This is where I think the divergence between "machine learning" and "artificial intelligence" really lays. These vehicles need to be able to generalize and understand motivations, not something that can be intuited via "more data," at least not practically.
Who wants to merge into highway traffic for 500 hours before a machine can do it well? Does it take a human 500 hours?
It's also interesting to think how they will deal with uncertainty. In my neighborhood, people don't follow the law at four way stop signs, and a rational and law-abiding agent might just end up being paralyzed by it.
The nice thing about these cars and all the dash cams we have from regular drives is that we can now tweak the laws to reflect reality. The robot cars should do what they are supposed to and drivers or law makers should adjust.
St. Ann $3,609,972 $9,115,012 39.6 13,020
St. John $1,126,763 $3,835,573 29.4 6,517
Ferguson $2,571,190 $12,746,894 20.2 21,203
Bellefontaine Neighbors $749,252 $4,918,310 15.2 10,860
City of Florissant $2,966,669 $23,120,332 12.8 52,158
City of Berkeley $1,047,536 $8,680,716 12.1 8,978
Creve Coeur $1,847,864 $16,365,796 11.3 17,833
Maryland Heights $2,077,689 $22,820,884 9.1 27,472
Jennings $679,787 $7,737,693 8.8 14,712
Hazelwood $1,919,022 $23,809,852 8.1 25,703
Overland $582,292 $8,695,835 6.7 16,062
Richmond Heights $763,523 $12,109,281 6.3 8,603
Black Jack $123,355 $2,621,186 4.7 6,929
Clayton $1,027,932 $21,761,741 4.7 15,939
Olivette $283,289 $7,046,302 4.0 7,737
Bridgeton $445,171 $12,887,494 3.5 11,550
St. Louis City $11,022,000 $441,426,000 2.5 319,294
University City $586,281 $26,639,562 2.2 35,371
St. Louis County $4,397,769 $341,291,336 1.3 998,954
This totals up to around $35MM in Fines and Forfeitures in St. Louis (chosen due to Ferguson and all the articles around Mike Brown). Anyway WAG numbers this looks to be about 80% of the neighborhoods on the highways but it's got the 2 major players (city & county) so lets call it good.
However if I just add all the pops from 21 up I get 146MM people, so that means that if scaled linearly for St. Louis (which may or may not be right), we're talking $1.763MM, so around 1.75BN for the 21 biggest cities in the US.
Easy to say that traffic stops in the US are well over a billion.
Way over analysis but it shows scale of traffic income. If you read the first article linked, it points out the share of revenue this. It can be up to 36% of revenue for small municipalities.
These cars remind me of the oblivious middle aged woman who leaves destruction in her wake but insists she's a good driver because it's never her fault and insurance for woman costs less...
Have they gotten driver-less cars to successfully navigate graded dirt and snow covered roads yet?
Yesterday I was driving past midnight through the city. I'd frequently come to an intersection where 2 arterials crossed at 90 degrees. 3 of the roads had red lights and 2 or more cars backed up. The 4th had a green and no traffic coming at all. And ..... we'd sit there a while, and eventually the light would go yellow, again, for nobody, turn red, pause a second for the non-existent red light runners, then finally turn green for us.
The intersections had cameras mounted pointing all 4 ways, and sensor wires in the roads.
You'd think that if we could build self-driving cars, we could devise a traffic light that would not be monumentally inefficient and incompetent. How hard can it be? And think of all the gas, CO2 and time saved.
There's a heluva lot of low hanging fruit in traffic lights.
> The initial design and construction cost of a roundabout can vary significantly depending on the roundabout size, right-of-way impacts, illumination requirements, and other design or aesthetic features that may be desired. A new single-lane roundabout intersection in an unbuilt environment can have construction costs comparable to a traffic signal. However, as the size of the roundabout increases, particularly in a fully built-out commercial or residential area, the cost of roundabout construction can be higher than that of a traffic signal, depending on the footprint of the roundabout relative to that needed for the signal. However, the ongoing operations and maintenance cost of a roundabout can be less than that for a signal, with the possible exception of increased illumination needs for a roundabout. Although the initial construction cost may be more, a roundabout can have less operating and maintenance costs than a traffic signal, and the service life of a roundabout is significantly longer, approximately 25 years, compared with 10 years for a typical signal (2). Roundabouts also provide substantial cost savings to society due to the reduction in crashes, particularly fatal and injury crashes, over their service life.
It also depends on how much you value people's time waiting for a light to change, as described in the parent scenario.
I guess I didn't phrase my question correctly, how much would it cost to replace all the traffic signals with roundabouts (per your original statement)? I am betting that would be cost prohibitive much less concerns for space.
I didn't say "all", I said "most". Even countries with roundabouts also have traffic signals.
My comment was not trying to establish what you are asking for. Instead, I was pointing out that there are many factors in addition to "cost to replace". There is also the yearly upkeep costs (power to the light, landscaping), signal timing maintenance every few years, and a shorter lifespan for signals vs. roundabout.
Plus, of course, the savings in human time and lives.
Then there's systemic issues which are harder to quantify. Roundabouts have higher throughput than signals, so switching to a roundabout might reduce the need to widen nearby parallel roads, which is an indirect cost savings. And roundabouts without signals work even when the power goes out.
> Roundabouts give superior traffic and safety performance to signalized intersections and to cross intersections almost everywhere, but roundabouts cost more, and not every signalized intersection or cross intersection is in great need of correction. Therefore, a highway agency should start at the top of its high-crash list and its high-congestion list and work down, often converting the problem location into an offset intersection where crossing flows are not too high. ...
> The modern roundabout should be recommended not only as a retrofit where there are present safety and capacity problems, but also as an intersection in new construction where roads join or cross each other for the first time.
That seems eminently reasonable to me.
Recall too that the challenge was to "devise a traffic light that would not be monumentally inefficient and incompetent", so the question is, how much would that hypothetical system cost per intersection, vs. a known, good solution to the given problem?
> much would that hypothetical system cost per intersection, vs. a known, good solution to the given problem?
Since the lights already exist, as well as cameras, and network connections, the incremental cost would be a box with a computer in it. The development cost of the software would be amortized over tens of thousands of intersections.
Yes, we are programmers, and are philosophically inclined to solve things through clever centralization.
That doesn't mean it's the only, or the optimal solution. Roundabouts depend on decentralized control to achieve very similar goal. (Though without, by the way, the pretty nasty failure modes should that box, or the power, go down.)
It's all handwaving without experimental studies and economic analysis. The numbers for roundabout are much better established than the smart systems, which are still experimental in nature.
> Recent roundabout projects in the United States have shown a wide range in reported construction costs. Assuming “1998 U.S. Dollars” in the following examples, costs ranged from $10,000 for a retrofit application of an existing traffic circle to $500,000 for a new roundabout at the junction of two State highways. National Cooperative Highway Research Program (NCHRP) Synthesis 264 (3) reports that the average construction cost of 14 U.S. roundabouts, none being part of an interchange, was approximately $250,000. This amount includes all construction elements, but does not include land acquisition.
Elsewhere I gave links to places which said the installation costs for a roundabout were more expensive than signals. I don't know if the extra 5-10% price of the "box with a computer" changes the balance.
> A modern signal can cost taxpayers between $80,000 and $100,000 to install, depending on the complexity of the intersection and the characteristics of the traffic using it. On top of this, there is the perpetual cost of the electrical power consumed in operating a signalized intersection 24 hours a day. This cost now averages about $1,400 per year.
> A request to spend $450,000 to install a new traffic signal on Easthampton Road remains an option for the City Council.
> And while the price tag might seem high for one traffic light, city officials insist that's the going rate to design and install such an item.
> "They're smarter controllers than they ever were in the past," Department of Public Works Superintendent William D. Fuqua said yesterday.
It's definitely not clear to me that the high-tech, interconnected system you desire is cheaper or more effective than roundabouts. (Again, there are clearly place where signals are more effective. I ...
Since your own figures show it is cheaper, especially since the lights, cameras, and network connections are already installed and paid for, I don't understand what more evidence you need.
As for their efficacy, the anecdote I posted shows there's plenty of low hanging fruit that would substantially improve things, even without an AI.
Let me correct what appears to be a miscommunication.
The numbers I gave don't prove anything, other than show that there is a huge range in cost, depending on the intersection type. You say "the lights, cameras, and network connections are already installed and paid for", but how many intersections are that way?
More specifically, the numbers I showed are that a fully equipped signal, like what you want, costs $450,000, while a roundabout is cheaper. If all lights already have "lights, cameras, and network connections" then it's a sunken cost. If 1% of the signals are that way, then it's entirely different.
I certainly don't know how interconnected the lights are. But even with hard numbers - and you haven't made any claims to any numbers - that's only a part of the point. (FWIW, the best numbers I found are from http://readwrite.com/2014/01/22/ai-traffic-lights-intelligen... which says "According to el-Tantawy, many of today’s traffic lights at intersections operate based on pre-programmed repeated cycles that run with little or no input from fluctuations in traffic.". That's not really helpful, other than to show there are a lot of dumb signals in the US.)
Yes, construction is cheaper on average for signals, but bear in mind that signals with 'lights, cameras, and network connections' are definitely not average signals.
But construction is only part of the cost-benefit analysis. The signals need to be powered, and replaced about every 10 years according to one of the URLs I gave. Roundabouts have a 25 year lifetime. Roundabouts are also safer, so there the benefits to human life must be included. And as I pointed out, you also have to consider the failure modes.
Sure, yes, there's 'low hanging fruit' (which is a misnomer, as the really 'low hanging fruit', like traffic light coordination and in-pavement detectors, have long since been picked). But that's because you have a goal in mind, which is to "improve traffic signals", not "improve traffic systems" (which roundabouts do), or "improve quality of life" (which 'complete streets' and livable community advocates want to do).
Thus, if you think that '"The reductions of 40 percent in vehicle wait time, nearly 26 percent in travel time and 21 percent in projected vehicle emissions realized in this pilot are remarkable,"' is a good justification, then you have to compare it to other solutions which achieve or improve upon those same goals, and not just cherry-pick (again with the fruit metaphors!) the numbers that best support your goal.
As I also pointed out, the pathway to more roundabouts isn't to wholescale replace existing signals, but to use them for new intersections, and for those with high accident rates where it's most worthwhile to switch. The over time, as the old signals need replacing, switch many of them to roundabouts.
That too "would substantially improve things, even without an AI".
Why do you need to conjecture when the data is easily available? The list of every traffic light in Seattle is at https://data.seattle.gov/Transportation/Traffic-Signals/dr6d... . You can unzip it and open WGS84/TrafficSignals.dbf in LibreOffice as a spreadsheet. There are 1077 traffic lights in the city, according to those records.
Of them, 85 are "FULL", 271 are "PRE", and the remaining 721 are "SEMI". Based on http://ops.fhwa.dot.gov/publications/fhwahop08024/chapter5.h... these appear to mean respectively "Fully-Actuated", "Pre-timed", and "Semi-Actuated", so 271 or 25% of your cities traffic lights are on a timer and don't have vehicle detectors.
Your original example was of an intersection with "cameras mounted pointing all 4 ways", making it fully-actuated. There are only 85 of those signals in Seattle, or 8% of the signals.
To be fair to you, you said "on an arterial". However, your definition of "arterial" appears to be different than Seattle's, which subcategorizes them as 'Principal Arterials', 'Minor Arterials', 'Collector Arterials'. See http://www.seattle.gov/transportation/streetclassmaps.htm . More specifically, the map at http://www.seattle.gov/transportation/streetclassmaps/planwe... shows more than 85 intersections of an arterial with an arterial, so the pidgeonhole principal say it shouldn't be hard to find an intersection that isn't fully actuated.
To be fair to me, this is completely besides the point. There are many solutions to your original problem of being stuck at a light. One is to add another layer of technology (smarter signals). Another is to switch to a different type of technology (roundabouts). All you've been doing is cherry picking your observations to justify your initial beliefs. I further believe is because you, like many other programmers, are philosophically inclined towards complex, centralized control systems, because that's what we understand and work with all day.
But the observational evidence is that roundabouts are better than signals for many situations. Please stop ignoring the evidence because your mind is set on the first solution you came up with while your were stuck at that traffic light!
> The City Council did not vote to eliminate the roundabout proposed at 51st Ave S, Renton Ave S and S Roxbury ST from the Transportation Capital Improvement Program (CIP). SDOT has secured a grant for $1.7 million to design and build the roundabouts; unfortunately, based upon preliminary design and cost estimate, additional funding is necessary. If built, it would be the first roundabout in Seattle. The project was identified in the Rainier Beach Neighborhood Plan as the most problematic intersection in SE for all modes. The roundabout could be a very cost effective solution as unlike a traffic signal, it will have minimal on-going operating and maintenance costs. Environmentally, it's better than a traffic signal because vehicles would not be idling at the light. It is also far friendlier to pedestrians, which would have the right away and not have to find the pedestri...
* Significantly more than $1.7m to build a roundabout, vs $40,000 for one that can be installed on existing infrastructure.
* Adding roundabouts requires more land, which is in tight supply in a built-up city like Seattle. I don't think shaving the corners off of buildings is going to go over very well.
* As I've noted, the buildup of the cameras is very recent. The traffic light data you cited is over a year old. They're creeping into the side streets now.
* Nobody needs a study to see the obvious improvements possible in cases I cited.
* Given the desire of the government to install cameras everywhere on the roads, the cost of them is going to be there with or without roundabouts.
* Far friendlier to pedestrians? How is that possible when cars don't stop? I've crossed roundabouts in England as a pedestrian. It involves pushing a button and a traffic light turns red so you can cross. I.e. the traffic lights are still there!
* In Romania the roundabouts also had traffic lights for the pedestrians, as well as pedestrian tunnels under the intersection. That can't be inexpensive.
* I can see a roundabout as cost effective in a rural area where land is cheap and there are few pedestrians. In a builtup city, not so much, as they'd be massive and expensive construction projects. A $40,000 box can be installed instead with no disruption and immediate improvements. Sounds like a winner to me.
"Significantly more than $1.7m to build a roundabout"
No. Look again. The quoted text says "$1.7 million to design and build the roundabouts". Plural. That was three roundabouts, not one, on "51st Ave S, Renton Ave S and S Roxbury ST". And some of the objections you just now raised have been addressed in links I gave earlier, including examples of SDOT proposals for roundabouts in urban Seattle.
I give up. You have decided on your answer and no real-world information will change your personal beliefs. You win.
It's still 10x the cost. Costs of the light system can be further reduced by instead of installing a computer at the light, have the computer centrally located to serve any number of lights. Since the cameras are already network connected, this is practical.
I've been in countries that are comfortable with roundabouts. They also still have traffic lights everywhere. Something is missing in your analysis that explains this choice.
All modern roundabouts have median islands separating incoming and outgoing auto traffic. Pedestrians don't have to find a gap in two directions of traffic, just one. This is safer for pedestrians, especially for younger or older ones, because they only have to concentrate on one direction of traffic at a time. This is what is meant by two-phase. Cross the first half, pause if you need to, then cross the second half. On multi-lane crossings pedestrian beacons or signals are often added if the auto (or pedestrian) traffic is too numerous. The signals can also be two phase, requiring the pedestrian to push a second button when they get to the median. The median can also have a Z path to reorient the pedestrian to view oncoming traffic. Also, the signals usually rest in off, so they are only activated if a pedestrian needs the help crossing. This way only motorists that need to stop are delayed.
Modern roundabouts are the safest form of intersection in the world (much more so than comparable signals). Visit http://tinyurl.com/iihsRAB for modern roundabout FAQs and safety facts. Modern roundabouts, and the pedestrian refuge islands approaching them, are two of nine proven safety measures identified by the FHWA, http://tinyurl.com/7qvsaem
The FHWA has a video about modern roundabouts on Youtube, or check out the IIHS video (iihs dot org).
First cost is the wrong way to compare projects. It would be like buying a car without knowing the fuel economy or safety of the thing, just its price to buy.
Present Value Life Cycle Cost Analysis (LCCA) is the best way to compare two or more choices. When comparing modern roundabouts to signals for a 20-year life cycle (the standard period), modern roundabouts usually cost less. Costs to compare include: first cost (design/land/construction), operation and maintenance (electricity, re-striping, upgrades, etc.), crash reduction (what’s your/your family’s safety worth?), daily delay (what’s your time worth?), daily fuel consumption (spend much on gas?), point source pollution (generated by stopped vehicles = health cost), area insurance rates (this costs more where it is less safe to drive). Each of these things, and others, can be estimated for any two choices and everyone near or using the project area will pay some portion of all of these costs.
Hmm, here in Australia, I can't recall a single time this has happened. I've actually tested how well the software for these lights work, for myself at night time when there's no traffic. Usually the lights on the main roads are green by default. In my case, it'd take 1-2 seconds after I approached the red light of the crossing road for the main one to turn yellow. Same when pressing the pedestrian crossing button.
You've obviously not tested them on a motorcycle in WA...
Grrr...I'm just waiting for the day I get picked off by some overzealous Judge Dredd cop because I ran a red arrow because I got tired of waiting for it to detect me.
Same way they do, you ask for the full tape, which you can get. Or you know just go around the wrong way. Anecdotally red light covered cameras seem to be much better maintained as othewise they lose revenue.
The red light cameras I've seen trigger when someone approaches a red light without slowing down quickly enough. Sometimes, late at night, I make a game of setting them off on purpose, by waiting til the last minute to brake. They wouldn't fire if you stopped, waited, then continued after determining that the detector hadn't noticed you.
Attach a largeish neodynium magnet (like a cow magnet, as a bonus they are coated in plastic so they don't corrode) to your bike frame. It will increase your magnetic signature massively and cause in wire-loop-in-road light controller to "see" you. (note, not all sensors are magnetic, some are weight based or IR based)
The demand sensor wires are too close to the lights, so you wind up having to stop before it turns green. Also, if cars are coming from both directions, the demand system tends to get 180 degrees out of phase with the traffic and makes everyone stop.
A proper system is, well, like one on a self-driving car. Detect traffic from further back using the cameras rather than the sensor wires. Estimate the number of cars, speed, and position, then adjust for max flow.
Given the position/speed data that Google collects through Google play services (which assists then with determining traffic conditions for instance), I think Google could do much better at controlling the traffic lights (or at least provide one source of input for).
In Western Australia the prevailing culture within the Main Roads Department is "If people get used to having green lights they won't stop for a red light so we must try make people get as many red lights as we can" its moronic but true, a technician friend of mine resigned over the stupidity of it.
To give you an idea where the tech is at, the local news did a report for Plymouth MN about Bob who is the person that manually goes around and adjusts the timing on the street lights. It is sneaker net applied to traffic control.
The timing was done in the 70's, and it never worked. The city would argue that its traffic engineers were highly competent and had timed it perfectly, but all the drivers knew better.
It's like the TV sound engineers who'd argue that the commercials are not set to be louder. Yeah, I know, dynamic range compression, but I'd put VU meters on the sound and the loudest sound in the TV show was quite a bit lower than the average sound on the commercials.
The first time I heard of autonomous vehicles I imagined intersections where no cars stop, they simply make minor adjustments to their speed so that vehicles could be continually flowing through in both directions. It would be quite the sight, but would obviously require a prohibition on human drivers and in practice would probably never work out.
That is what I'm talking about. Note that it is a 3 year old article. You'd think environmentalists would get all over "The reductions of 40 percent in vehicle wait time, nearly 26 percent in travel time and 21 percent in projected vehicle emissions realized in this pilot are remarkable," but nope.
http://www.iihs.org/iihs/topics/t/roundabouts/qanda and others cite "Retting, R.A.; Mandavilli, S.; Russell, E.R.; and McCartt, A.T. 2006. Roundabouts, traffic flow and public opinion. Traffic Engineering and Control 47(7):268-72". The above URL summarizes the paper as:
> A study of three locations in New Hampshire, New York and Washington state where roundabouts replaced traffic signals or stop signs found an 89 percent average reduction in vehicle delays and a 56 percent average reduction in vehicle stops.
That URL then goes on to say:
> Because roundabouts improve the efficiency of traffic flow, they also reduce vehicle emissions and fuel consumption. Installing roundabouts in place of traffic signals or stop signs has been found to reduce carbon monoxide emissions by 15-45 percent, nitrous oxide emissions by 21-44 percent, carbon dioxide emissions by 23-37 percent and hydrocarbon emissions by 0-42 percent. 6, 11, 12 Constructing roundabouts in place of traffic signals or stop signs reduced fuel consumption by an estimated 23-34 percent. 6, 11, 13
and
> A 2005 Institute study documented missed opportunities to improve traffic flow and safety at 10 urban intersections suitable for roundabouts where either traffic signals were installed or major modifications were made to 10 intersections with signals. 14 It was estimated that the use of roundabouts instead of traffic signals at these intersections would have reduced vehicle delays by 62-74 percent.
Sounds to me like existing roundabouts at least competitive to state-of-the-art experimental smart traffic signals ... and have the advantage of working even when the power goes out, and don't require more expertise in order to maintain.
I call bullshit on roundabouts improving the efficiency of traffic flow in New Hampshire.
The roundabouts on US 4 between Portsmouth and Concord will back up for miles, and the Portsmouth traffic circle is an enormous cluster, with more accidents than any signaled intersection I've ever seen (Doesn't help that one of the highest-volume liquor stores in New England is located right on the circle...).
> Often there is considerable confusion about roundabouts and traffic circles or rotaries. People often say that they don't work in the Northeast (especially in New York and New Jersey) and are being removed, so why build them here [in Florida]? ...
> A study done by Per Gardner at the University of Maine found that traffic circles have 3.5 to 6.5 times more crashes than roundabouts. ... Because of these growing concerns, the last design reference to traffic circles in the U.S. design standards was in 1965.
> Then came roundabouts. The first roundabout in America was built in 1992 at an intersection in Gainesville. Although traffic circles and roundabouts use a circular design, they operate very differently. Traffic circles are very large and are designed for high-speed vehicle operation. Roundabouts are designed as small as possible, 16 to 180 feet wide, and operate at 15 mph to 25 mph.
This might defeat the purpose, but why not have remote controlled cars. A digital taxi driver could alleviate a lot of the problems that we currently have with non-autonomous cars, except bypasses the whole open-research problem of having roads full with autonomous vehicles. What's being done right now with digital assistants is a similar strategy.
The obvious solution is to forbid human drivers but that will take many years. We will learn cool things until it happens. It is, imho, incidentally the reason why all scifi shows and movies are unrealistic though: there will not be drivers and pilots (of spaceships, planes and cars); it will actually be forbidden.
It won't be forbidden to drive. People will still watch Nascar even if they can't afford to own cars. The demand for drivers can't be satisfied solving for the demand for rides. Robots are going to have to learn to deal with human drivers, we're not going to be able to define them out of existence.
I hate that "dilemma" if whether or not driverless cars should kill its passengers in order to avoid a deadlier collision. If a group of people walk onto the road as my car approaches, why should I be the one to die? They made the dangerous action, they should bear the consequences. Even if they didn't intend to, how is my car supposed to determine that? I think asking "who is at fault" is a better question than "how many people will live" when asking who should die.
Vehicles should be programmed for preservation of the passengers. It's the only unambiguous rubric that makes sense to everyone. Preserve yourself, and if you are in the wrong, bear the legal consequences.
Moreover, why are we asking computers to make decisions we don't ask humans to make? In a split second, a person does not weigh the pros and cons of hitting an old lady vs a schoolbus filled with boy scouts. We just react, and ideally react according to how we were taught in driver's ed. A computer would do just that, only better and more consistently.
It's all silly philosophical argument. The heuristics will be more like: reduce kinetic energy as much as possible, avoid pedestrians as much as possible in the event of likely collision. If that heuristic is sufficient to save lives in most situations, it'll be good enough.
I'm skeptical of the "crash rate double that of those with human drivers"
The figures are "17 minor crashes in 2 million miles." That's a crash every 117,000 miles or roughly once a decade for a typical driver. From my experience most drivers average more than one minor crash per decade. It's probably that they just don't bother formally reporting them. (My most recent one was about 6 months ago. Another car reversed into mine while I was stationary - small dent. Didn't bother with reporting, insurance claims etc.)
My father was very cautious driver, he never broke speed limit intentionally, and I remember the same argument I had with him when I was in driver's school. I was of opinion that speed limits should be hard enforced, and he told me, you don't want that, sometimes you might need to break the speed limit, for example when you're exiting the crossing a little too late for some reason. I think he was right.
But in the case of self-driving cars software - I would much prefer it to just stick to the law, rather than try to match human nuances (or not!) of breaking it. I think general predictability is of value. I hate to say that but the human drivers that cause these accidents are the ones who have to learn better. And if it is still desirable for the autonomous cars to behave differently in certain place, you can always change the speed limits (or other signs) in that location.
The only problem with the article is the speed limit is much more complicated than the number printed on the sign. The actual speed limit depends on the flow of traffic and weather conditions. Merging on to a high way where people are driving 80 then driving 65 is violating the law.
In which jurisdictions is it illegal to obey the speed limit if traffic is going faster? I know some jurisdictions have charges for impeding traffic but that's always been for when you're going under the limit. I'm skeptical that to conform to the law you have to exceed the posted limit.
Toronto, Ontario, Canada at least. One guy got a speeding ticket for doing 120 kph in the merge lane. He got mad and decided he was going to strictly follow the law and do 100 kph in the fast lane, which normally travels at 130 or higher. He caused a traffic jam, got an obstruction ticket, fought it and lost.
There's a big asterisk there, and that was that he got the ticket for being in the fast lane. That's different than forcing someone to speed (i.e. break the limit even in the slow/only lane).
> There are several circumstances in which drivers may be ticketed for illegally blocking or impeding traffic by driving too slowly or failing to yield to a long line of vehicles behind them. ... Driving Too Slowly in Left Lane ... Impeding Traffic ... Failing to Use "Turnouts"
Regarding the first of these,
> Your state's law will say something like:
> Any vehicle proceeding upon a highway at a speed less than the normal speed of traffic moving in the same direction at such time shall be driven in the right-hand lane for traffic or as close as practicable to the right-hand edge or curb, except when overtaking and passing another vehicle pro ceeding in the same direction or when preparing for a left turn at an intersection or into a private road or driveway. If a vehicle is being driven at a speed less than the normal speed of traffic mov ing in the same direction at such time, and is not being driven in the right-hand lane for traffic or as close as practicable to the right-hand edge or curb, it shall constitute evidence that the driver is operating the vehicle in violation of this section.
Note how this uses "normal speed", not "speed limit".
In an "absolute" speed limit state, you can defend against a 'Driving Too Slowly in Left Lane' by showing you were were traveling at the posted speed limit. (Quoting NoLo - I have no legal training.)
Again, that is not saying you have to exceed the limit, it's just saying if you're slower than the flow of traffic, then you have to move over and let people pass.
My interpretation is that in a state has an "absolute" speed limit law then you have a successful defense against a ticket for going at the speed limit in the left-most lane when there is faster traffic.
If you are in a state which does not have an absolute speed limit law, then you do not have that defense, and can be fined for going at the speed limit in that situation.
Here is a breakdown of the state laws (undated, and no promises that it's correct or still true) - http://www.mit.edu/~jfc/right.html . The bottom has the following:
> The Uniform Vehicle Code states:
> Upon all roadways any vehicle proceeding at less than the normal speed of traffic at the time and place and under the conditions then existing shall be driven in the right-hand lane then available for traffic ...
> This law refers to the "normal" speed of traffic, not the "legal" speed of traffic. The 60 MPH driver in a 55 MPH zone where everybody else is going 65 MPH must move right. Contrast Alaska's rule, 13 AAC 002.50, allowing vehicles driving at the speed limit to use the left lane, and Colorado rev. stat. 42-4-1103, prohibiting blocking the "normal and reasonable" movement of traffic.
> Enforcement is inconsistent. Toledo police used to ticket truck drivers for driving at the 60 MPH speed limit in the left lane. Police looking for criminal activity frequently use the "keep right" law as a pretext to stop a suspicious car. On the other hand, a New York judge announced that he would not convict drivers for blocking speeding traffic, People v. Ilieveski, 175 Misc. 2d 943; 670 N.Y.S.2d 1004 (Monroe County N.Y. 1998).
You're confusing two different moving violations here: exceeding posted speed limit and failure to keep right. (AKA impeding traffic; the exact term varies from state to state. Also, I'm assuming American traffic law here.)
If the speed limit is 65, the flow of traffic is at 80, and you drive 65 in the left lane - you are failing to keep as far right as practicable, because you're moving slower than the flow of traffic in other than the rightmost lane.
If you're going 80, you are still speeding, but you're not failing to keep right. Yes, it's unlikely you'll be pulled over, but you are committing a moving violation nonetheless.
If you're going 65 and you're in the rightmost lane, there is no violation even though you're travelling slower than the flow of traffic. Posted speed limits are still do-not-exceed values; the requirement to keep as far right as practicable is a completely separate issue.
> Ten days later, a Mountain View motorcycle cop noticed traffic stacking up behind a Google car going 24 miles an hour in a busy 35 mph zone. He zoomed over and became the first officer to stop a robot car. He didn’t issue a ticket -- who would he give it to? -- but he warned the two engineers on board about creating a hazard.
> “The right thing would have been for this car to pull over, let the traffic go and then pull back on the roadway,” said Sergeant Saul Jaeger, head of the police department’s traffic-enforcement unit. “I like it when people err on the side of caution. But can something be too cautious? Yeah.”
I believe that this reflects the 25 mph maximum speed that Google has set. I don't know CA law, but in some places it's illegal to go under 25 mph in a 35 mph zone. It's always safest to minimize speed difference with other vehicles. Even if that means exceeding the limit. That won't protect against citations in speed traps, however.
By California law, NEVs are vehicles which are incapable of driving faster than 25 mph, and are allowed to drive (unless barred by local ordinance) on any street with posted limit ≤ 35 mph.
Thanks for the cite. According to Wikipedia, 25 mph NEVs can use roads posted at 45 mph in some places. But laws notwithstanding, even 10-15 mph speed differential is hazardous.
Yep. It would be great to get urban street speed limits lowered to max out at 25–30 mph everywhere, with strict enforcement against driving above 35–40mph. In collisions between cars and pedestrians or cyclists, everyone mostly walks away alive at 25 mph, whereas at 45+ mph the pedestrian or cyclist dies almost every time.
With smarter road design organized to minimize waits at traffic signals, it wouldn’t even necessarily result in slower trips or less road capacity overall.
> The Caddy performed perfectly, except when it had to merge onto I-395 South and swing across three lanes of traffic in 150 yards (137 meters) to head toward the Pentagon.
This is probably one of the most stressful maneuvers you can make on the Capital beltway. Honestly I feel like once we hit critical mass with driverless cars this action will be much more manageable. I and other drivers have to do some real aggressive things to make this work.
Possible reasons you might find more slow-speed minor rear-end collisions among self-driving cars than among human-driven cars on a per-mile basis:
1. There are only a tiny number of self-driving cars on the road, driving a relatively small number of miles, and accidents in general are fairly rare on a per-mile basis, so there might not yet be enough data to get a reliable idea of average crash rate. The linked study is based on analysis of only 11 (!!) crashes, 8 of which were at <5 mph, with only 2 causing “possible injuries”. As the study itself says, “[...] the corresponding 95% confidence intervals overlap. Therefore, we currently cannot rule out, with a reasonable level of confidence, the possibility that the actual rate for self-driving vehicles is lower than for conventional vehicles.” (Personally I think their confidence intervals for self-driving cars are useless given the data so far, but whatever.)
2. The Google cars which account for all of the cited crashes are AFAIK doing all of their driving on city streets, where intersections, stops, and thus these types of collisions are more common than on highways.
3. I suspect most tiny bumper bumps which don’t damage either car or injure anyone are never reported to the police when both drivers are people, but self-driving cars record every case.
4. Self-driving cars have better sensors to detect road hazards, and as a result might stop in cases where a human driver would do something which risked a more serious accident. People tail-gating these cars will end up unable to react in time and bumping them, but even these 11 crashes might have been in situations where the autonomous car stopping prevented more severe car damage or injuries.
5. Driving cautiously and strictly limiting speed to 25mph might make following human drivers frustrated and more likely to tail closely. This might be exacerbated by the Google cars’ goofy appearance. (Would be interesting to see statistics for various skills/styles of human drivers, or for human-driven NEVs.) I’m not sure NEV accidents can be meaningfully compared to standard car accidents.
Overall I think this topic is worthy of study, but the data is not really all that useful yet. Keith Naughton’s Bloomberg piece calling this a “key flaw” seems like high-order FUD / clickbait.
Interesting. Coincidentally (or not?) yesterday Bloomberg posted an article about George Hotz building his own Self Driving AI and he talks about the very problem Google is having with the car following the rules too strictly. Looks like he's right at least about this one thing.
“They’re a little bit like a cautious student driver or a
grandma.” --- Dmitri Dolgov, principal engineer of the program.
No one likes getting stuck driving behind grandma. I expect as the tech sees more of the light of day, they will go from 'cute curiosity' to 'extremely frustrating to drive near'. They will pin you, in some cases, well below the speed limit, which is often set dangerously low.
When someone dies trying to pass one of these things, of course it won't be the robots fault...
Robots should drive to minimize their impact on the road. That would be my guiding philosophy. That means driving appropriate speeds at appropriate times, and not creating a dangerous situation for other drivers. "Slower traffic ahead" is a big danger on the freeway.
A human driver would be expected to speed up and merge with traffic and not create an impedance. There is no question the right thing for the software to do is to exceed the posted limit.
Imagine a government so willing to help spur investment in driver-less cars they set a different statutory limit for the AI to follow (class-by-class)? Sounds like a very cool way to solve two problems at once. Based on the specifics of the road, visibility, and telemetry, weather, etc. the tech gets a huge boost if the robot can travel significantly faster.
85 comments
[ 5.2 ms ] story [ 154 ms ] threadWho wants to merge into highway traffic for 500 hours before a machine can do it well? Does it take a human 500 hours?
A tough problem to solve, no doubt.
http://www.governing.com/topics/public-justice-safety/gov-fe...
This is all fines and forfeitures but:
St. Ann $3,609,972 $9,115,012 39.6 13,020 St. John $1,126,763 $3,835,573 29.4 6,517 Ferguson $2,571,190 $12,746,894 20.2 21,203 Bellefontaine Neighbors $749,252 $4,918,310 15.2 10,860 City of Florissant $2,966,669 $23,120,332 12.8 52,158 City of Berkeley $1,047,536 $8,680,716 12.1 8,978 Creve Coeur $1,847,864 $16,365,796 11.3 17,833 Maryland Heights $2,077,689 $22,820,884 9.1 27,472 Jennings $679,787 $7,737,693 8.8 14,712 Hazelwood $1,919,022 $23,809,852 8.1 25,703 Overland $582,292 $8,695,835 6.7 16,062 Richmond Heights $763,523 $12,109,281 6.3 8,603 Black Jack $123,355 $2,621,186 4.7 6,929 Clayton $1,027,932 $21,761,741 4.7 15,939 Olivette $283,289 $7,046,302 4.0 7,737 Bridgeton $445,171 $12,887,494 3.5 11,550 St. Louis City $11,022,000 $441,426,000 2.5 319,294 University City $586,281 $26,639,562 2.2 35,371 St. Louis County $4,397,769 $341,291,336 1.3 998,954
This totals up to around $35MM in Fines and Forfeitures in St. Louis (chosen due to Ferguson and all the articles around Mike Brown). Anyway WAG numbers this looks to be about 80% of the neighborhoods on the highways but it's got the 2 major players (city & county) so lets call it good.
St. Louis is 21st (https://en.wikipedia.org/wiki/List_of_metropolitan_areas_of_...) in Metro area size, so that gives us $755MM if everything was actually the size of St. Louis.
However if I just add all the pops from 21 up I get 146MM people, so that means that if scaled linearly for St. Louis (which may or may not be right), we're talking $1.763MM, so around 1.75BN for the 21 biggest cities in the US.
Easy to say that traffic stops in the US are well over a billion.
Way over analysis but it shows scale of traffic income. If you read the first article linked, it points out the share of revenue this. It can be up to 36% of revenue for small municipalities.
Have they gotten driver-less cars to successfully navigate graded dirt and snow covered roads yet?
The intersections had cameras mounted pointing all 4 ways, and sensor wires in the roads.
You'd think that if we could build self-driving cars, we could devise a traffic light that would not be monumentally inefficient and incompetent. How hard can it be? And think of all the gas, CO2 and time saved.
There's a heluva lot of low hanging fruit in traffic lights.
> The initial design and construction cost of a roundabout can vary significantly depending on the roundabout size, right-of-way impacts, illumination requirements, and other design or aesthetic features that may be desired. A new single-lane roundabout intersection in an unbuilt environment can have construction costs comparable to a traffic signal. However, as the size of the roundabout increases, particularly in a fully built-out commercial or residential area, the cost of roundabout construction can be higher than that of a traffic signal, depending on the footprint of the roundabout relative to that needed for the signal. However, the ongoing operations and maintenance cost of a roundabout can be less than that for a signal, with the possible exception of increased illumination needs for a roundabout. Although the initial construction cost may be more, a roundabout can have less operating and maintenance costs than a traffic signal, and the service life of a roundabout is significantly longer, approximately 25 years, compared with 10 years for a typical signal (2). Roundabouts also provide substantial cost savings to society due to the reduction in crashes, particularly fatal and injury crashes, over their service life.
It also depends on how much you value people's time waiting for a light to change, as described in the parent scenario.
My comment was not trying to establish what you are asking for. Instead, I was pointing out that there are many factors in addition to "cost to replace". There is also the yearly upkeep costs (power to the light, landscaping), signal timing maintenance every few years, and a shorter lifespan for signals vs. roundabout.
Plus, of course, the savings in human time and lives.
Then there's systemic issues which are harder to quantify. Roundabouts have higher throughput than signals, so switching to a roundabout might reduce the need to widen nearby parallel roads, which is an indirect cost savings. And roundabouts without signals work even when the power goes out.
Quoting from http://www.roundaboutresources.org/when-should-roundabouts-r... :
> Roundabouts give superior traffic and safety performance to signalized intersections and to cross intersections almost everywhere, but roundabouts cost more, and not every signalized intersection or cross intersection is in great need of correction. Therefore, a highway agency should start at the top of its high-crash list and its high-congestion list and work down, often converting the problem location into an offset intersection where crossing flows are not too high. ...
> The modern roundabout should be recommended not only as a retrofit where there are present safety and capacity problems, but also as an intersection in new construction where roads join or cross each other for the first time.
That seems eminently reasonable to me.
Recall too that the challenge was to "devise a traffic light that would not be monumentally inefficient and incompetent", so the question is, how much would that hypothetical system cost per intersection, vs. a known, good solution to the given problem?
Since the lights already exist, as well as cameras, and network connections, the incremental cost would be a box with a computer in it. The development cost of the software would be amortized over tens of thousands of intersections.
That doesn't mean it's the only, or the optimal solution. Roundabouts depend on decentralized control to achieve very similar goal. (Though without, by the way, the pretty nasty failure modes should that box, or the power, go down.)
It's all handwaving without experimental studies and economic analysis. The numbers for roundabout are much better established than the smart systems, which are still experimental in nature.
For that matter, how much is that cost? The proposal at http://readwrite.com/2014/01/22/ai-traffic-lights-intelligen... is "$20,000 to $40,000 per intersection" for 'a box with a computer in it'.
From http://safety.fhwa.dot.gov/intersection/innovative/roundabou... , "The average construction cost of roundabouts is estimated at approximately $250,000". That gets its numbers from http://www.fhwa.dot.gov/publications/research/safety/00067/0... which says:
> Recent roundabout projects in the United States have shown a wide range in reported construction costs. Assuming “1998 U.S. Dollars” in the following examples, costs ranged from $10,000 for a retrofit application of an existing traffic circle to $500,000 for a new roundabout at the junction of two State highways. National Cooperative Highway Research Program (NCHRP) Synthesis 264 (3) reports that the average construction cost of 14 U.S. roundabouts, none being part of an interchange, was approximately $250,000. This amount includes all construction elements, but does not include land acquisition.
Elsewhere I gave links to places which said the installation costs for a roundabout were more expensive than signals. I don't know if the extra 5-10% price of the "box with a computer" changes the balance.
I tried to get numbers for the cost of a intersection. from http://azdot.gov/business/engineering-and-construction/traff...:
> A modern signal can cost taxpayers between $80,000 and $100,000 to install, depending on the complexity of the intersection and the characteristics of the traffic using it. On top of this, there is the perpetual cost of the electrical power consumed in operating a signalized intersection 24 hours a day. This cost now averages about $1,400 per year.
However, that appears to be for the dumb light, and not one with 'cameras and network connections' or even magnetic sensors. From http://www.masslive.com/news/index.ssf/2009/02/450000_traffi... :
> A request to spend $450,000 to install a new traffic signal on Easthampton Road remains an option for the City Council.
> And while the price tag might seem high for one traffic light, city officials insist that's the going rate to design and install such an item.
> "They're smarter controllers than they ever were in the past," Department of Public Works Superintendent William D. Fuqua said yesterday.
It's definitely not clear to me that the high-tech, interconnected system you desire is cheaper or more effective than roundabouts. (Again, there are clearly place where signals are more effective. I ...
As for their efficacy, the anecdote I posted shows there's plenty of low hanging fruit that would substantially improve things, even without an AI.
The numbers I gave don't prove anything, other than show that there is a huge range in cost, depending on the intersection type. You say "the lights, cameras, and network connections are already installed and paid for", but how many intersections are that way?
More specifically, the numbers I showed are that a fully equipped signal, like what you want, costs $450,000, while a roundabout is cheaper. If all lights already have "lights, cameras, and network connections" then it's a sunken cost. If 1% of the signals are that way, then it's entirely different.
I certainly don't know how interconnected the lights are. But even with hard numbers - and you haven't made any claims to any numbers - that's only a part of the point. (FWIW, the best numbers I found are from http://readwrite.com/2014/01/22/ai-traffic-lights-intelligen... which says "According to el-Tantawy, many of today’s traffic lights at intersections operate based on pre-programmed repeated cycles that run with little or no input from fluctuations in traffic.". That's not really helpful, other than to show there are a lot of dumb signals in the US.)
Yes, construction is cheaper on average for signals, but bear in mind that signals with 'lights, cameras, and network connections' are definitely not average signals.
But construction is only part of the cost-benefit analysis. The signals need to be powered, and replaced about every 10 years according to one of the URLs I gave. Roundabouts have a 25 year lifetime. Roundabouts are also safer, so there the benefits to human life must be included. And as I pointed out, you also have to consider the failure modes.
Sure, yes, there's 'low hanging fruit' (which is a misnomer, as the really 'low hanging fruit', like traffic light coordination and in-pavement detectors, have long since been picked). But that's because you have a goal in mind, which is to "improve traffic signals", not "improve traffic systems" (which roundabouts do), or "improve quality of life" (which 'complete streets' and livable community advocates want to do).
Thus, if you think that '"The reductions of 40 percent in vehicle wait time, nearly 26 percent in travel time and 21 percent in projected vehicle emissions realized in this pilot are remarkable,"' is a good justification, then you have to compare it to other solutions which achieve or improve upon those same goals, and not just cherry-pick (again with the fruit metaphors!) the numbers that best support your goal.
As I also pointed out, the pathway to more roundabouts isn't to wholescale replace existing signals, but to use them for new intersections, and for those with high accident rates where it's most worthwhile to switch. The over time, as the old signals need replacing, switch many of them to roundabouts.
That too "would substantially improve things, even without an AI".
In the Seattle metro area, it's hard to find one on an arterial that doesn't have them. (This is just in the last handful of years.)
Of them, 85 are "FULL", 271 are "PRE", and the remaining 721 are "SEMI". Based on http://ops.fhwa.dot.gov/publications/fhwahop08024/chapter5.h... these appear to mean respectively "Fully-Actuated", "Pre-timed", and "Semi-Actuated", so 271 or 25% of your cities traffic lights are on a timer and don't have vehicle detectors.
Your original example was of an intersection with "cameras mounted pointing all 4 ways", making it fully-actuated. There are only 85 of those signals in Seattle, or 8% of the signals.
To be fair to you, you said "on an arterial". However, your definition of "arterial" appears to be different than Seattle's, which subcategorizes them as 'Principal Arterials', 'Minor Arterials', 'Collector Arterials'. See http://www.seattle.gov/transportation/streetclassmaps.htm . More specifically, the map at http://www.seattle.gov/transportation/streetclassmaps/planwe... shows more than 85 intersections of an arterial with an arterial, so the pidgeonhole principal say it shouldn't be hard to find an intersection that isn't fully actuated.
To be fair to me, this is completely besides the point. There are many solutions to your original problem of being stuck at a light. One is to add another layer of technology (smarter signals). Another is to switch to a different type of technology (roundabouts). All you've been doing is cherry picking your observations to justify your initial beliefs. I further believe is because you, like many other programmers, are philosophically inclined towards complex, centralized control systems, because that's what we understand and work with all day.
But the observational evidence is that roundabouts are better than signals for many situations. Please stop ignoring the evidence because your mind is set on the first solution you came up with while your were stuck at that traffic light!
For examples drawn from Seattle, from these meeting notes in 2006 at http://www.seattle.gov/transportation/docs/setscctnov06%20mt... :
> The City Council did not vote to eliminate the roundabout proposed at 51st Ave S, Renton Ave S and S Roxbury ST from the Transportation Capital Improvement Program (CIP). SDOT has secured a grant for $1.7 million to design and build the roundabouts; unfortunately, based upon preliminary design and cost estimate, additional funding is necessary. If built, it would be the first roundabout in Seattle. The project was identified in the Rainier Beach Neighborhood Plan as the most problematic intersection in SE for all modes. The roundabout could be a very cost effective solution as unlike a traffic signal, it will have minimal on-going operating and maintenance costs. Environmentally, it's better than a traffic signal because vehicles would not be idling at the light. It is also far friendlier to pedestrians, which would have the right away and not have to find the pedestri...
* Adding roundabouts requires more land, which is in tight supply in a built-up city like Seattle. I don't think shaving the corners off of buildings is going to go over very well.
* As I've noted, the buildup of the cameras is very recent. The traffic light data you cited is over a year old. They're creeping into the side streets now.
* Nobody needs a study to see the obvious improvements possible in cases I cited.
* Given the desire of the government to install cameras everywhere on the roads, the cost of them is going to be there with or without roundabouts.
* Far friendlier to pedestrians? How is that possible when cars don't stop? I've crossed roundabouts in England as a pedestrian. It involves pushing a button and a traffic light turns red so you can cross. I.e. the traffic lights are still there!
* In Romania the roundabouts also had traffic lights for the pedestrians, as well as pedestrian tunnels under the intersection. That can't be inexpensive.
* I can see a roundabout as cost effective in a rural area where land is cheap and there are few pedestrians. In a builtup city, not so much, as they'd be massive and expensive construction projects. A $40,000 box can be installed instead with no disruption and immediate improvements. Sounds like a winner to me.
No. Look again. The quoted text says "$1.7 million to design and build the roundabouts". Plural. That was three roundabouts, not one, on "51st Ave S, Renton Ave S and S Roxbury ST". And some of the objections you just now raised have been addressed in links I gave earlier, including examples of SDOT proposals for roundabouts in urban Seattle.
I give up. You have decided on your answer and no real-world information will change your personal beliefs. You win.
I've been in countries that are comfortable with roundabouts. They also still have traffic lights everywhere. Something is missing in your analysis that explains this choice.
First cost is the wrong way to compare projects. It would be like buying a car without knowing the fuel economy or safety of the thing, just its price to buy. Present Value Life Cycle Cost Analysis (LCCA) is the best way to compare two or more choices. When comparing modern roundabouts to signals for a 20-year life cycle (the standard period), modern roundabouts usually cost less. Costs to compare include: first cost (design/land/construction), operation and maintenance (electricity, re-striping, upgrades, etc.), crash reduction (what’s your/your family’s safety worth?), daily delay (what’s your time worth?), daily fuel consumption (spend much on gas?), point source pollution (generated by stopped vehicles = health cost), area insurance rates (this costs more where it is less safe to drive). Each of these things, and others, can be estimated for any two choices and everyone near or using the project area will pay some portion of all of these costs.
Grrr...I'm just waiting for the day I get picked off by some overzealous Judge Dredd cop because I ran a red arrow because I got tired of waiting for it to detect me.
Also this is why you have a dash cam.
A proper system is, well, like one on a self-driving car. Detect traffic from further back using the cameras rather than the sensor wires. Estimate the number of cars, speed, and position, then adjust for max flow.
It's like the TV sound engineers who'd argue that the commercials are not set to be louder. Yeah, I know, dynamic range compression, but I'd put VU meters on the sound and the loudest sound in the TV show was quite a bit lower than the average sound on the commercials.
https://vimeo.com/106226560
http://www.iihs.org/iihs/topics/t/roundabouts/qanda and others cite "Retting, R.A.; Mandavilli, S.; Russell, E.R.; and McCartt, A.T. 2006. Roundabouts, traffic flow and public opinion. Traffic Engineering and Control 47(7):268-72". The above URL summarizes the paper as:
> A study of three locations in New Hampshire, New York and Washington state where roundabouts replaced traffic signals or stop signs found an 89 percent average reduction in vehicle delays and a 56 percent average reduction in vehicle stops.
That URL then goes on to say:
> Because roundabouts improve the efficiency of traffic flow, they also reduce vehicle emissions and fuel consumption. Installing roundabouts in place of traffic signals or stop signs has been found to reduce carbon monoxide emissions by 15-45 percent, nitrous oxide emissions by 21-44 percent, carbon dioxide emissions by 23-37 percent and hydrocarbon emissions by 0-42 percent. 6, 11, 12 Constructing roundabouts in place of traffic signals or stop signs reduced fuel consumption by an estimated 23-34 percent. 6, 11, 13
and
> A 2005 Institute study documented missed opportunities to improve traffic flow and safety at 10 urban intersections suitable for roundabouts where either traffic signals were installed or major modifications were made to 10 intersections with signals. 14 It was estimated that the use of roundabouts instead of traffic signals at these intersections would have reduced vehicle delays by 62-74 percent.
Sounds to me like existing roundabouts at least competitive to state-of-the-art experimental smart traffic signals ... and have the advantage of working even when the power goes out, and don't require more expertise in order to maintain.
The roundabouts on US 4 between Portsmouth and Concord will back up for miles, and the Portsmouth traffic circle is an enormous cluster, with more accidents than any signaled intersection I've ever seen (Doesn't help that one of the highest-volume liquor stores in New England is located right on the circle...).
Traffic circles are not roundabouts (nor are they 'traffic calming circles'). Eg, http://www.heraldtribune.com/article/20090621/columnist/9062... (warning: that's the "print" page, rather than having three differet pages):
> Often there is considerable confusion about roundabouts and traffic circles or rotaries. People often say that they don't work in the Northeast (especially in New York and New Jersey) and are being removed, so why build them here [in Florida]? ...
> A study done by Per Gardner at the University of Maine found that traffic circles have 3.5 to 6.5 times more crashes than roundabouts. ... Because of these growing concerns, the last design reference to traffic circles in the U.S. design standards was in 1965.
> Then came roundabouts. The first roundabout in America was built in 1992 at an intersection in Gainesville. Although traffic circles and roundabouts use a circular design, they operate very differently. Traffic circles are very large and are designed for high-speed vehicle operation. Roundabouts are designed as small as possible, 16 to 180 feet wide, and operate at 15 mph to 25 mph.
See also http://www.wsdot.wa.gov/Safety/roundabouts/BasicFacts.htm and http://www.wcroads.org/sites/default/files/pdf/media/Modern%... .
The figures are "17 minor crashes in 2 million miles." That's a crash every 117,000 miles or roughly once a decade for a typical driver. From my experience most drivers average more than one minor crash per decade. It's probably that they just don't bother formally reporting them. (My most recent one was about 6 months ago. Another car reversed into mine while I was stationary - small dent. Didn't bother with reporting, insurance claims etc.)
But in the case of self-driving cars software - I would much prefer it to just stick to the law, rather than try to match human nuances (or not!) of breaking it. I think general predictability is of value. I hate to say that but the human drivers that cause these accidents are the ones who have to learn better. And if it is still desirable for the autonomous cars to behave differently in certain place, you can always change the speed limits (or other signs) in that location.
> There are several circumstances in which drivers may be ticketed for illegally blocking or impeding traffic by driving too slowly or failing to yield to a long line of vehicles behind them. ... Driving Too Slowly in Left Lane ... Impeding Traffic ... Failing to Use "Turnouts"
Regarding the first of these,
> Your state's law will say something like:
> Any vehicle proceeding upon a highway at a speed less than the normal speed of traffic moving in the same direction at such time shall be driven in the right-hand lane for traffic or as close as practicable to the right-hand edge or curb, except when overtaking and passing another vehicle pro ceeding in the same direction or when preparing for a left turn at an intersection or into a private road or driveway. If a vehicle is being driven at a speed less than the normal speed of traffic mov ing in the same direction at such time, and is not being driven in the right-hand lane for traffic or as close as practicable to the right-hand edge or curb, it shall constitute evidence that the driver is operating the vehicle in violation of this section.
Note how this uses "normal speed", not "speed limit".
The NoLo site at http://www.nolo.com/legal-encyclopedia/free-books/beat-ticke... describes how there are three types of speed-limit violations: "Absolute", "Presumed", and "Basic".
In an "absolute" speed limit state, you can defend against a 'Driving Too Slowly in Left Lane' by showing you were were traveling at the posted speed limit. (Quoting NoLo - I have no legal training.)
However, that is not true of all states.
If you are in a state which does not have an absolute speed limit law, then you do not have that defense, and can be fined for going at the speed limit in that situation.
Here is a breakdown of the state laws (undated, and no promises that it's correct or still true) - http://www.mit.edu/~jfc/right.html . The bottom has the following:
> The Uniform Vehicle Code states:
> Upon all roadways any vehicle proceeding at less than the normal speed of traffic at the time and place and under the conditions then existing shall be driven in the right-hand lane then available for traffic ...
> This law refers to the "normal" speed of traffic, not the "legal" speed of traffic. The 60 MPH driver in a 55 MPH zone where everybody else is going 65 MPH must move right. Contrast Alaska's rule, 13 AAC 002.50, allowing vehicles driving at the speed limit to use the left lane, and Colorado rev. stat. 42-4-1103, prohibiting blocking the "normal and reasonable" movement of traffic.
> Enforcement is inconsistent. Toledo police used to ticket truck drivers for driving at the 60 MPH speed limit in the left lane. Police looking for criminal activity frequently use the "keep right" law as a pretext to stop a suspicious car. On the other hand, a New York judge announced that he would not convict drivers for blocking speeding traffic, People v. Ilieveski, 175 Misc. 2d 943; 670 N.Y.S.2d 1004 (Monroe County N.Y. 1998).
If the speed limit is 65, the flow of traffic is at 80, and you drive 65 in the left lane - you are failing to keep as far right as practicable, because you're moving slower than the flow of traffic in other than the rightmost lane.
If you're going 80, you are still speeding, but you're not failing to keep right. Yes, it's unlikely you'll be pulled over, but you are committing a moving violation nonetheless.
If you're going 65 and you're in the rightmost lane, there is no violation even though you're travelling slower than the flow of traffic. Posted speed limits are still do-not-exceed values; the requirement to keep as far right as practicable is a completely separate issue.
> “The right thing would have been for this car to pull over, let the traffic go and then pull back on the roadway,” said Sergeant Saul Jaeger, head of the police department’s traffic-enforcement unit. “I like it when people err on the side of caution. But can something be too cautious? Yeah.”
I believe that this reflects the 25 mph maximum speed that Google has set. I don't know CA law, but in some places it's illegal to go under 25 mph in a 35 mph zone. It's always safest to minimize speed difference with other vehicles. Even if that means exceeding the limit. That won't protect against citations in speed traps, however.
https://www.dmv.ca.gov/portal/dmv/?1dmy&urile=wcm:path:/dmv_...
This traffic stop was caused more by cop ignorance than any problem of Google’s. No harm done though.
Human drivers of NEVs go the same top speed on the same roads, it just doesn’t make the news if one gets stopped and has a conversation with a cop.
With smarter road design organized to minimize waits at traffic signals, it wouldn’t even necessarily result in slower trips or less road capacity overall.
This is probably one of the most stressful maneuvers you can make on the Capital beltway. Honestly I feel like once we hit critical mass with driverless cars this action will be much more manageable. I and other drivers have to do some real aggressive things to make this work.
1. There are only a tiny number of self-driving cars on the road, driving a relatively small number of miles, and accidents in general are fairly rare on a per-mile basis, so there might not yet be enough data to get a reliable idea of average crash rate. The linked study is based on analysis of only 11 (!!) crashes, 8 of which were at <5 mph, with only 2 causing “possible injuries”. As the study itself says, “[...] the corresponding 95% confidence intervals overlap. Therefore, we currently cannot rule out, with a reasonable level of confidence, the possibility that the actual rate for self-driving vehicles is lower than for conventional vehicles.” (Personally I think their confidence intervals for self-driving cars are useless given the data so far, but whatever.)
2. The Google cars which account for all of the cited crashes are AFAIK doing all of their driving on city streets, where intersections, stops, and thus these types of collisions are more common than on highways.
3. I suspect most tiny bumper bumps which don’t damage either car or injure anyone are never reported to the police when both drivers are people, but self-driving cars record every case.
4. Self-driving cars have better sensors to detect road hazards, and as a result might stop in cases where a human driver would do something which risked a more serious accident. People tail-gating these cars will end up unable to react in time and bumping them, but even these 11 crashes might have been in situations where the autonomous car stopping prevented more severe car damage or injuries.
5. Driving cautiously and strictly limiting speed to 25mph might make following human drivers frustrated and more likely to tail closely. This might be exacerbated by the Google cars’ goofy appearance. (Would be interesting to see statistics for various skills/styles of human drivers, or for human-driven NEVs.) I’m not sure NEV accidents can be meaningfully compared to standard car accidents.
Overall I think this topic is worthy of study, but the data is not really all that useful yet. Keith Naughton’s Bloomberg piece calling this a “key flaw” seems like high-order FUD / clickbait.
Article here for the curious: http://www.bloomberg.com/features/2015-george-hotz-self-driv...
When someone dies trying to pass one of these things, of course it won't be the robots fault...
Robots should drive to minimize their impact on the road. That would be my guiding philosophy. That means driving appropriate speeds at appropriate times, and not creating a dangerous situation for other drivers. "Slower traffic ahead" is a big danger on the freeway.
A human driver would be expected to speed up and merge with traffic and not create an impedance. There is no question the right thing for the software to do is to exceed the posted limit.
Imagine a government so willing to help spur investment in driver-less cars they set a different statutory limit for the AI to follow (class-by-class)? Sounds like a very cool way to solve two problems at once. Based on the specifics of the road, visibility, and telemetry, weather, etc. the tech gets a huge boost if the robot can travel significantly faster.