Vehicle automation is increasingly showing up on the radar of urban planning and transportation planning professionals. Technologies are developing rapidly, and some news stories report that fully self-driving cars are just a few years away. It’s tempting to envision automation ushering in a bold new era in urban transportation, where driverless cars whisk passengers between destinations safely and conveniently, use roads with great efficiency, and make public transit as we know it obsolete.
However, a closer look at vehicle automation reveals a more nuanced picture of the future. Automation capable of replacing human drivers in any situation may be many years away from the market. The traffic flow improvements enabled by automation will be limited in several ways. Buses and other forms of public transit will still be needed to efficiently move large numbers of travelers around cities. And various forms of automation in buses could enable major improvements in service.
The last two points have come up on this blog before (here, here and here), but since there are a variety of opinions on the implications of automation for transit, it’s useful to dig a bit deeper into these issues and take a critical look at when various forms of automation will arrive, how automation will affect traffic flow, and how it will affect travel behaviour. This post will delve into those questions to shed a bit more light on what automation means for the future of public transit.
According to some, vehicles that can drive themselves anywhere, anytime, without any human intervention – described as “Level 4” vehicles by the National Highway Traffic Safety Administration (NHTSA) – are just around the corner. In 2012, Google co-founder Sergey Brin said of their famous self-driving car: “you can count on one hand the number of years until ordinary people can experience this.” Many others have made bullish predictions. For example, the market research firm ABI Research foresees Level 4 cars on the roads by around 2020, and panelists at the Society of Automotive Engineers (SAE) 2013 World Congress predicted arrival between 2020 and 2025.
On the other hand, some point to a number of challenges that suggest Level 4 will emerge further down the road, perhaps not for several decades. Steven Shladover of the California Partners for Advanced Transportation Technology, a leading expert on vehicle automation, argues that Level 4 will be much more technically difficult to achieve than many optimists acknowledge (see Vol. 7, No. 3 here). According to Shladover, huge advances in technology would be needed to progress to systems capable of driving safely in the vast range of complex and unpredictable situations that arise on roads. In addition, such systems would have to be far more reliable than products like laptops or mobile phones, and extensive – and expensive – testing will be needed to prove reliability. While Google’s vehicles have driven long distances in testing – over 500,000 miles as of late 2013 – and have not caused any crashes while in automated mode, Shladover points out that this proves very little because their vehicles are monitored by drivers who take over when risky or challenging situations arise.
Legal and liability issues could also delay the emergence of Level 4 vehicles. A few American jurisdictions now explicitly allow automated vehicles on public roads for testing, and Bryant Walker Smith, a leading authority on the legal dimensions of vehicle automation, has found that automated vehicles are “probably” legal in the US; however, he also cautions that their adoption may be slowed by current laws. Laws will have to be clarified before Level 4 vehicles hit the mass market in the US and in other countries. Liability for crashes could also be a thorny question. If a human isn’t driving, presumably blame would shift to the manufacturer, or perhaps a supplier of system components, or a computer programmer. Resolving these issues could stall the emergence of automation.
While there is dispute as to when Level 4 vehicles will be on the road, most in the field agree that more limited forms of automation are coming soon. Some are already here. For example, Mercedes S-Class vehicles can simultaneously control speed and steering when road and traffic conditions allow, though the driver must continuously monitor the road. This is just shy of “Level 2” automation, since Mercedes’ system also requires the driver to keep their hands on the wheel. Numerous other vehicle manufacturers are developing advanced technologies that promise to take over driving duties, at least some of the time, on some roads. As technologies advance, “Level 3” vehicles could be on the market by 2020 to 2025, according to most experts. These vehicles would allow drivers to forget about monitoring the road and instead read or watch a movie, with the caveat that when the automated system is out of its depth, it would ask the driver to take over. (The takeover time is a matter of debate – anywhere from several seconds to several minutes has been suggested.)
Automation could be a boon for safety – or it could create new problems. On the plus side, it appears that crash avoidance systems already on the market may be effective. Of course, as machines take over more of the responsibility of driving, safety will only improve if the machines are in fact less fallible than humans. This might seem an easy task, considering the foibles of humans, but it’s worth remembering that some automation experts believe otherwise. And where driving is shared between human and machine, the safety impacts are especially open to question. A driver in a Level 2 vehicle might fail to continuously monitor the road, or a driver in a Level 3 vehicle could be engrossed in their movie and fail to take over control quickly enough when requested. In either case, automation could actually decrease safety.
After safety, one of the biggest selling points of vehicle automation is its potential for improving traffic flow, especially through increased road capacity. With their slow reaction times, human drivers can’t safely follow other vehicles closely, so even at maximum capacity, around 90 percent of the length of a freeway lane is empty. If machines could react quickly enough, road capacity would increase enormously. Some studies appear to suggest huge increases are in fact possible – for example, one study estimates that capacity would almost quadruple, and another finds quintupled capacity. However, their calculations consider endless streams of densely-packed vehicles. More realistic estimates assume that several vehicles, say four to twenty, would follow each other in tightly packed groups or “platoons”, with each group separated from the next by a large gap. These interplatoon gaps would provide safety and allow vehicles to change lanes and enter and exit the freeway. Studies that account for these gaps estimate that automation would increase capacity in the range of 50 to 100 percent (for examples, see here and here).
While the more realistic estimates of capacity increases are still very impressive, there are a number of caveats. First, short headways are possible only when automated vehicles are equipped with V2V, or vehicle-to-vehicle communication. Vehicles that rely completely on on-board sensors – such as the Google self-driving car, in its current form – cannot react quickly enough to the movements of other vehicles, so they would enable relatively small capacity increases. A second caveat: large capacity increases would come only when automated cars dominate the road. Studies have found that when fewer than 30 to 40 percent of vehicles on the road are capable of platooning, there would be little effect on capacity, and large increases would come only after the proportion of equipped vehicles exceeds 60 to 85 percent (e.g., see here). This is important, since new vehicle technologies will take some time to become commonplace. Imagine that as soon as automated vehicles hit the market, every new vehicle purchased is automated: it would then take two decades for automated vehicles to account for around 90 percent of vehicles on the road. If the rate of adoption is more realistic, but still rapid, it would take three decades or more before automated vehicles make possible large road capacity increases. A third major caveat: platooning is only feasible on freeways. Changing lanes, stopping at red lights, making left turns, parallel parking, stopping for pedestrians – such manoeuvres would make platooning impractical on city streets.
For city streets, however, there is the prospect of using automation to improve flows at intersections by coordinating vehicle movements. A good example is the “reservation-based” intersection, where there are no stop lights or stop signs – instead, cars equipped with V2I (vehicle-to-infrastructure communications) technology “call ahead” to a roadside computer that orchestrates the movements of vehicles and assigns time and space slots for vehicles to cross the intersection. Simulations show such an intersection could move almost as much vehicle traffic as an overpass – but so far, simulations haven’t included pedestrians and cyclists. Accommodating these road users in a reservation-based intersection would require signals with sufficiently long cycles, so capacity increases would be limited.
Vehicle automation would also bring a very direct impact: reduced or eliminated labour in driving. Time spent traveling in Level 2 vehicles could be less stressful, and could become more productive and enjoyable in Level 3 and especially in Level 4 vehicles. Profound changes in travel behaviour would result. As people increasingly let their robot chauffeurs deal with road congestion and other hassles of driving, travel by motor vehicle would become more attractive. Trips would tend to be longer and more frequent and travel at peak times would increase. Trip routes would also tend to make greater use of freeways with Level 2 and 3 vehicles, since it is primarily on these roads that the vehicles will be able to operate in automated mode.
These induced demand effects would tend to increase road congestion. Freeways would be the exception – if platooning-capable technology becomes widespread, freeway capacity would increase and congestion would drop. That is, until the surplus capacity is taken up by the “triple convergence” of mode shifts, route changes, and change of time of day of travel. However, the increase in freeway traffic would be constrained by capacity limitations on the rest of the road network – as freeway travel increases, new bottlenecks would form on streets near freeway entrances and exits, where automation does not boost capacity, thus restricting the volume of traffic that can access the freeway.
The upshot of the above observations on the capacity effects of automation is that even when the potential freeway capacity increases enabled by platooning are fully realized, automated cars would nevertheless be able to carry far fewer people than bus or rail on a given right-of-way. And, as mentioned, capacities on streets will be largely unaffected. Because the capacity improvements made possible by automation would be limited, we will still need buses and trains when space is in short supply and we need to transport large numbers of people. Larger vehicles will still fit a lot more people into a given length and width of right-of-way than platoons of small vehicles will be able to carry. As Jarrett would say, it’s a simple fact of geometry.
So, vehicle automation will not render large transit vehicles obsolete. On the contrary, it could enable significant improvements in bus service and increases in ridership. Automated steering enables bus operation at speed in narrow busways, which reduces infrastructure and land costs. It also enables precise docking at passenger platforms, which improves passenger accessibility and reduces dwell times. Automated control of speed enables bus platooning, allowing buses to effectively act like trains. Automation can be taken further yet: a driver in a lead bus can lead a platoon of driverless buses, thus providing high capacity with low labour costs. Similarly, individual buses or platoons can operate driverlessly, thus enabling increased frequency with low labour costs. “Dual mode” operation is also possible: imagine a busway where chains of buses leave the city running like a train until they separate at a suburban station, where drivers board and take them onward onto various local routings.
Some of these forms of automation have already been implemented in BRT systems. For example, a system in Las Vegas employed optical sensors to enable precise docking at passenger platforms, BRT buses in Eugene, Oregon used magnetic guidance to facilitate precision docking and lane-keeping in a pilot project, and systems in Paris and Rouen, France, and in Eindhoven, the Netherlands, use various types of guidance systems. While bus platooning and driverless operation have not been deployed so far, these applications could be achieved given sufficient technological advances – or by using a low-tech shortcut. The simple solution is to keep other vehicles or humans out of the way of the automated bus. If buses operate on busways with adequate protection, platooning and driverless operation is possible with existing technology. (Similarly, current driverless train systems are able to operate driverlessly, even with decades-old technology, by virtue of the well-protected guideways they run on.) Developing a vehicle capable of driving itself in the simplified environment of a protected busway is a considerably easier task than developing a vehicle that can drive itself on any road, anytime.
With the arrival of Level 4 automation, driverless buses could operate on the general road network. This would make it possible to operate smaller buses at higher frequencies, since labour costs would no longer constrain frequency. If you shrink driverless buses small enough – and provide demand-responsive service for individual travelers – you end up with driverless taxis. This points to the possibility that public transit service may be more efficiently provided by driverless taxis (or driverless share taxis) in low-density areas, thereby replacing the most unproductive bus services and improving transit productivity overall. (Of course, while automation could boost productivity, even driverless demand-responsive service would still have low productivity where densities are low.)
While it’s a seductive story that driverless cars will transport us to a realm of much improved safety, convenience, and efficient road use – and where public transit has dwindled away – the future is likely to be more complicated. Advanced automation is indeed coming soon, though we might not see Level 4 technologies for a while. Automation could improve safety, though it could also generate new problems. It could also improve road capacity, but the improvements would be limited in several ways. All this suggests that we needn’t worry about (or celebrate) how vehicle automation will make public transit obsolete. Instead, let’s focus on how to use automation to the advantage of public transit.
Fantastic post. Great info on driver-less vehicles in general, but also a wealth of reminders about how dependent transportation is on so many other factors.
This is a very well thought out post. I’m sure your research is even more so. Indeed, I can’t immediately find any points of disagreement.
But…as a cyclist, a pedestrian–as a human among machines, I can promise you that for the rest of my life, I will be as violently opposed to ‘driverless’ anything as it is possible to be. (except in extremely limited cases like Vancouver’s fully grade-separated metro)
I do not see anything good that can come of this. Not one single thing. Nothing but ‘pedestrian fatalities’ mindlessly tallied in a database and rationalized by statisticians. Nothing but an even more mindless whirring to want I want to be the most human of places.
A driverless car won’t _intentionally_ try to run a cyclist off the road or have road rage. That right there will be a great improvement for cyclists.
We could argue timetable or desirability all day and not come to an agreement. So let’s consider just a level four world.
If it’s dominated by personally owned autonomous vehicles (AV’s), unshared beyond the household unit, then we could very well find ourselves with a massive increase in VMT and even more growth in VHT as up to half of all vehicle trips could be deadheading between missions. We would then have to work even harder to make sure that public transit were attractive and available to take up part of the load.
Alternatively, if the Earth Institute’s “Transforming Personal Mobility” is anywhere close to being on target and shared AV’s at an unsubsidized total cost-per-mile of 41 cents or less were a smartphone swipe and less than a minute or two away, then it’s over for transit as we know it. Transit could still be an effective tool during commute hours, as circulators in congested areas, or in certain specialized situations.
The centuries-old model of fixed-route, fixed-schedule, fixed-price transit just isn’t where it’s at.
Why no discussion of the massive socio-economic impact that shared Level 4 automated vehicles will have? As soon as we raise the average vehicle per person from around 1.1. upwards then we see all sorts of capacity benefits irresespective of close following and platooning. There’s the Earth Institute, Columbia University study that suggests one sixth the number of private cars would be required and the average transportation cost for a person would fall by 40%.
A recent Singapore study shows that they could serve their entire transportation needs with only one third of the current number of passenger vehicles by using shared automated fleets.
This study is a great introduction, but it doesn’t tell a complete story.
A recent Morgan Stanley study estimated that autonomous vehicles when fully deployed will save the US $1.3 trillion/year – and that is the base case figure – which is 8% of the 2012 GDP.
In my presentations I keep making the point that fully automated vehicles are both the biggest opportunity and the biggest threat to transit.
One thing that’s not mentioned about vehicle platooning here, though, is that it’s also bounded by road cleanliness. The higher the speed of the vehicles the more road grit is thrown up. Trailing vehicles suffer more and more impacts from road grit the closer they are to the vehicle in front. Vehicle damage becomes an issue.
The world is a messy place. In areas that see snowfall it’s going to be difficult to keep the roads clean enough to allow platooning at all at certain times of year. Some areas that don’t see snowfall are remote yet still have high traffic roads (central California, freeways around Las Vegas and other cities in the U.S. south west, etc.) so there’d be increased infrastructure costs to keep them clean.
Besides all this, debris often falls from vehicles so platoons are likely to be thus affected at completely random times, on random patches of road. V2V can help, here, but there’s still going to be some initial vehicle or vehicles that are impacted.
Due to this reality, the road capacity gains from platooning are frequently overstated.
You are correct that claims of imminent increased capacity from reduced headways alone are overly optimistic. Automated no-signal intersections are very far away as they don’t allow human drivers.
But this misses the point about other technologies which come much sooner and have dramatic effects on congestion and transit.
Congestion arises primarily from traffic demand surpassing road capacity, compounded by two human behaviours — accidents and irrational and irregular driving patterns. Reduce accidents, along with slowing down to look at accidents, and you cut congestion.
The big wins, however, can come from road metering capabilities that go far beyond what we can do today, even in places like Singapore. These don’t require robocars — smartphone equipped human driven cars can and must participate. With proper metering, so you never allow more cars on a road than it can handle, congestion can be seriously reduced, and induced demand is no longer the problem it was.
Robocars can easily handle dynamic lane and street redirection (today we insist on barriers for this most of the time for humans so don’t do it a lot.) I also predict robocars will move a lot of traffic to half-width vehicles for 1-2 people which can pack 2 to a lane when they encounter one another.
However, to get to transit. Transit will have an impossible time competing with cheap, personal door-to-door transportation on your own schedule, except at times where congestion is so bad that only the private ROW of the transit line saves it. Even with the major subsidies it receives it may have trouble competing on price and can’t compete on energy efficiency — small private vehicles are much more energy efficient per passenger than transit vehicles with non-peak load factors.
This presents a problem for transit lines as they will thus only run at rush hours. Current transit economics don’t work well (in most places) with buying trains or even buses only to run a few hours a day.
(The need for drivers only at rush hour would also be a problem but self-driving buses may solve that.)
I have envisioned a different type of transit, which consists of small single-person vehicles which operate in local areas, and coalesce simultaneously at a transfer station where a van awaits. Everybody moves to the van which goes to the rough area they all wish to travel to, where the van stops beside a group of single person robotaxis which take the passengers the last mile. This vision is almost as fast as door-to-door, and is very efficient in use of energy and road space. Current modes will have trouble competing.
Great article! Nate Wessel, I couldn’t disagree with you more – as someone who cycles everywhere, driverless cars can’t come soon enough. They will not hit the roads until they are 100x better at not hitting pedestrians and cyclists. Every single time I’ve been in a collision or a close call (none serious, thankfully) it was driver error, usually by breaking the law.
I do have one niggling issue that never gets addressed in all these discussions: variation in vehicle speed. Right now I presume there is large statistical variation in speed except during total gridlock. Every day I see the fastest, most aggressive drivers travelling at twice the average traffic flow speed. Presumably, driverless cars won’t be allowed to speed (they may raise freeway speeds but not city speeds), weave aggressively in and out of traffic, run yellow lights, etc., so the average speed will be pretty much the same. I wouldn’t even criticize the “speeders” necessarily: we all rush when we’re late, and I would trust most of my friends to drive a little bit faster if they had a solid reason to. I cycle faster when I’m late.
So I predict in some situations, driverless cars will be slower because they will obey the traffic laws and drive slower. AND I predict the speed of the fastest cars will be much lower too, making it harder to speed and creating a bizarre market for used non-driverless cars as a faster way to get around, once most new cars are driverless.
You can draw a rough analogy to the aviation industry: it used to be that if you wanted your plane to go faster, you flicked the turbocharger switch, fired up the afterburners on Concorde, or flew more aggressively. “We’re late can you slam the throttles forward and throw caution to the wind?” just isn’t something you hear much at airports (it’s why we all love Top Gun so much.) Automation has brought great leaps in aviation safety and efficiency, but not speed. Any variation in speed is usually a delay to weather or congestion.
I realise this was probably beyond the scope of research, but when zipcars become robotaxis the politics of sprawl repair get a heck of a lot easier.
It’s harder to moan about your god-given right to a free parking space everywhere you go, when e-robotaxis are dropping aged customers off for a few bucks a journey, and only stopping to charge off-street.
Interesting that, on this of all blogs, the word ‘coverage’ doesn’t appear. (Ridership does, I see.)
The e-robotaxis + e-robobuses world envisioned by the end would seem to map pretty directly to coverage + ridership goals. Good thing.
Now, about those unions…
(Actually on that, I think the easiest transition would be to replace retiring drivers with robots, i.e. shrink the pool at the natural rate. To the extent that driverless buses can be deployed more quickly, drivers could be redeployed as roving conductors among the much-enlarged fleet, spot checking tickets, and meeting and greeting.)
I’m not sure I buy the geometry argument for most North American cities, which have population density levels far below that of Manhattan. Car2Go employs the Smart Fortwo, which is small to begin with. If their service offered driverless vehicles that also happened to be narrow microcars no wider or longer than a motorcycle (e.g., the Tango Commuter Car), that would free up significant road space. The geometry argument assumes a future world in which everybody rides driverless mid-sized sedans. While people generally purchase a car that’s large enough for any conceivable task that they might have throughout the course of the year, the majority of driving is single occupancy, and people can select purpose-specific vehicles when using on-demand car rental or “roboxtaxi” services. Car sharing services that people use for small errands and commuting can offer vehicles half the size of a Smart Fortwo, which can allow people to bypass congestion in dense close-in areas. After driverless technology becomes mainstream, such services also become economically feasible in the suburbs. For cities no denser than Portland or Minneapolis, I think such an amenity would undermine the case for further investment in public transit, although the geometry argument still might apply in places like Hong Kong or Tokyo.
One also needs to ask what incentive people who can currently afford cars have to ever use public transit. I think it’s because city centers are too expensive to park in regularly and because commuters would rather focus on activities other than driving. The introduction of driverless cars would remove both of these two incentives to use public transit. Absent those, what other incentives exist? You’d be left with a public transit system that exists solely to serve the underclass.
Given the apparently promising state of driverless car technology, why is it that driverless trains are still rare and require billions of dollars and years of upgrades on existing lines? (Optimistic plans suggest driverless operation might be possible in London by 2025 and New York by 2029.) Trains have only one degree of freedom and no need to deal with pedestrians or cyclists. Recognising railway signals is surely not harder than recognising stoplights. Safely closing the doors (without platform screen doors) may still be an issue but a single employee can monitor and control door closing for many stations.
I understand there are organisational obstacles to driverless operation in many parts of the world, but I haven’t heard of any system anywhere trying to replace its drivers with cheap PCs and webcams (even in places with extraordinarily good transit organisation like Japan and Switzerland). Bizarrely it’s instead sounding like many cities will have driverless cars long before they have driverless trains. What am I missing here?
Anon256 – public sector is less efficient than private sector?
Eric – the various railway companies in Japan are all private (and mostly very profitable) and still use drivers on the vast majority of lines, as do most other private rail operators all over the world, so this doesn’t really explain it at all.
@Anon256 What about Line 1 in Paris? That got converted with no problem, and I think there is an example in Germany where they built an extension of an existing line and bought new driverless trains, which operated together with driverful ones until the latter were converted. Driverless metros are quite common among new-build systems (DLR in London, Line 14 in Paris, JFK Airtrain in NYC). And even on metro systems that have a driver, the trend has been toward complete automation since at least the 60s. The real problem is in mainlines and street railways, both of which are by definition much less controlled environments than metro systems, and which have thus not been automated to nearly the same extent.
anonymouse – The automation of Paris Line 1 was decided in 2005 and work began in 2007, but the first automated train did not run until 2011 and the line was not fully automated until 2012. €100M of signalling etc upgrades were required (which goes a lot further in France than in the US), as well as all-new rolling stock. Why was this necessary? Why can’t existing vehicles be retrofitted (as the Google cars are) and use computer vision to interface with the existing signalling systems (as the Google cars do) rather than requiring signalling upgrades? And why should mainline railways particularly more difficult to automate (assuming grade-separation as present on e.g. most mainlines around Tokyo), aren’t the train’s allowable movements still fully determined by speed limits and visible signals?
It’s never going to happen. It’ll be hundreds of years beyond these wildly optimistic projections.
The thing is, automation has been *completely solved* for trains. And yet no country allows an automated train on a route with grade crossings.
No country will allow an automated car on a route with grade crossings (i.e. anything other than a freeway) if they won’t allow an automated train in the same situation. That’s the bottom line.
“They will not hit the roads until they are 100x better at not hitting pedestrians and cyclists”
Which won’t happen, unless they’re made *very* conservative.
“So I predict in some situations, driverless cars will be slower because they will obey the traffic laws and drive slower. ”
Correct. Which will make them unpopular.
Ain’t happening by 2030, no way no how.
“Due to this reality, the road capacity gains from platooning are frequently overstated.”
Basically nonexistent. In any real-world situation platooning is unsafe. The catalog of reasons is long enough that they’re not all going to be dealt with.
It’s so easy to estimate the gains from fantasy technology really high. In fact, the gains are… low.
The Technology Review article is an excellent summary of how impossible “driverless cars” are. Thanks for the link.
” Automated driving will at first be limited to relatively simple situations, mainly highway driving,…”
So, useless for most purposes. I doubt it can even handle our snow-covered, pothole-filled, black-ice expressways, let alone the country roads.
Maps are no good; the map WILL NOT be up to date and the “robot car” will drive right off into a river, not realizing that the bridge was replaced by one 20 feet over last week. We already have idiotic people “following their GPS” doing this.
Northeastern weather makes the sensors pretty dumb; the problem of detection in arbitrary types of snow is difficult even for a human (who are actually VERY VERY GOOD at this sort of visual pattern matching) and quite impossible for a programmer.
No programmer will *ever* be able to spot all kinds of road signs, especially not the improvised ones used for urgent warnings. Nor will they realize that the silhouette of a deer at the *side* of the road means “slow down to 10 mph before it jumps out at you”.
Please note that the BMW test car proceeded to unsafely run up towards a swerving car, and the driver had to take over and hit the brakes.
These things aren’t even close to ready for use in real life.
Honestly, I think that driverless cars are the most overhyped technology out there. All the predictions of greater capacity supposes flawless operation from thousands of cars with not a single mistake or failure of communications. I just don’t see it.
Here’s the thing: computers are idiots. They rely on simple algorithms to simulate “thinking” and can do an extremely large amount of calculations, but they rely on inputs from sensors and the like. If a sensor tells them they are now traveling at 2000 miles per hour because of a glitch, the computer will assume it is true and react accordingly, a human would doubt the result and verify, computers do not have that ability.
It seems to me that automated cars would rely on a vast quantity of sensors and wireless communications to do anything, any of these sensors could fail, causing the car to act irrationally and dangerously. These sensors would also be extraordinarily expensive, because reliability and quality are very expensive to provide. I’m not even talking of possible software failure.
I’ve seen images of the Google car, it’s driven mainly on roads and streets out in the suburbs with very low congestion, often in clear weather and with well-made lane markings. And always with a driver at the helm.
The reality is that there is a reason why humans drive far apart on freeways, at those speeds you need some buffer to be able to maneuver. Even with a reaction time of 0 second or close to it, you still have inertia to deal with. People talk of platoons of cars with almost no space between each other, but if only one of those cars behaves erratically, whether because of software or hardware failure, or the failure of a mechanical component, there would be no margin for error, all the cars in the platoon would then be involved in an accident… and then, who is responsible? The owner of the car? The maker of the vehicle? The programmer of the software?
The sensors Google put on its car cost 250 000$. Even if they are able to lower the quality of them and the cost, those sensors will likely remain very expensive. Are people ready to pay thousands more for an automated car?
@Anon256 (and fellow commentators on him)
Unmanned train operation (UTO), which is the most advanced stage of automated trains, are a bit more complicated because most train systems start from a baseline paradigm that allows tighter limits than that of road vehicles, where visual human input is, so far, almost the only way of assessing the course of action.
Trains are not driven on basis of visual reckoning of other trains, at least not heavy rail, subways etc. Only mixed-traffic trams and streetcars are usually that ‘dumb’. Subways, heavy trains, high speed trains cannot possibly be operated at such speeds on that basis of ‘see and be seen.
Just take a look at the speeds train travel, their braking distances and so on. Add to that the fact trains are, by their very nature, fixed on running over their tracks (they can’t swerve to avoid an imminent collision, for instance).
Trains also have a much steeper failure threshold than road vehicles. They are incredible safe and reliable while on track, but derailment is often a catastrophic event, while road vehicles deal with different stages of grip, skid and drift.
Now modern train operation is already heavily automated: automatic breaking if a red signal is violated, speed limiters, controlled acceleration and what else. These ATO/ATP systems rely on modern signaling that, for virtue o the specifities of rail, need to be fit into the infrastructure. You cannot rely on trains-talking-with-trains on busy systems like subways, unless you want to slash capacity.
When you move towards a fully driverless train, you need precise platform alignment control. The final step (removing the operator) is not really the most difficult one: installing proper automated train operation systems are.
Upgrades to driverless operations also ensue much higher safety and, often, increased reliability and sometimes capacity, as trains are no longer subject to fixed blocks, but dynamic signaling that keeps them as close as possible.
Here’s the thing: computers are idiots. They rely on simple algorithms to simulate “thinking” and can do an extremely large amount of calculations, but they rely on inputs from sensors and the like. If a sensor tells them they are now traveling at 2000 miles per hour because of a glitch, the computer will assume it is true and react accordingly, a human would doubt the result and verify, computers do not have that ability.
Hogwash. Computer control systems can and do, today, reject dubious or otherwise bogus sensor inputs; the problem of sensor failure is a well-known issue in the design of control systems.
There are plenty of obstacles to Type 4 operation, but this is not one of them.
@Andre Lot: The fact that trains are not “see and be seen”-based surely makes automation easier, the camera just has to recognise signals and not other trains/hazards. Similarly I would think platform alignment (stopping at a precise marker) is even easier for computers than it is for human drivers. What exactly do train drivers do today that a computer with a camera can’t? Trains-talking-with-trains is exactly the sort of expensive and complicated system I’m asking about avoiding, with a simple system that does exactly what the driver does today.
@Anon256 Thank you for your detailed analysis.
On driverless cars, I think this thread does not recognize the singular significance of dead-running for automated personal vehicles. Given that there is no indication that driverless taxis would fundamentally behave any differently than driven ones, (except perhaps in greater scale and approximating the nature of Zipcar), our discussion should focus on automated personal vehicles (APV’s).
APV’s would still require facilities similar to that of today and the past: regulated parking, both at origin (home) and destination (work, store, etc.). In order for APV’s to contribute as minimally as possible to congestion (certainly law by the time the idea becomes relevant), dead-running would have to as well be minimized; i.e, parking, as it is today and has been in the past and into the foreseeable future, would still have to be in proximity to the final destination. Even if cars are self-parking, and chauffeur riders to the front door, cars would be performing the act of valet, which by definition requires quick call times when the riders are leaving (If a driverless car has to dead-run 5 miles to get from the riders to a garage on the edge of town, it will take just that long to retrieve it). Train systems are affected by this same phenomenon. Secondly, there is no indication that streets dominated by automated taxis and personal vehicles would be any less congested than today (especially when you add other users like freight and service vehicles), when speaking in terms of speed of vehicle per unit of road space. Dead-running would only add to this. It is by these constraints, that as cities grow and densify, the need for the accommodation of parked cars will still be as relevant and restricted as today, if not more.
Thus, wholly substituting the popular use of APV’s and automated taxis for modern rapid transit would require an urban structure similar to that dominated by today’s automobile culture (which, if the dominant form of transport–much as the use of cars has been for the past half century in the U.S–would, by nature, resemble large parking structures and strip malls flanking wide blvd.’s [or, as a Chinese model, ^that scaled up an order of magnitude]); a model that is incompatible with a healthy, dense, highly pedestrianized urban core that relies on the critical mass of millions of human beings per day in a relatively small area (which, I believe, is the central setting of our discussion), and thus which is most suitable for public transit. Think London, or Paris, or Berlin, or Moscow, or Istanbul, or Mumbai, or Shanghai, or Tokyo, or Melbourne, or San Francisco, or Vancouver, or Chicago, or Toronto, or Washington D.C, or New York, or Boston, or Santiago, or Curitiba, and you will see my point.
Therefore, I believe transit will never go out of style so long as people live in cities denser than suburbs. @Anon256 made a very good case that the bounteous possibilities of automation are not alien to guided-rail transport, both historically and into the future. Technology scales across media, and whatever improvements are made to automation will make transit in general just that much more efficient, which will in turn make it, in its myriad forms, feasible long into the future.
Whether driverless cars can be granted the kind of presence that many people visualize on the pedestrianized and bicyclized streets of tomorrow is a subject for a different discussion. I also think this thread should talk more about the effects of automation on taxis and buses, both of which have a much longer way to come than trains, and which are central to the premise of Mr. Loro’s concluding argument. In the meantime, chew well.
Thank you for your time.
This made me laugh for ten minutes when I read it last night-
From ‘Transport for Suburbia’ by Paul Mees (2010): “The most popular exhibit at the 1939 World’s Fair was Futurama, a General Motors sponsored depiction of the city of 1960: ‘trains’ of electronically controlled cars ran efficiently on freeways, with vehicles de-coupling to operate individually on local roads. Despite promises, name-changes and lavish government funding, the concept, most recently re-christened Intelligent Vehicle-Highway Systems, has gone nowhere in 70 years. Futurama is now better known as a cartoon series satirizing technology-based utopias.”
and here I thought the wait for my connecting bus was too long. Glad I’m not waiting for that driverless taxi.. ha ha ha.
I read Antonio’s article with interest and don’t disagree with most of his summary. However, technology develops much faster than we give it credit for. And while we don’t know exactly how close we are to full autonomy, it will likely be sooner than later – let’s not underestimate the rate of technology development. When the technology is mature enough it will have an impact on both road transportation, road density and transit – particularly in urban areas. Autonomous technology for trains is already here (SkyTrain), albeit on a segregated guideway. In urban areas, for example, parking requirements and demand will decrease significantly relieving congestion. Platooning of vehicles on major corridors could be a first phase of the roll-out of mature technology, increasing road capacity – this could apply to cars, trucks and transit.
There will always be a need for public transit; particularly in urban areas – it is difficult to beat the acrrying capacity of public transit vehicles, road or rail. Whether buses will be driverless is a longer term issue to resolve – one still needs to assist persons with disability and ensure fares are being paid, these issues will be a challenge.
One last point I would like to make is that we are spending billions on transit and transportation infrastructure every year. This is infrastructure that will last 100 years or more if properly maintained – surely strategic urban and transportation planners should be looking at how autonomous technology will impact those projects and plan for accommodating it in original design as opposed to expensive retrofits down the road.
“Automated steering enables bus operation at speed in narrow busways, which reduces infrastructure and land costs. It also enables precise docking at passenger platforms, which improves passenger accessibility and reduces dwell times. Automated control of speed enables bus platooning, allowing buses to effectively act like trains. Automation can be taken further yet: a driver in a lead bus can lead a platoon of driverless buses, thus providing high capacity with low labour costs. Similarly, individual buses or platoons can operate driverlessly, thus enabling increased frequency with low labour costs. ”
Or, you can just build an automated train, which does all the same things, but is cheaper and more foolproof. Angled steel wheels on steel tracks allows a huge “platoon” of train cars to run and “dock” perfectly.
Why do people love to chase after silly gadgetry?
France built some idiotic “rubber tired metros” because of Michelin, which wanted to sell rubber. It turns out to be a dumb idea.
Given the spatial and environmental damage that cars inflict on our urban spaces and the planet overall, why would cars of any stripe (driverless or otherwise) be desirable at all?
We should be building cities, towns and villages where cars are not needed or wanted. These places can then be intra and interconnected by high quality trains or trams or buses.
It would be a better world for everyone.
Great blog.Thanks for sharing your information with us
What do you think of Tesla Motors plan for driverless cars?
It’s a wonderful thought, but we’re probably several years if not decades away from fully automated vehicles. The public is too scared right now to let computers take over the roads.
Its a amazing though but When the technology is mature enough it will have an impact on both road transportation, road density and transit – particularly in urban areas.