Saturday, July 5, 2014

Driverless car: transport revolution? End of transit? A dud?

Reading Charlie Gardner's recent post on driverless cars has motivated me to write about my own opinion on the subject. It's a topic that keeps coming up again and again, a lot of people are very bullish on automated cars and making all kinds of predictions. That they will eliminate the need for transit, that they will increase road capacity and speed in major ways, that they will allow suburbs to sprawl more/or eliminate the need for parking and allow instead denser cities.

First of all, I'm still skeptical of the technology. For one thing, it will likely add major costs to cars, since computers work on a "garbage-in-garbage-out" basis, meaning their outputs are only good as their inputs, so they will require high-quality and well-maintained sensors. Quality and durability are very expensive and likely to remain so. Google's self-driving car relies on 100 000$ sensors that are constantly being tested and maintained. As to the ideals of increased capacity and speed, that's only possible in roads where there are no other users, no cyclist, no pedestrian and no human-driven vehicles (cars, buses, etc...). Possibly doable on highways, but not on city streets.

Even if we assume that all goes well on the technological front, it doesn't correct the main problems of cars. Which are:
  1. They take a lot of space in the street when on the move, much more than any other mode
  2. They need to be parked somewhere when not in use, and they take a lot of space there too
There are essentially two ways I see them being implemented. Let's see what the possible impacts of each would be by thinking rationally about it.

Self-driving as an option of private cars

In this case, people keep owning private cars, the self-driving ability is merely an option on the cars. This actually doesn't change things much, if only maybe increase drivers' tolerance to long trips as they can do other things than drive. People still need residential parking to store their cars, and would likely still need just as much parking at their destination (work, stores, etc...). So using a car doesn't cost any less money, people don't need any less parking, cars still occupy just as much space as before, there is just nothing revolutionary here.

Theoretically, they could dispense with parking by having their cars drive back home or at holding spots outside of dense areas. However, this would be nothing short of disastrous for the environment, for the economy and for congestion. What I mean is that just because cars are driving around empty, doesn't mean they use less fuel, and doesn't make their presence unfelt by other cars on the road.

So taking an extreme example, let's suppose a suburban resident driving to the city every day with a car with a fuel consumption of 10 L/100 km (around 24 MPG), his trip is 20 km one way (12 miles). The way he's doing it now, he's driving 20 km in the morning, leaving his car parked at his workplace, then 20 km in the evening, for a total of 40 km each day (24 miles). Overall, he uses 4 liters of fuel every day (around 1 gallon). What happens if he sends his car back home? Well, his car is actually driving to work once in the morning, then driving back home, then driving back to work again in the evening, and finally one last time back home. So instead of being driven 40 km each day (24 miles), the car drives itself 80 km (48 miles) and burns up 2 gallons of fuel instead of 1.

This means twice the pollution, twice the cost of running the car, it's a lot of wasted energy, that's what it is. Not only that, but if you think current traffic is bad, just wait and see how it looks like if peak hour traffic become symmetrical. Currently, traffic looks a bit like this:
What traffic looks like on most streets, blue is toward the city, red is towards suburbs
Which means that roads are congested in one direction, but not both at the same time. But if you send cars back empty (what is called in transit "dead heading"), then you just congest roads in both directions at once. What's the big deal? Well, we engineers dread a particular situation called "interblocking". Interblocking is when congestion occurs in two directions at once and the cars waiting in one direction block the cars going in the other direction, especially turning maneuvers, which can reduce tremendously road capacity and make a bad situation much, much worse.

So, all in all, if people keep using their own cars, I don't see things changing much, and if self-driving is an expensive option, many people will just not bother with it.

Driverless taxis

This is the second option. Essentially, driverless cars are used as taxis without drivers, going to meet clients, then dropping them off and going to meet the next ones. Cars are now shared rather than private vehicles, yet still offer many advantages of private vehicles for those who like them, like privacy. If this became the default mode of transport, it would effectively be a revolution, primarily in term of parking spaces, which would become much less needed, and the number of cars would plummet, but not as much as people think. Some studies showed driverless cars replacing as many as 15 cars (if I remember correctly) based on actual trip data... but this was really the best case scenario. Meaning, an urban environment, with trips in all directions which minimized dead heading (Cars traveling empty) and with very few driverless cars.

Adding more driverless cars makes them less "effective" as their likelihood of idling and not being in use increases quickly. Furthermore, the efficiency of the driverless car totally collapses once one takes into consideration suburban commutes, which are very highly directional. If traffic flow in one direction is 6 000 cars per hour, but only 2 000 in the other, it means that 67% of cars will dead head at least half of the time, and create the risk of interblocking as they make the car flow symmetrical.
Traffic flow in a day per direction, the graph to the left is current flows, the graph to the right is the result of driverless taxis replacing private vehicles
This is the sum of flows in both direction, in green the current situation, in black the driverless taxi hypothesis
In the scenario pictured, which is typical of most roads in metropolitan areas, the number of car trips increases by 25% daily, and up to 50% during peak hours. This means total fuel consumption would increase by 25%. Imagine if this happened all over the world, what it would do for fuel prices.

The number of driverless cars needed to respond to the demand would be equal to the highest demand during the day. If in the worst time of the day, there is 1 million cars on the streets and roads at the same time, then you need 1 million cars. In the scenario I illustrated, I'd estimate that the needed number of cars would fall by about 75%, each driverless car would replace on average 4 cars, and maybe even less. These cars would also need to park somewhere when they are not currently being used, to avoid having them fill out streets, they would probably require holding areas outside of cities, and trips to and from these holding areas would add to the dead heading distance they would have to travel.

Okay, so this scenario leads to massive bidirectional congestion and massive increase in energy/fuel consumption. But is it a plausible scenario? Would people actually use driverless taxis to the exception of everything else? Would they kill both fixed route transit and private vehicles?

In a nutshell, no.

First, if they replaced transit, they would increase even more the number of cars on the roads, further increasing congestion and fuel consumption. Driverless cars do not solve the spatial problem of cars, they still require very wide roads, wasting valuable urban space.

Second, with all the dead heading they're going to do, it means that every time one uses one of them, they're going to need to pay for all the dead miles the taxis have to travel. Currently, the CAA (the Canadian version of the AAA) estimates that a car costs about 50 cents per km to use (80 cents per mile), without counting parking (which if counted would easily add another 20-30 cents). Much of the cost is depreciation, so let's say 40 cents per kilometer (64 cents per mile) assuming a very used driverless taxi, which has increased maintenance and acquisition costs because of the sensitive sensors. Now let's assume a reasonable amount of 50% of deadheading, that means that every mile traveled inside a driverless taxi has to pay for two miles. So now the cost is 80 cents per kilometer (1,28$ per mile). If the taxi companies are private, add 10% profit, we're up to nearly 90 cents per kilometer (1,40$ per mile). That's still much cheaper than current taxis, I've taken a taxi recently that cost me about 35$ to go 15 kilometers, so 2,33$ per kilometer (3,60$ per mile), but it's much more expensive than using one's own car or taking transit. For a more in-depth analysis of what driverless taxis might cost click here.

Therefore, the value proposition of these driverless taxis isn't that great, not unless many people take place in one. Transit operating costs vary between about 10 cents per kilometer per passenger (15 cents per mile) for heavily used subways (like in Tokyo, subways are profitable and can charge 3$ to go 25 km, but North American costs are probably more around 20-30 cents per passenger-kilometer) to 60-80 cents per passenger-kilometer for local buses (1,00-1,30$ per mile), with LRT and BRT falling somewhere in the middle. Subways still remain an immensely more efficient way to carry a lot of people around very fast, and buses aren't more expensive than driverless taxis and take much less space, but are much slower in general. Buses can also be made cheaper by making a better use of them. I know in Japan some local bus companies are often profitable while demanding the equivalent of 50 cents per kilometer (80 cents per mile) from users, and that's without articulated buses.

Also, one must consider the fact that if cars can be made driverless, so can buses. And big gains can be expected from this. If I go by the STM's budget, about 40% of bus operating costs comes from the drivers' wages. So remove drivers and buses' costs can fall to about 50 cents per kilometer (80 cents per mile) or less. Fixed route transit would thus be more efficient, at least in urban areas. In suburban areas, I could see driverless taxis replace the infrequent, inefficient suburban buses, or maybe some kind of driverless van, an hybrid between buses and taxis, with a fixed route but coming on call rather than being on a schedule.

Yes, I hear people whine "you're using passenger-kilometer for transit and vehicle-kilometer for cars!". Yes, I am, because most people travel alone by car, even in places where carpooling is more common, it's still a minority activity, because it is incredibly inflexible, you need to come from roughly the same origin and going to the same destination as someone else, at the same time. And I assume that people will not want to be with strangers inside a taxi without a driver. Alone, we feel safe. In a crowd, we feel safe. But there is a middle, when we're alone with one or two strangers in a closed space, where we really, really do not feel safe.

Worse, the value proposition of driverless taxis is terrible compared to private vehicles. Not only is the cost from private vehicles about 40% less, but private vehicles have the advantage of having most of the cost be fixed costs, with only limited marginal costs. In other words, once you've bought the car, paid the insurance, paid the regular maintenance, and all those fees that are not directly connected to your use of it... cars seem cheap to run, you just need to gas it up once in a while. Meanwhile, the driverless taxi would be a full-on marginal cost. Travel 10 miles, pay 14$. Ouch, right?

So that means that people who want to keep traveling by private vehicle are likely to not use the taxis but buy their own car instead. So this scenario of shared driverless taxis as the dominant mode of transport is just not that plausible to me. The economics are not there. In dense areas, transit is, or can be, much more economical and just as good in some cases (subways with development around stations so most destinations are within walking distances of them). In less dense areas, people who commonly use private cars to get around are much better off still using private vehicles rather than driverless taxis.

In conclusion

So, overall I do not expect self-driving cars, even when the technology is mature, to significantly change how people get around. People who use private vehicles to get around will still use private vehicles to get around, they just will be less likely to drive them themselves. Meanwhile, people who use efficient mass transit in urban areas, who are the majority of transit users in North America, are likely to keep using mass transit. Driverless taxis can make taxis much more affordable, but taxis are likely to remain a marginal transport mode, though they may make owning a car for occasional trips less worthwhile and increase car-less households in cities, even if these households would still mostly rely on transit and active modes of transport.

Transit in less dense areas may be more affected, especially if the technology filters to transit buses, as it would make them much more economical to run, and hybrid bus-taxis on fixed routes could really open up transit in less dense areas.


  1. Thanks for your very interesting post, much more realistic that all the fuzzy things we read about autonomous vehicle impact. Anyhow, the impact could be a bit higher that what you proposed if there are solutions less costly that what has been developed by Google, based on simple camera for example instead of Lidar. Of course today those solutions are less mature than Lidar, and speed limits are lower. Anyhow a solution like VIPA developed by Ligier has already been tested in real life in an hospital parking lot and soon on other "last mile" situations. Of course there are still many tests to be done and problem to be solved before it could a large service (of shared vehicle... because, as you mentioned, it has to be shared). Your realistic vision of transport and urbanism would be very interesting to determine the potential of this kind of solution.

  2. I'm glad I was able to provide some motivation to write about the topic, but you've done a much more thorough job. I would only add that, as alluded to in my own take on this, even though the "owned" car is cheaper to operate, per mile, than the driverless taxi, this may understate the difference, since the owned vehicle independently offers, or can offer, other advantages that a driverless taxi cannot. To name just a few: proximity of access, immediate convenience, storage use, carseat availability, feature customization, and so on. These are things that are essentially perceived as free for the suburban vehicle owner (they are fixed costs -- e.g. one's parking space at home, paid for as part of the purchase price of the residence, or the cost of an infant carseat), but which add significant value to the vehicle. The value of these advantages may vary greatly person to person, but should tip the scales even further in the direction of ownership.

    1. Indeed, there are still many advantages to private vehicles, people can have a vehicle they can choose tailor-made for their needs and even play around with and trash a little.

      Proximity access is an interesting point to bring up, many of the AV promoters seem to assume that an AV will always be right outside people's doors and ready to go at a press of a button, but that is quite impossible, especially in lower density suburbs where you can't afford to have an AV on every street. Well, you can, but once this vehicle has been called, another needs to get mobilized to replace it, coming from some other holding spot. So there may be quite a lot of wait time sometimes, 5 to 10 minutes easily if there is a random "burst" of demand and all cars around are already being used or moving to pick up customers. In order to reduce wait times, they would need to buy a lot more of them than they actually need to have excess capacity laying around, because travel trips can't be perfectly predicted. Which reduces the advantage of sharing vehicles.

      Also, your message reminds me that I forgot to point out one particular economic aspect of driverless taxis: the cost of their parking. I pointed out one reason cars look relatively cheap is that parking costs are not included in their costs by the CAA or the AAA, but parking is more likely to be paid directly by AV managers, just like transit companies have to pay for garages. In the Driverless Taxi Hypothesis, we have to put vehicles somewhere when they're not being used. Some will be on the streets, just like taxis, and just like taxis, they may have to pay for the privilege. There's a reason why most cities limit the number of taxis or put high prices on taxi licenses, it's because it comes with the right to use specific parking that comes at a cost to the city. Otherwise, taxis could theoretically submerge street parking and create congestion if they're too numerous, or not find parking and have to cruise around aimlessly. That's not a problem the AVs will solve, but instead make worse, unless they cruise around when unoccupied, then, hello! Dead miles!

      Some parking spots will also be needed, and they are more likely to be paid directly by whoever runs the AVs. If they want some holding spots in cities, it's going to cost a pretty penny. If they build them where land is cheap far from cities, then the number of dead miles will increase. So they're caught between a rock and a hard place: increase parking costs significantly or increase dead miles. Either way, it drives costs up.

      The more I think of the logistics, the more it sounds really implausible to implement efficiently. Someone on an article compared the AV dream to the PRT idea, the idea of individual small transit vehicles on guided rails that could be programmed for point-to-point transport. That's another idea that works extremely well on a small scale, but has way too little capacity for actual real-world implementation. The capital costs are too high for low-density areas, and capacity is way too low for high-density areas, meaning it only works well for very specific situations like airport shuttles.

  3. This comment has been removed by the author.

  4. Hi,
    Thanks for sharing this information.


    Car Transport