Proponents of driverless cars often suggest the technology will make all sorts of significant changes to transport. Gone will be car ownership with people just hiring a car when they need one, like taxi’s only easier and cheaper. As such they say gone too will be the need for public transport, especially when you take away some of the benefits PT currently has like being able to do other things such as work, read, use a phone or even sleep. Further congestion will also be a thing of the past with these smart vehicles able to better work together rather than the randomness of humans. Of course the biggest and likely most accurate prediction will be safety as for a start these cars will obey the road rules so no speeding, no running red lights or any of the other bad habits human drivers have.

That all sounds wonderful however an article from CityLab highlights research showing that at least for some time driverless cars could actually make things worse on the roads.

A new simulation-based study of driverless cars questions how well these two big secondary benefits—less traffic and more comfort—can coexist. Trains are conducive to productivity in large part because they aren’t as jerky as cars. But if driverless cars mimic the acceleration and deceleration of trains, speeding up and slowing down more smoothly for the rider’s sake, they might sacrifice much of their ability to relieve traffic in the process.

“Acceleration has big impacts on congestion at intersections because it describes how quickly a vehicle begins to move,” Scott Le Vine of Imperial College London, who led the research, tells CityLab via email. “Think about being stuck behind an 18-wheeler when the light turns green. It accelerates very slowly, which means that you’re delayed much more than if you were behind a car that accelerated quickly.”

For their study, Le Vine and colleagues simulated traffic at a basic four-way urban intersection where 25 percent of the vehicles were driverless and the rest were standard. In some scenarios, the driverless cars accelerated and decelerated the way that light rail trains do—more comfortable than, say, riding in a taxi, but still a little jerky at times. In other scenarios, the cars started and stopped with the premium smoothness of high-speed rail.

Within these broad scenarios the researchers also tested alternatives that reduced speeds but improved smoothness, such as longer yellow lights or following distances. All told they modeled 16 scenarios against a baseline with all human-driven cars. The researchers then ran each simulation for an hour, repeated it 100 times, and calculated the average impact that scenario had in terms of traffic delay and road capacity.

In every single test scenario, driverless cars designed to create a comfortable, rail-style ride made congestion worse than it would have been in a baseline scenario with people behind every wheel.

So cars with fast acceleration and deceleration are obviously easy to make but that’s not what people are likely to want if you’re also trying to do some of the other activities mentioned earlier. Regardless traffic generally moves at the pace of the slowest vehicles so all it takes is one slow driver or driverless vehicle and many others will be slowed down too. I bet they won’t say that in the marketing brochures.

The interior of the Mercedes-Benz driverless car concept

I suspect this isn’t the only aspect of driverless cars that could create congestion. As an example the driverless taxi model that most people say will happen, is likely to result in a lot more vehicle movements as cars reposition themselves to pick up additional fares. That means that where roads are generally congested in one direction only, with driverless cars congestion could occur in both directions.

Share this


  1. I’m actually curious to know how big the impact of driverless *buses* would be; what proportion of the bus cost is attributable to the driver component?

      1. Given that the farebox recovery is currently around 1/2, if that kind of cost reduction made it ~100% then you could probably *more* than double the services for *less* money.
        (depending on available demand really)

    1. Actually it shows the opposite. Smooth and consistent driving at optimal speed and vehicle-separation for max traffic throughput is what reduces congestion. Of course with driverless vehicles it may be possible to safely reduce separation to less than the currently recommended “2-second rule”, which could also help reduce congestion.

        1. The effects of the demise of the rubber-necking driver would be quite spectacular. No more would everyone be stuck behind idiots who are more interested in gawking at someone else’s misfortune than in paying attention to the road ahead.

      1. Only on open roads or motorways. On urban/suburban streets with traffic signals everyone behind the driverless vehicle that brakes early and accelerates slowly is potentially held up by missing the light or the gap in the crossing traffic. The example of being stuck behind a heavy truck is apposite; on the motorway it doesn’t matter, but when that same truck is the lead vehicle at the queue at traffic lights it can reduce throughput of that lane through the intersection quite significantly because of its much slower acceleration.

        1. You’re right, but possibly only when it comes to driver-controlled vehicles. The normal start-up process for a traffic queue is a messy affair as following drivers have to carefully gauge their acceleration to avoid ramming the vehicle in front. Even if the vehicle at the front accelerates hard away, subsequent vehicles are bound by driver-response lag and the need to allow a considerable factor-of-safety in manually regulating separation under acceleration.

          With automated vehicles the acceleration process of a queue is likely to be much more homogeneous and my hunch is that traffic throughput could be greater even at lower acceleration rates.

        2. All cars should be able (in theory) to accelerate at the same speed immediately when the lights go green, which clearly doesn’t happen at the moment. So, I am very confident your intuition is correct.

          Furthermore, a driverless car can react faster to a green light than a human. Further, furthermore, safety delays between one light turning red and the next turning green can be reduced due to driverless cars obeying the stop signals. Sure, running red lights saves _some_ people time, but it’s inneficient.

          None of these improvements are going to happen soon. I don’t think driverless cars are the silver bullet to congestion, but it seems unlikely they are going to make it worse.

        3. Not to mention, if the LIGHTS are tied into the network, the cars could potentially regulate their speed so that they never stop at a red light

  2. I think this study can be a little misleading as it seems like they are assuming that speed, rule adherence etc. is the main cause of traffic congestion.

    I would like to point out that driverless cars can potentially reduce driver error on the road, and combined with research that shows that many sources of traffic jams are caused by driver error, it would suggest that at least part of the cause of traffic congestion can be effectively cancelled out by reducing driver error dramatically.

    Related reading:

  3. I see that the full study document is only available for purchase (which I won’t be doing), so I can only base my comments on the free abstract. However my immediate response is that a common trap has been fallen into, and that is to assume a direct relationship between acceleration rates (or deceleration rates) and passenger discomfort, while taking no account of rate-of-change of acceleration – i.e. “jerkiness”. The reality is that within reasonable bounds, acceleration per-se is not the main determinant of passenger discomfort, provided it ramps up and ramps down smoothly. If people can sense it coming and going in time to comfortably brace themselves there is usually no problem. It is sudden changes in acceleration (either forward/backward or side/side lurching) that tends to make a ride uncomfortable. Thus if driverless vehicles are smoothly controlled with limits to rate-of-change in acceleration, the actual acceleration rate can be a fair bit higher than typical urban transit systems without being uncomfortable.

    In this respect, I believe many transit systems sell themselves short by imposing unnecessarily restrictive acceleration caps in the name of “passenger comfort”, while some systems successfully offer higher acceleration rates but with carefully-controlled rates of onset/decline. In Auckland’s case, the EMU’s are restricted to a conservative 1.0m/s² service acceleration (0.9m/s² for Wellington’s Matangis ) but passengers regularly experience more than this in the average bus (or car), where discomfort is typically caused by excessive rate-of-change.

    A smooth vehicle-driver who avoids making sudden changes should be able achieve at least 2m/s² (= 0 to 100K in about 14 sec, assuming steady acceleration all the way) without upsetting passengers. Therefore I think it is fair to say that driverless cars should not need to be limited to typical rail transit acceleration rates.

    I also posit that higher-accelerations-with-gentle-onset could likewise be applied to trains without significant degradation in comfort, though whether this would achieve a meaningful performance gain would depends on the frequency of stops.

  4. Have to be careful that you don’t start to sound like a negative Nancy here TB. Driverless cars are pretty cool, no reason to get defensive as they are unlikely to threaten good, functional and affordable PT.

  5. Meh. Simulation is next to useless in the real world. Too many incorrect assumptions, adjustments, modifiers and factors to give any real answer. We won’t know until it happens and it is still decades away before we get real market penetration.

    If the 25% of driverless cars actually set the speeds and prevented all the shockwaves from happening, you could reduce congestion.

    However I said it before, driverless cars probably won’t reduce congestion. It just means you can fit more cars on the roads.

    1. “still decades away before we get real market penetration.” esp in NZ which is historically at least half (if not a full ) century behind most technological innovations when it comes to transport.

  6. Most of the often stated benefits of driverless cars cant actually occur until all cars are driverless, and that is a long way away.
    But one advantage that could immediately reduce congestion is driverless car pooling taxis. It would be quite common at rush hour to have a multitude of people leaving from a similar origin and arriving at a similar destination; a car pooling taxi could turn a lot of currently separate journeys into one.
    As for acceleration, the real advantage of driverless cars is that they can all start accelerating together from traffic lights, compared to humans who tend to start accelerating after the person in front of them has started moving rather than when the light turns green. Of course this can only happen when all cars are driverless.

    1. So you catch a taxi to work and it goes around your local streets picking up others for what? 10-15 minutes? What if it gets to each one and has to wait for a few minutes for each person to come outside and jump in? It would be a nightmare like an airport shuttle everyday back and forward. The only way I can see that working is via bus stops, in which case you might as well just use a driverless bus network instead of thousands of meandering driverless cars.

      1. I think you order the car on an app, you get a notification when car is 2 mins away. Car can see where you are based on GPS on your phone. You meet on street with a no-show cutoff (1-2 mins) whereby you loose the ride and are charged a fee. Driver picks up others in a similar fashion, or you pay a levy to be sole occupant. With each occupant the cost goes down, which provides an incentive for the waiting. App gives you timing of drop off at destination which is contingent on information such as number of picks ups, traffic flow (gleaned from the online network) etc. If this drop off is not met within a given time frame, you get a reduced fee which is staggered etc.

        1. I also imagine different styles of car being available, you might have some with completely separated compartments for each occupant which would be great for car pooling, then some might be like a standard car for use when you need to take the family, or you can order up a van if you’ve got the family and some of the kids friends. It would be so much more flexible than owning a car or than taking PT.

    2. I can’t see it taking off in any big way with our current road pricing setup. The benefits of carpooling are mainly vehicle operating costs, but these are pretty small. The limited take-up of carpooling in today’s cars shows that the small saving is just not worth the hassle for most people. Something like HOV lanes where people can personally recoup some of the travel time benefits of their vehicle sharing would get more people on board with it.

  7. The key to fixing congestion is the vehicle’s width in proportion to the width of the road. I would be surprised if the study simulated automated driving with motorcycles, scooters, and tandem seated narrow commuting vehicles which are the most likely type of vehicles to use self-driving tech. Building a self-driving car looking like that picture? Nope.

    1. Traffic in Saigon is 95% single width vehicles and congestion is horrendous. It hasn’t fixed a thing, except crossing the street. Fixed that up real good.

        1. Well that proves my point, narrow vehicles do nothing for congestion. All they might do is let more people be stuck in congestion at the same time. Yay.

          And that’s if you convince ~100% of drivers to take them up while banning regular width buses, trucks and vans. If you only get a smaller percentage you’ll simply have more congestion by complicating all intersections and turning movements.

        2. This article disagrees with your assessment: “The answer to the world’s urban traffic congestion may be as simple as creating policies to promote motorcycle (narrow commuter vehicle) commuting…Quite clearly a change to smaller (narrow) road vehicles offers the cities that can engineer it a more hospitable environment than those that don’t or can’t. The inhabitants of those cities will waste less time in transit and the air will be cleaner.”

          Check out this Tango filtering through traffic in California: So you don’t want to lanesplit? How about creating a narrow lane on the edge of traffic for NCV’s, motorcycles, and scooters? In both of these videos, it’s clear that the narrow vehicles are absolutely breaking congestion and lessening congestion, even on roads dominated by cars with side by side seating.

          This isn’t complicated. Narrow vehicles pass traffic congestion easier and ease congestion.

        3. Your idea of narrow cars to me is in the same category as car pooling and everyone using motorbikes. Yes, in theory both those things could reduce congestion. But in the real world where we all have to live, people aren’t choosing those options so they are not a solution.

          Anyway, induced demand means that any free space created by your narrow cars would just then be filled by someone who has just figured out that driving the SUV to work is now much quicker than the bus/train. The roads will always be full – you can’t fix congestion in a successful city.

          You show the same frustration as Betamax users had when everyone kept using VHS. Human nature can be a bitch.

      1. Exactly. From being there, the main problem with Saigon/Ho Chi Minh is the lack of separated public transport. Same as it used to be for Bangkok until the MRT was built.

        You are not in congestion, you are congestion. Doesn’t matter if you are a wide vehicle or not.

  8. I find the results hard to believe….. The researches must have excluded the 25% less of human factor in their inputs. e.g. The driver that falls asleep at the intersection, is preoccupied and does not see the cars in front move, accelerates into the car in front of them, stalls the car, struggles to put the car into gear, decides to exit the back of the queue and push into the middle of the queue, runs the red light into another car, decides to stop and talk to a pedestrian, stops to pick up/drop off a pedestrian, slows down to look at a street address/signs/people/accident, remains stationary to break up the arguing kids/pets etc…

  9. As you say, “traffic generally moves at the pace of the slowest vehicles so all it takes is one slow driver or driverless vehicle and many others will be slowed down too”. In Wellington cyclists cause congestion because on our narrow streets cars and buses often can’t pass cyclists and are compelled to travel at the cyclist’s speed. Does the simulation take account of cyclists, pedestrians and roadworks, which all slow traffic?

    1. Hey Pam. It may be semantics, but cyclists in Wellington do not CAUSE congestion on “our narrow streets”. Those streets are rarely so narrow that vehicles cannot pass a cyclists safely, especially on bus routes.

      The real cause of the conflict is PARKED CARS, which can significantly reduce the effective width of an already narrow street, or almost block it altogether if parking happens on both sides. A street that presents no problem to cyclists and overtaking vehicles when free of parked cars becomes a major source of danger and frustration when it is thus obstructed.

      This is NOT the fault of cyclists, and while I can understand the point of your comment in illustrating the ‘slow-vehicle’ principle, it is not helpful to suggest that cyclists are the main cause of this in Wellington. Too many people unthinkingly blame cyclists and remain blind to the REAL cause, which is loss of road-space due to (often free) on-street parking on totally unsuitable streets.

      1. Parked cars, cyclists…one is moving, one is stationary, but both cause loss of road space. My comment that traffic is reduced to the speed of the slowest vehicle (i.e. the cyclist) is still valid.

        1. Your idea that a cyclist on the road causes a ‘loss’ of road space is interesting one. Surely they have a valid right to use the road? So the road space isn’t lost to them?

        2. My observation was an agreement with the original post that traffic has to move at the pace of the slowest vehicle, which is often a cyclist. I believe that cyclists contribute to road congestion because other vehicles often cannot pass them on our narrow roads. Dave B said that loss of road space is caused by parked cars. In response to Dave, I say that yes, parked cars, reduce the effective width of narrow streets, but so do cyclists. Cyclists probably more so – at least a parked car is not going to do anything unusual, whereas extra caution is needed when travelling by motorised vehicle behind a moving cyclist because one cannot predict whether the cyclist will wobble, swerve or move to the right. Obviously cyclists have a right to travel on the road, but they do effectively narrow the road width available for larger vehicles, and thereby hold up other traffic a lot.

        3. Pam. My point was that in terms of cyclists holding up traffic:-

          – narrow roads are not to blame, as these are generally wide enough to accommodate motor-traffic and cyclists WHEN NOT OBSTRUCTED.
          – cyclists are not to blame, as they are narrow enough not to conflict with motor-traffic on roads – even narrow ones, WHEN NOT OBSTRUCTED.
          – problems generally only arise when parked vehicles create what is effectively a ‘semi-permanent obstruction’ on the road.

          What is the primary purpose of roads? Is it to permit the movement of traffic (including cyclists)? Or is it to provide semi-permanent vehicle-storage? And if these two purposes conflict and create a hazard (which they often do even without cyclists present), then which objective should prevail? If roads need to be widened to safely accommodate both, then who should pay? Or who should subsidise who? These are pertinent questions which all-too-often get shrugged off. On-street parking is of course prohibited in some situations such as on certain arterial roads or on particularly narrow residential streets, – i.e. places where traffic authorities finally concede that action against inappropriate parking must be taken. So recognition is clearly there, that parked vehicles can be an unacceptable problem. However working against this is the popular but misguided notion that parking anywhere is a right, regardless of the hazards or costs it may impose.

          And in terms of road-space occupied, a cyclist will often be sandwiched in as little as 1m of space between the kerb and an overtaking vehicle (effective separation about 0.5m), though at normal ‘around-town’ driving-speeds this seems to work most of the time (even though safety guidelines recommend a minimum separation of 1.5m!). Cyclists randomly wobbling or swerving to the right is unusual. It has to be, as cyclists need to be prepared for the many drivers who allow them very little wiggle-room. However it is not true that “a parked car is not going to do anything unusual”, as any cyclist who has been knocked off by a door flying open will attest. Cyclists passing parked cars have to give them a wide berth to allow for this very real possibility, and this of course forces them even further out into the path of motor traffic.
          So in practice, cyclists themselves do not take up much space, and unlike parked cars they do not tie up this space for the duration.

          But a car, say 1.8m wide (more for an SUV), parked 0.2m out from the kerb, will fully block out at least 2m of road-width. Add to this 1.0m of separation for the “door width rule” between stationary and passing vehicles and that’s 3.0m effectively gone. And on a road where multiple vehicles remain regularly parked, this reduction in road-width is effectively permanent. Note that if roadworks or some other activity were to cause such an obstruction, then temporary traffic management measures would be mandatory!

      2. Yet another reason to support narrow commuting vehicles is that when they park to the curb they don’t open their doors to the street/bike paths and thus eliminate the possibility of bikes crashing into their car doors. Narrow commuter vehicles driving and parking and automated driving make the roads safer for bicycles.

  10. Actually, the best person to be would be the very last person with a real drive-your-own car, scooting between all the auto-bot drone cars, which will be computer controlled to avoid an accident, and so they will shuffle to make a space for you if you barge into the queue….

  11. This doesn’t pass the sniff test.

    The advantages of self-driving cars are great for efficiency:
    – Better reaction times to green lights
    – Closer following distances
    – No more driver inattention is queue situations
    – Uniform speed thus avoiding wave congestion

    I find it hard to reconcile this piece

    1. Most of those points only apply if every single car on the road is self-driving. I think the point Matt’s trying to make here is that spurning PT in favour of a fabulous new technology that might not deliver is a dud move.

    2. One has to consider that if driverless cars replace public transport then there will be more cars on the roads. The advantages of driverless cars will have to prove to be significant enough to still reduce congestion despite the large increase in the number of cars.

      And for those who say that car sharing will compensate for this I say no. While there will probably be a lot of car sharing. one of the reasons why many people prefer to drive over public transport is to enjoy a private journey. So there will still be a large number of single occupant vehicles on the road.

  12. I am amused by the backward facing seats. They are unpopular on trains and supposedly also on planes where they would theoretically be safer. People do seem to want to ‘drive’ even when they aren’t the driver.

  13. They haven’t considered that the intersection model they look at the moment may be able to be modified. Rather than one stream of traffic at a time, with a driverless system it could be possible to allow two perpendicular flows of traffic to flow and to pass safely between each other. Not as crazy as it sounds when you take driver error out of the equation..

Leave a Reply

Your email address will not be published. Required fields are marked *