high-quality settings on) in real time. Not even nation sponsored multimillion dollar supercomputers.
posted by lastobelus at 4:54 PM on September 7, 2017


Nobody thinks that computing power is the bottleneck here.
posted by tobascodagama at 4:57 PM on September 7, 2017


I misspoke, that should be 30 times increase not three orders of magnitude. In any case, I have yet to see a compelling case that we need a mathematical/mystical breakthrough. We didn't need one to beat Go. 25 years ago you would've had trouble finding anyone who would've thought self driving was a tougher nut to crack than Go. Turns out it is, but I am very skeptical that it's anything more than more computing density required.
posted by lastobelus at 5:13 PM on September 7, 2017


Yes, which is why I'm not too keen to add "Well, it's just the algorithm, what can you do about it?"

I don't really get this objection; the algorithm can be changed. It's human behavior that's well-nigh impossible. In each case of an autonomous vehicle causing a fatality, we could—if the interest existed, and one would hope it would—do a root cause analysis, figure out why the collision happened, decide how it should have been avoided, write a fix for that edge case, patch the software, add it to the list of a few million other edge cases that we'd run every new patch through, test the hell out of it, and do some phased rollout to the actual vehicles on the road. (If Congress really does feel like they need to get in on the autonomous-vehicle thing, making sure that the auto manufacturers can't say "oops sorry your car is unsupported, we only do software updates for 3 years" would be a good place to start. As would making it a crime to operate a car with software that hasn't been brought up to date.)

Sure, that process assumes a lot of sanity and good management, but it's all within the real of feasibility. It's not even the hard part of the problem; it's something we know how to do already. (To head off a lot of potential "lulz but software sucks harhar" comments: consumer and enterprise software sucks because most people aren't willing to pay for high-quality software. Which is why developing autonomous vehicle software should be handled the same way that Boeing does software for its flight control systems, not the way Microsoft creates the newest manure pile of Windows.) The very few times when modern aircraft have crashed in a way that's attributable to a software defect, the defect has gotten fixed—that's one of the good things about having systems that use software, as opposed to older generations of hardware logic.

In contrast, good luck trying to "patch" the drivers on the road. If you get run down by a person who's just too distracted to not drive their fucking car over you, well, that's just the human condition. Nothing to be done. People get tired, after all. And they do dumb things, and then they kill each other. And this is something we just totally accept. In a way that we do not tend to accept deterministic systems, or machines, killing people.

So while I don't really want to get run over in either case, if I have to go by becoming a red stripe on the pavement, I'd much rather go in a way that leads to a bunch of engineers sitting in a conference room and figuring out how to make sure that it never happens again, and not in a way that leads to a bunch of lawyers shrugging and basically saying to my family "welp, here's a few bucks, sorry for your loss, best we can do."
posted by Kadin2048 at 5:18 PM on September 7, 2017 [17 favorites]


I'm also really skeptical that "what Americans like" is going to somehow stop self driving cars. Who cares what Americans like! If the rest of the world switches to self driving, and you don't, how do you expect to be competitive with 3-500 hours less per year per person and $1k to $5k higher capital cost.
posted by lastobelus at 5:18 PM on September 7, 2017 [1 favorite]


I am very skeptical that it's anything more than more computing density required.

There are (at least) two major issues:

1) Getting the algorithms right, so they recognize other things on the road and fail to crash into them - that part's do-able; we'll get there. (I am not certain that we can get there on a road containing both human-driver and computer-driver cars, because there may just be too many variables.) and

2) Coping with troll attempts, wherein some pack of teenagers decides to "fool the car." That part's less likley to be fixable via computer, even with excellent AI.

How do you put out a call for bug testing that's basically, "please try to discover all possible methods to get this ton-and-a-half 50 MPH device to swerve uncontrollably into a crash, so we can fix the coding so they can't happen?"
posted by ErisLordFreedom at 5:22 PM on September 7, 2017 [3 favorites]


Those are issues, and will require lots of work and engineering iteration. But that's not the same as saying they require breakthroughs.
posted by lastobelus at 5:30 PM on September 7, 2017


Eliminate the edge cases by only engaging the self-driving system on single-mode roadways, rather than mixed-use roadways where the driving algorithms have to account for pedestrians and bicyclists.

At least in parts of the west, bicycles use sections of freeway. I've also seen people pushing wheelbarrows, bicycles covered in bags of stuff, and pulling homemade wooden carts. Still probably easier for a computer than a city street, but plenty of weird stuff to deal with even so.
posted by Dip Flash at 5:37 PM on September 7, 2017 [1 favorite]


About the fact that the risks of self-driving seem somehow intrinsically scarier (a machine deciding who lives!, etc.): this is just a temporary artifact. Humans put unfamiliar risks in a completely different category than familiar risks. Many (most?) people have trouble even emotionally recognizing familiar risks as real risks. As soon as it starts to become familiar, the public's feelings will shift dramatically. Humans have a lot of trouble properly weighting a low-likelihood unfamiliar risk against a high-likelihood familiar risk. They do a lot better at weighting when both risks are familiar.
posted by lastobelus at 5:39 PM on September 7, 2017 [6 favorites]


I can see this thread is going to go about as well as trying to convince Elon Musk that Skynet won't kill us all, so whatever.
posted by tobascodagama at 6:38 PM on September 7, 2017 [1 favorite]


Kadin2048: "I've always been a little surprised that the guidewires or magnetic lane departure technology never really took off."

BeeDo: "It really illustrates that we have chosen to take the absolutely hardest route to solve this problem."

Absolutely. When you get down and think about it, which makes more sense?:
  • Vehicles augmented with just enough sensors & smarts to detect & avoid hitting unforseen obstacles, all connected to a centralised control system (or set of systems, with hand-offs in-between) which knows the location, destination, and speed of every vehicle in the road and directing them all in the most efficient manner, or
  • Attempting to build all that smarts into each individual autonomously-controlled vehicle and create a completly decentralised control system, with each vehicle trying to negotiate its optimum path against the different algorithms & standards of multiple other vehicles from wildly different manufacturers, and hoping that what emerges is efficiency rather than a "fuck you, got mine" arms race where different software versions & algorithms - not only factory-standard, but aftermarket 'mod chips' and weekend-hacker tinkered - try to game the emergent behaviour to prioritise themselves at the expense of other traffic.
Yes, there's ancillary problems with both - privacy, for starters (but you'd be kidding yourself if you believed it wouldn't be a problem either way; the only difference is who holds the data - the city/state/country-wide traffic control system, or the vehicle manufacturer) - but I'd bet the first option is the simpler to roll out, quicker to expand, less able to be gamed, & vastly more efficient…
posted by Pinback at 6:38 PM on September 7, 2017 [1 favorite]


Thanks to filthy light thief for the high-quality FPP at least.
posted by tobascodagama at 6:38 PM on September 7, 2017 [3 favorites]


Yes, which is why I'm not too keen to add "Well, it's just the algorithm, what can you do about it?" to the list of reasons why some cyclist totally deserved to die that day.

Your first link talks about potential dangers of bicycle detection, which seems to ignore the fact that there are self driving cars on the road right now which have (to my knowledge) not killed a single cyclist yet. You follow it up with the quote above, which seems to reference an accident where a self driving car mass murdered cyclists. What is your agenda here? Are you deliberately referencing nonexistent accidents?

As far as I can see there has been exactly one collision between a bike and a self driving car, and in that case the bicycle ran into the car. Stop fear mongering.
posted by Literaryhero at 7:00 PM on September 7, 2017 [5 favorites]


so, yeah, the hypothetically dead baby. about that.

a) if it's a human driver, jailtime for manslaughter (driver)

b) if a robot driver, jailtime for manslaughter (ceo, board, director of engineering)

as long as we have that understanding before deployment, i'm cool.
posted by j_curiouser at 8:35 PM on September 7, 2017 [1 favorite]


Trains follow two steel rails and have a driver/passenger ratio in excess of today's SDC but are deeply not sexy due to reasons. The cynic in me thinks that society is spending a lot of capital to make sure that we do not have to share buses and trains with minorities.

There are tons of SDCs here in Mountain View. The youth at the local cinema have learned that fake charging the cars whilst on foot causes the vehicle to panic stop. Oh, the hilarity. I wonder what other fun, unintentional hilarity lurks in these vehicles? Nobody knows, but hey! we are going to find out nationwide.

The Waymo cars have 5 LIDAR, 4 RADAR, many optical sensors, ultrasound, a massive point cloud of the roadways, and tons of map data. All feeding into a really nice TPU system. The Tesla system is 8 (12?) cameras, one radar, some COTS mapping data, and some graphics cards. It is hard to believe that these two diverse systems offer the same level of performance and safety. Yet they are equally SDCs in the eyes of the public.

To be clear, I think that cars shouldn't have drivers. They're a menace! However, my preference would be to remove the cars. In the event, society is at low risk for adopting my strategy. So, SDCs it is.
posted by pdoege at 8:45 PM on September 7, 2017 [5 favorites]


so, yeah, the hypothetically dead baby. about that.

a) if it's a human driver, jailtime for manslaughter (driver)

b) if a robot driver, jailtime for manslaughter (ceo, board, director of engineering)

as long as we have that understanding before deployment, i'm cool.


Should vaccine creators be imprisoned because one child had an allergic reaction and died? I don't think this is a good rationale. If self driving cars reduce accidents by 50 percent (number pulled out of my ass), we shouldn't be stopping the technology just because someone needs to be punished in the case of an accident.
posted by Literaryhero at 9:21 PM on September 7, 2017 [3 favorites]


Literaryhero: Should vaccine creators be imprisoned because one child had an allergic reaction and died?
If they're trying to "disrupt" vaccines by pushing this week's petri dish into a million injections with zero testing? Yes, as this is unreasonably reckless.

Why do people think driverless cars will automatically reduce deaths? I think they'll increase deaths, just like cars did.

After the Florida death, Tesla claimed 130 million miles with no fatalities on autopilot (the earlier fatality in China was claimed but not confirmed as autopilot), as opposed to the 93 million miles on average per driver death in the US. Even at face value, that's a thin margin. Yet those miles were nearly all on very safe freeways and drivers can instantly intervene. Clearly autopilot is much more dangerous than an average driver.

Humans likely follow a logistic distribution where 20% of drivers result in 80% of accidents, where computers should be uniform. Even when driverless cars manage to be better than the average driver, those 20% are more likely to choose to keep driving themselves. Younger drivers can't afford these cars. Impatient drivers will disable or modify them. Whatever pushes the worst drivers into that category will keep doing so. Conversely safe drivers will attempt to drive safe cars in a safe manner, and self-driving cars will be sold as such. Therefore, driverless cars will displace better-than-average driving with average driving, and so increase death rates.
posted by netowl at 12:11 AM on September 8, 2017 [2 favorites]


A consequence of SDCs is the reformation of the insurance industry. Volvo has said that when it sells SDCs, it will insure them - having run the numbers, which show >90% decrease in accidents, it reckons that it can roll the price of lifetime insurance into the sales ticket. If those are the economics, then car insurance as we know and love it is over.

Those are the economics.

And good luck trying to drive a self-driving vehicle into a crowd of people, terrorists.

I'm reminded of a paragraph from an otherwise totally-forgotten SF short story I read as a kid; the plucky Earthmen have made planetfall and have encountered a technically-advanced, peaceful and totes hoopy alien race. But one thing's missing - no cars. Why's that, asks an astronaut. "We invented them, of course", says Tharg, "but we couldn't work out how to stop them killing people. Doubtless you've solved that one."

(I lost enough eyesight five years ago to also lose my licence. Guess how I feel about SDCs.)
posted by Devonian at 5:10 AM on September 8, 2017 [5 favorites]


[first comment, hi everyone]

I love the self-driving cars threads! The thing I want to add is that there's a massive economic incentive to get delivery vehicles, at least, driving themselves. Just the ability to drive all the time is going to make the delivery companies more profitable. No more trucks sitting at truck stops while drivers take their minimum mandated rest period. Those vehicles will drive and drive and drive and not stop, ever. UPS could do the same amount of road-oriented business with a fleet half the size.

And it's not just the delivery vehicles. Buses too. All manner of service vehicles. The incentives are just too great for these firms to not push as hard as they can for driverless vehicles. Not to be cynical, but when there's that much money on the table, it's hard to see how self-driving vehicles (at least delivery and service vehicles) aren't inevitable
posted by breezytimes at 5:17 AM on September 8, 2017 [1 favorite]


Clearly autopilot is much more dangerous than an average driver.

That isn't clear to me from that example, but it's not a stretch to think that a first-draft version of an autopilot is going to need a lot of improvement. So if this is the safety baseline from which self-driving cars are refined and improved, I think it is going to be pretty good and honestly I'd like to see us get there sooner rather than later.

Just selfishly, I would love to have long highway trips and stop/start rush hour traffic jams become situations where I am relaxed with a book or my laptop, and maybe a drink, while the car takes care of the travel. Also selfishly, I would love to be able to walk to work or walk to meet friends at a restaurant and know that every car is programmed to stop at red lights, obey walk signs, and so on, rather than just hoping that the driver is not texting, drunk, or cognitively impaired and happens to notice that a pedestrian is crossing.
posted by Dip Flash at 5:19 AM on September 8, 2017 [6 favorites]


a couple things:
cost of highways(to understand relative cost of adding sensors or magnets or something):
"What are the actual costs for projects around the nation?
Of the 15 projects studied from around the country the cost per lane mile varied from $1.9 million per lane mile for simple widening project in Virginia without interchanges and large structures to $188 million per lane mile for the Big Dig in Boston that was constructed in a high density urban area with a considerable amount of tunnels and bridges "

on congestion:
this seems best solved more HOV lanes with more penalties for single occupant cars. also maybe cars with more occupants are also safer? but i don't know if that has been studied.
the few times i was in central or south america, hitchhiking in rural places is crazy easy compared to hitchhiking here. hitchhiking "bus stops" at all on-ramps would be a wonderful thing to add to our highway system.
posted by danjo at 5:58 AM on September 8, 2017


SDC's solve the problem that humans are very, very, bad at driving and no amount of training or law enforcement is going to fix that. In the USA alone there are around 5,500,000 auto accidents every year which result in around 2,900,000 injuries and around 33,000 deaths.

SDC's don't have to be perfect, but if they can reduce that by even 10% it'd be great. If htey can reduce that by even 50% it'd be fantastic! And as Devonian noted, Volvo thinks their SDC implementation can cut that by over 90%.

Moreover, SDC's are likely to make traffic a lot less problematic. Almost all traffic flow problems are due to human drivers doing stupid shit, not due to the number of cars on the road. One driver swerves across a couple of lanes of freeway causing people to hit their breaks and the next thing you know you've got a moving traffic stoppage that lasts for miles.

A centralized system and smart roads would probably be a technologically easier way to solve the problem, but that requires fairly massive upfront investment by municipalities, state government (for state roads) and the Federal government (for interstate highways), which is why it never happened. SDC's allow for incremental replacement of human drivers rather than a single massive upfront payment by the government followed by people having to upgrade their vehicles to be compatible.

More important, while I agree that public transit is a very good thing, I think SDC's also provide a way to make public transit work better.

Finally, I work with blind people, and the hoops they have to jump through, the insane and awful public transit they have to try to work with, is isolating and expensive for them. Most have to leave for work at least two hours before work starts, and don't get home until two hours after work ends. And that's with microbus public transit that does curb to curb service, if they use the regular buses its even worse.

For elderly, blind, and otherwise physically disabled people SDC's represent freedom and independence.

And, so far, the results of SDC's look good. Google's cars have driven for millions of miles (city miles, not highway miles) and the only accidents they've been involved in were the result of human drivers hitting the Google car.

Further, we already have a very successful model for bug fixes with every SDC caused crash: the airline industry. Every airline crash involves a forensic engineering team going over the crash with a fine tooth comb and figuring out **EXACTLY** what went wrong and engineering that problem out of airplanes so it doesn't happen again. As a result airplanes (in the first world anyway) are the single safest means of travel that exists. We can do the same with SDC's.

More trains would be great, I'm 100% down with more trains. And I think that SDC's and car as a service will make passenger rail in the US vastly more appealing. Right now the problem with passenger rail in the US is the same as the problem with air travel. You get to your destination and then you're in an American city that was designed around cars and has shitty public transit because basically the city wasn't designed with public transit in mind. Car rentals and taxis cost a fortune. But car as a service would, presumably, be cheaper and thus make intercity travel by public transport significantly more attractive.

Moreover, I think car as a service is a good solution to the problem of public transit in American cities that were designed all but explicitly to make public transit unworkable. Computer pathing can easily plot a path that gets you to your destination quickly and also picks up additional passengers. People could pay a premium to get to their destination without others in the vehicle, or pay less and get there a few minutes later by sharing the vehicle.

But to my mind the three biggest problems SDC's solve are the 33,000 people who die from auto accidents in the USA every year, giving more independence to people unable to drive, and making a commute less hellish by letting me read or play games or do something other than spend my attention driving.
posted by sotonohito at 6:04 AM on September 8, 2017 [8 favorites]


A centralized system and smart roads would probably be a technologically easier way to solve the problem...

A pretty hefty proportion of the work on automated vehicles in the UK (or rather a pretty hefty chunk of the government innovation funding budget) is around the idea of Connected and Autonomous Vehicles (or CAVs).
I think this is very much the right direction, because a lot of the biggest benefits come not from the driverlessness but from the connectedness. It allows you to do real time traffic analysis and rerouting (of course even more so with autonomous, but still possible with manual) and things like the aforementioned platooning/fleeting behaviour or on demand vehicles etc. etc.
Making the road and the car work together as part of a smart system is the idea.
posted by Just this guy, y'know at 7:22 AM on September 8, 2017 [2 favorites]


There's already some rudimentary connectedness and centralized route planning in regular use: Google Maps and especially Waze. Those systems already use position and density data from participating devices to generate efficient routes and relay that back to individual cars. With enough participating cars, the system could maximize road use, sending some cars on the highway, others on the frontage road, etc.
posted by notyou at 7:35 AM on September 8, 2017 [1 favorite]


Clearly autopilot is much more dangerous than an average driver.

That's not really clear at all IMO, but even if we were to stipulate that right now, the trajectory of autonomous vehicles is likely quite good; human drivers have flatlined, in terms of performance, generations ago. As I said earlier in the thread, one of the biggest benefits of automation is its ability to not make the same mistake twice.

I feel strongly that we should be nudging the market towards solutions that facilitate the sharing of control logic and rules, which the market may not necessarily do on its own, but even if we fail to do that and have to depend on the much-slower diffusion of information via engineers moving from one company to another, rather than the fast diffusion of actually sharing source code, it's still improvement over time. People, by and large, do not improve. Someone today is basically as good a driver as someone in 1960, who was as good as someone in 1910, who was probably no better or worse than Ug The Caveman, if you gave him a car and made him sit through Drivers Ed. (It's possible that drivers today are worse than drivers in past generations, due to increased distractions, but this is hidden under the effects of better safety features. Once again, the machines are improving, not the people.)

I'm also not convinced at all that autonomous vehicles are going to replace the best drivers on the road and leave the worst. There might be some of that — CDL drivers, mile for mile, are actually pretty good, despite what you might think if you've ever had a truck try to merge into you — and there's a big economic incentive to automate some of those miles, because they're also the easiest for a machine to handle. (I think there will be a long period of semi-automation with platooning and driver-supervision-machine-control, though.)

But, and this is particularly true in rural areas, the lack of an alternative to cars means that lots of people who everyone acknowledges shouldn't drive are allowed to do so. It's beyond ridiculous how many DUIs you can rack up in a rural area and yet still be allowed to get behind the wheel—and it's not just that people don't regard driving drunk as a problem, it's that there's literally no other option that doesn't render someone destitute, so they're reluctant to actually for-reals no-exceptions take away someone's license and then enforce it. In urban areas this is not so much the case, because people involved in the system know they can make some shithead take the bus (or a "DUIcycle"). I think you'll see more bad drivers pushed into autonomous vehicles once they become a valid option.

In the same way that the Japanese have a slight obsession going with "companion robots" as a result of demographic trends, I could also see a generation of aging drivers who aren't interested in giving up personal mobility and have the final spoils of their generation's looting of the economy retirement funds to burn, basically bootstrap the consumer market for cars. Younger and working people will need to do fractional-ownership or pooling, but there are a lot of people who might see a $120k first-gen SDC as an alternative to a house in a golf-cart community in Florida.

Here's your Monocle Polish Magazine™ Vile Business Model of the Day: build really shitty, just-barely-adequate autonomous vehicles — 30 MPH max, no freeways, yield to all other vehicles, remote supervision by kids in a sweatshop in India, whatever it takes — and then rent them out at extortionate pay-by-the-day terms to people who have DUI license suspensions, taking their existing vehicles as collateral against nonpayment and damage. Grease up some politicians and get MADD to help you craft some legislation to get paid directly from the state, with them extracting the money via fines and wage garnishments. (Still taking the user's car though, duh.) Use check cashing-places that are being slowly regulated out of existence as your depots, since they're in your target neighborhoods anyway. You could build in neat "features" like a hose-out all-hard-surface, vandal-resistant interior, and remote door locks for automatic enforcement of missed court dates. The vehicle sensor package can also double as an enforcement mechanism to make sure the user isn't jumping in another car in the morning, too. (They'd have to pay for the service of having someone, probably in India, watch them.) Oh, and "luxury" features like heat and AC would be extra, of course, payable on-demand from inside the vehicle. If anyone looks askance at you, call them 'elitist' and tell them you're 'democratizing' technology.
posted by Kadin2048 at 7:40 AM on September 8, 2017 [4 favorites]


Just this guy, y'know I think in the US we'll see smart roads coming after smart cars rather than the other way around. I agree both are good and that we should have gone smart road first. But you can always count on Americans to do the right thing only after they've exhausted all the other possibilities.

It's a bit like the way we did the moon shot. The **SMART** way to do a moon shot would have been to build a reusable launch vehicle (not like the Shuttle which was a compromise, I mean one without a very expensive fuel tank thrown away each launch), use it to build a decent sized space station (like crew of 20 or 30 minimum size), build a moon lander in orbit, and then use it to get to the moon and start a colony.

But that won't get you to the moon by Kennedy's 10 year time limit, and it won't beat the Reds in the space race, so instead we did it the dumbest, most dangerous, way possible. NASA didn't want to do it that way, they just were forced to by political considerations.

So it goes with SDC's. We should have spent money on smart roads first. Instead we'll do the hard part first, build really damn smart cars, and then after they've become common as a sort of afterthought we'll make the roads smart to make it easier for the cars.
posted by sotonohito at 8:26 AM on September 8, 2017 [1 favorite]


we already know that self-driving cars are perfectly content to kill the shit out of bicyclists.

Do we know that? The article discusses simulations of systems that are far from the best. The only real life example I found was an SDC *rear ended* by a cyclist tailgating. Really are we going to blame irresponsible actions on the software? (probably)

I saw the video shown by the former head of the google program, a messy intersection at night with bike and he suggested that if he were driving in place of the SDC he'd probably have hit the bike. That was a real life example.

The existing google level SDC's would save thousands of lives now, today, but a single edge case that is barely the software fault will certainly make headlines, cause célèbre.
posted by sammyo at 8:38 AM on September 8, 2017 [2 favorites]


Trains follow two steel rails and have a driver/passenger ratio in excess of today's SDC but are deeply not sexy due to reasons. The cynic in me thinks that society is spending a lot of capital to make sure that we do not have to share buses and trains with minorities.

Trains are deeply not sexy? Really? Have you ridden CalTrain? Or the LIRR? Or Amtrak? Or the NYC subway? Or the Philadelphia regional rail? Trains are an extremely popular form of transit whose primary limitation is the real estate, easements and cost required for track construction. It is not an either/or between trains and SDCs, as SDCs are more likely to replace buses, which, yes, carry a lot of social stigma.
posted by grumpybear69 at 8:50 AM on September 8, 2017


I mean, if SDCs really take off, there's going to be an excess amount of lanes on the interstate. Maybe we'll end up with trains and busses afterwards? With the amount of people urbanizing, it'd be great to infill parking lots and change the parallel parking spots into bike lanes.

Imagine what changes a place like Atlanta would undergo.
posted by Trifling at 9:02 AM on September 8, 2017


We do have smart roads by default now, though. If you're driving a Model T down a dirt track with a mobile phone in your pocket and you've got cell coverage, you're already highly visible to the global AI and it can talk back to you. It can know where you are, it knows about the road you're on, and it has real-time information coming in from other road users in your area. The precision, speed and latency for all this will only improve.

Moreover this sort of smart doesn't need you to dig up the road to upgrade it, which is a big problem if you're actually embedding stuff - and in places where they can't be bothered to keep the bridges to code, then spending millions (billions - there's a lot of road) for something that everyone knows will be obsolete in a decade is not going to happen. In an ideal world, well, you'd invest in an upgradeable infrastructure design first, with lots more up-front costs before any experiments can even start, and you'd carefully plan and iterate your way though complex cost-benefit analyses stretching far into the future, and wonders would result.

This is not our world. Ask the Soviets; they tried to make it their world, and it didn't happen. NASA went to the Moon because of the very factors that dictated how they'd do it: take those away and you don't get your sane reusable flexible baseline spacegoing infrastructure. You don't get to the Moon either.

We are getting SDCs now because the factors that make them feasible have come good - Moore's Law, basically, operating over fifty years. As the FPP makes clear, there's never been a lack of vision to get here, but the path is littered with failed experiments. It turns out that, just like the Internet, the technology that makes it happen demands a very great deal of decentralised autonomy working to open rules within an ad-hoc and agnostic communications infrastructure. We're only just learning how to purposefully work that way.

As for 'are SBCs safer', the data is hugely yes. The same is true of aviation automation, where fully automated passenger aircraft would improve things stiff further - most incidents and accidents are pilot error or worse - but wouldn't be publicly acceptable. Such research as is being done is disguised as working towards single-pilot commercial flight in order to avoid pre-emptive controversy, but as one of the reasons for dual-pilot standards is to guard against one of them becoming incapacitated, there's really no such thing as commercial single-pilot focussed automation.

One of the knock-on effects of SDCs becoming commonplace may be a softening of public attitude to fully automated aviation. Which would be a good thing.
posted by Devonian at 9:16 AM on September 8, 2017


we already know that self-driving cars are perfectly content to kill the shit out of bicyclists.

Do we know that? The article discusses simulations of systems that are far from the best. The only real life example I found was an SDC *rear ended* by a cyclist tailgating. Really are we going to blame irresponsible actions on the software? (probably)


Yeah... the problem isn't that automated vehicles will kill cyclists and pedestrians as much as human operated cars, it's that they actually obey traffic laws which is annoying for everybody else who doesn't. People are arguing that AVs need their own traffic laws, and that will probably come with regulation. Congress just took steps in that direction this week. Even still, there's no way it would be legal for the manufacturers or automakers to design cars that put bicyclists and pedestrians at risk the same way typical human drivers do. That would be a huge liability issue and really stupid.

/a regular pedestrian and bicycle commuter who indexes connected and automated vehicle research
posted by kendrak at 9:23 AM on September 8, 2017 [3 favorites]


Longer term a whole bunch of infrastructure will be changing in ways it's unlikely that the best futurist could possibly imagine. Streetlights, gone. Crosswalks probably gone. Gas stations, well any that a non-engineer could find, gone. Parking lots, moved to a radically different structure but regular people will never need to find one. No more parking meters!!! Emergency vehicles traveling full speed (an old Dick Tracy cartoon visualized the busy city streets clearing instantly, well people react slow) a system that sends a message to cars to clear could actually do that.

Like the ragamuffins in the valley "playing" tag with the google cars, the social change will be slower than the technological, but look at the super computer (ya ya almost) everyone caries around to tweet, tech continues to ramp up and change whats possible. First insurance will become prohibitive to any without an absolutely perfect record, then the car insurance industry will evaporate. What's after that?
posted by sammyo at 10:11 AM on September 8, 2017 [1 favorite]


Assassination made easy. Freedom of travel revoked. No explanations, you can't travel that road, to where the:
1. Rich live
2. infrastructure exists
3. People of one color live
4. People of one belief live
5. Land crimes are ongoing
6. The border is close by
7. Experiments are carried out
8. Where the locals claim land that is not theirs
posted by Oyéah at 10:16 AM on September 8, 2017


You could if you took a bike.
posted by Devonian at 10:55 AM on September 8, 2017


There's an interesting part in David Marusek' Mind Over Ship where a bunch of distributed algorithm augmented wheeled pedestrians become caught in a traffic standing wave, where they go around a roundabout without anybody being able to leave it, and the bots in charge of keeping order yell at everybody to calm down, and of course they don't, some panic, and start clumping up in groups where the ones in the middle get crushed to death as en emergent behavior of the system, others spin around fast enough to pass out, etc.
Sounds like fun.
posted by signal at 11:07 AM on September 8, 2017



It's beyond ridiculous how many DUIs you can rack up in a rural area and yet still be allowed to get behind the wheel

When I lived in Arkansas in the 80s, I knew of several people who had never gotten licenses, because the penalty for driving without one was lower than the penalty for driving after a DUI - and of course they weren't going to stop drinking and driving. (Arkansas has a whole bunch of "dry counties" where booze can't be purchased, so you have to drive 50+ miles away to find a bar; if you're driving that far, you're not going to drink one beer and head home.)

Really are we going to blame irresponsible actions on the software?

Sometimes, yes. The person driving 45 MPH on the freeway can cause plenty of accidents without being the one who hit other cars. The blame for an accident isn't "which car's front hit another car;" it's "which car failed to follow the rules of the road AND notice changing circumstances in time to react to them." A car that always follows the law can still cause deaths because it wasn't making adjustments for "hey, the car next to me is being erratic and swerving a lot, so I'll just... slow down so it can not be next to me anymore." An SDC will need a lot of special programming to deal with "nearby car/bike/pedestrian isn't breaking the law but is acting in a way that could be dangerous."

A human driver can notice "kids playing kickball next to the street; watch out for someone being stupid." I'm not sure SDCs have that built in.

if SDCs really take off, there's going to be an excess amount of lanes on the interstate. Maybe we'll end up with trains and busses afterwards?

Why would anyone think this? Are we assuming that SDCs will make carpooling more likely? (I'm not - as much as I'd love SDCs to replace buses and encourage lots of 3-5 person shared transit, I expect most of the push is coming from rich white dudes who want a hands-free commute so they can be on the phone.) Or is it just that several SDCs in a row can stack up, lock their software into sync and remove the space in between them?
posted by ErisLordFreedom at 11:17 AM on September 8, 2017


> Why would anyone think this? Are we assuming that SDCs will make carpooling more likely?
That seems a very reasonable conclusion. Assuming a (ideally nationalized but let's be real, more likely corporate owned) shared car pool, there is a major economic incentive on the owners to encourage Lyft Line style transport instead of an SDC going from point A to point B with a single person. The more people that use SDCs, the more efficient this style of transport gets.

> I expect most of the push is coming from rich white dudes who want a hands-free commute so they can be on the phone.
Yes, rich white dudes are the only class of people who don't enjoy driving and like using their phone. This is a take that makes sense.
posted by ReadEvalPost at 11:51 AM on September 8, 2017 [2 favorites]


A human driver can notice "kids playing kickball next to the street; watch out for someone being stupid." I'm not sure SDCs have that built in.

Not kickball but a google presentation a couple years ago had an example that was not hard coded in. The graphic showed some cars as cubes (simple representation of data from sensors) and a smaller cube going in circles in the middle of the road with a bunch of even smaller dots weaving around. The car handled it fine. It was a woman in an electric chair chasing ducks in the street. No ducks were injured in the incident.
posted by sammyo at 12:15 PM on September 8, 2017 [1 favorite]


> I expect most of the push is coming from rich white dudes who want a hands-free commute so they can be on the phone.
Yes, rich white dudes are the only class of people who don't enjoy driving and like using their phone. This is a take that makes sense.


No, rich white dudes are the ones most likely to throw a whole lot of money into projects with the goal of "I get more time to spend on my interests, and if that interferes with anyone else's interests, they should spend money of their own."

I don't believe SDCs are mostly of interest to rich white dudes - I believe rich white dudes are working to make them happen, and not because they care about accessibility for people who can't drive, which means they're working hard to make SDCs happen in a way that's most convenient for them:

* passenger cars that look like today's 4-5 person vehicles, or even mini-versions that only hold 2, rather than minivans designed to hold 7-10 people as the default version,
* focus on freeway driving rather than inner-city shopping/pick-up-kids routes (which are harder to do safely, but the cars could be kept to 25 mph and less, and that would work fine for dense urban traffic, and the safety measures get easier at those speeds),
* personal or corporate ownership, or service/rental plans, rather than treating them like a public utility.

I see lots of articles about how terrific it'll be to call a car to take you to work, and call another one to take you home, but not about how they'll pick up 4 random other people heading to the same office. I see mentions of "send a car to pick up the kids at school" but not "send a driverless bus that picks up 20 kids and takes them all home." (Which you can't do unless you have a human adult on the bus to manage the kids - but at least that adult wouldn't need to also watch the road.)
posted by ErisLordFreedom at 1:22 PM on September 8, 2017 [2 favorites]


SDC's will also eliminate a lot of traffic problems because a lot of traffic problems are caused by human failure to drive optimally.

Merging, for example, tends to produce traffic snarls rather than perfect zipper merges. Lane changes on freeways can often be so disruptive of traffic patterns that they create miles long patterns of people coming to a complete stop then accelerating.

If you've ever been driving on a highway in a city and had the odd occurrence of traffic stopping, then resuming, and there being no apparent cause the most likely explanation is that some time in the past (usually only ten or fifteen minutes at the most) someone changed lanes in a way that caused another driver to slam on the breaks, which echoed back and amplified into the snarl you ran into. The actual cause most likely took place five or ten minutes in the past and several miles up the road. Enough SDC's in traffic would keep that sort of thing from happening and greatly speed up traffic flow.
posted by sotonohito at 1:33 PM on September 8, 2017 [1 favorite]


Many people say that long haul trucks on interstate highways will be the first economic application of self-driving vehicles because that is the easiest to solve. But I'm made some rough calculations of cost savings.

A high paid driver will earn about $70,000 a year. That works out to about $250 a day. Let's say they carry a load of beer worth $25,000. That's 1% of the cost of the load.

That's kind of hard to justify automating given the fact that most drivers don't make $70,000 a year and most loads are worth a lot more than beer.

Human drivers are still pretty cheap.
posted by JackFlash at 7:05 PM on September 8, 2017


An anecdote that may or may not mean anything. I recently went to a local auto body repair shop to take care of a minor dent. Turns out this shop is certified to repair of lot of high end cars like Maserati, Jaguar, Austin Martin, Bentley, Rolls Royce, Ferrari, etc.

I noticed there were over a dozen Teslas sitting in the lot waiting for body repairs. I asked the woman at the front desk what all these Tesla owners were doing to their auto-pilot cars.

Her laconic reply: "Wreckin' 'em."
posted by JackFlash at 7:16 PM on September 8, 2017


Tesla "auto pilot" isn't actually any sort of a self driving car. It's a feature that a lot of high end cars have had for a while now that just keeps the car in a lane and (sort of) staying with traffic. It doesn't navigate, it's not equipped with the necessary sensor package for true self driving, and it is the subject of some serious doublespeak by Tesla.

On the one hand they call it "autopilot" and give the carefully calculated impression that it's self driving.

On the other hand they explicitly state that it is not autonomous driving and that the driver is instructed to never, ever, take their attention off the road while autopilot is active.

Don't judge SDC's by Tesla's autopilot, in other words.
posted by sotonohito at 4:46 PM on September 9, 2017 [1 favorite]


Most truck drivers (at least, long-haul owner-operators) are paid per mile, not salaried, so the calculation is actually easier than that. The US BLS says most drivers get between $0.28 and $0.40/mi. Most trucks get between 6 MPH (average) and 8 MPG (new trucks), which is 0.31 to 0.42 per mile at an average price for #2 Diesel of $2.50/gallon.

So, intriguingly, fuel is actually a bigger cost component of long-haul trucking than driver labor. Electric trucks, which would remove or substantially decrease fuel costs, would be a bigger deal to the industry than removing drivers.

That's not what I expected, TBH.
posted by Kadin2048 at 10:54 AM on September 11, 2017


the vaccine analogy sux. and if we're redoing all the infrastructure for nsdc, we could redo it for anything. is a nsdc the right choice?

safety: a 10 % reduction in highway deaths has a lot of solutions. reduced speed limits, better enforcement, more stringent dl reqs, raise the driving age to 18...

i am bored of the safety argument. i think

a) nsdc would not be all the rage w wall street and the valley if they were doing it for safety. they're doing it for profit and justifying it with safety

b) corporations, as always, will be indemnified.

c) ain't gonna work. not how anyone thinks they will.
posted by j_curiouser at 9:33 PM on September 12, 2017


Intel reveals it’s been working with Google on self-driving cars since 2009 - Intel processors, ethernet, and FPGAs have powered Waymo’s cars since 2009. (Jonathan M. Gitlin for Ars Technica, Sept. 18, 2017)
Working with (Google's) Waymo isn't Intel's only foray into autonomous driving. For example, Ford went with Intel processors for its autonomous R&D vehicles. Last year, Intel announced that it was partnering with BMW and Mobileye to collaborate on a self-driving car called the iNEXT that would launch in 2021. Then in March of this year, Intel bought Mobileye for $15.3 billion and integrated its own Automated Driving Group into the Israeli company. And in August, Fiat Chrysler Automobiles revealed that it, too, was joining the BMW/Intel autonomous vehicle program.

Obviously, Intel's rivals aren't content to leave it alone in this space. Nvidia, in particular, has been working hard on this area, partnering with the likes of Tesla, Audi, Bosch, and others to make its Drive PX the go-to platform for autonomous vehicles.
Lots more links in the article.


JackFlash: Human drivers are still pretty cheap.

Their pay is actually worse than that, and the job sucks, so there's been a shortage of qualified drivers for years now. The industry faces a high turn-over rate, too. So autonomous trucks keep looking like a boon to companies that are trying to stay competitive against each-other and rail, all which keep the actual pay for truckers down. And with a hour of service limit to be enforced by electronic logging devices (ELDs) instead of paper logs that can be "stretched" (i.e. drivers can cook the books) starting on December 18, 2017, which will (make) more truckers stop more frequently, and for longer periods (11 hours of driving in a 14 hour day, with 10 hours off). That limits how far a single driver can go, so tandem drivers get more distance, or truckers can swap out and keep the load moving.

Or it could keep going without a pause with automated trucks, which wouldn't get drowsy or distracted, and wouldn't (ab)use substances while on the job or drive too fast or too slow. Freight movement is all about reliability, and in theory, automation is peak reliability.
posted by filthy light thief at 12:49 PM on September 18, 2017


« Older a comic book about delivering newspapers and also...   |   They probably think you should eat cake, but you... Newer »


This thread has been archived and is closed to new comments