Thursday, September 11, 2014

Driverless Cars and Uneasy Riders

I had forgotten we've been having this discussion for over three years.
Tyler Cowen has a piece in the New York Times on how regulation inhibits innovation in transportation using the example of driverless cars. I'm not sure he's made his general case (that's the subject for an upcoming post), but his specific case is particularly problematic.

In case you haven't been following this story, Google has been getting a lot of press for its experiments with self-driving cars, especially after statements like this from Stanford professor Sebastian Thrun:
"Think about the car as a medium of mass transit: So, what if our highway-train of the future meant you go on the highway, and there's a train of very close-driving cars with very low wind drag, fantastic capacity, is twice as efficient as possible as today, and so there is no congestion anymore?"
Cowen is clearly thinking along the same lines:
Furthermore, computer-driven cars could allow for tighter packing of vehicles on the road, which would speed traffic times and allow a given road or city to handle more cars. Trips to transport goods might dispense with drivers altogether, and rental cars could routinely pick up customers. And if you worry about the environmental consequences of packing our roads with cars, since we can’t do without them entirely, we still can make those we use as efficient — and as green — as possible.
Putting aside the question of the magnitude of these savings in time, road capacity and fuel effeiciency (which, given the level of technology we're talking about here, aren't that great), where exactly are these savings coming from?

Some can certainly be attributed to more optimal decision-making and near instantaneous reaction time, but that's not where the real pay-off is. To get the big savings, you need communication and cooperation. Your ideal driving strategy needs to take into account the destination, capabilities and strategies of all the vehicles around you. Every car on the road has got be talking with every other car on the road, all using the same language and rules of the road, to get anything near optimal results.

Throw just one vehicle that's not communicating (either because it has a human driver or because its communication system is down or is incompatible) into the mix and suddenly every other vehicle nearby will have to allow for unexpected acceleration and lane changes. Will driverless cars be able to deal with the challenge? Sure, but they will not be able to able to do it while achieving the results Thrun describes.

A large number of driverless cars might improve speed and congestion slightly, but getting to the packed, efficient roads that Cowen mentions would mean draconian regulations requiring highly specific attributes for all vehicles driving on a major freeway. The manufacture and modification of vehicles would have to be tightly controlled. Motorcycles would almost certainly have to be banned from major roads. Severe limits would have to be put on when a car or truck could be driven manually.
Based on the conversation that followed that post (check out the comment section), I should probably add that much of the benefit described by Thrun and Cowen could achieved by making special lanes and sections of road driverless-only. One the whole though, I stand by the point that much of what we've been promised (speed, fuel efficiency, road capacity) require an all driverless group of cars working together.

One point I made in passing could probably use more elaboration. Motorcycles are small, accident prone vehicles. They can accelerate very quickly, they often behave erratically, and they tend to function under a somewhat looser set of traffic laws. Their small size and low cost make them more difficult to regulate. And finally, as far as I can tell, there is no serious plan to introduce fully autonomous versions. If you want to get close to the level of performance Thrun promises, you do not want motorcycles on the road.

It's hard to see this not becoming a typical convenience-of-the-many argument for regulation. As autonomous vehicles become more common, it is pretty much inevitable that, while overall accidents and traffic jams will go down, those that still occur will be disproportionately caused by vehicles that don't lend themselves to autonomous control or those which routinely have to do things that are difficult to explain to a computer. The first group would include motorcycles and classic/antique cars. The second would include real pick-ups* or SUVs that actually leave the pavement. I would hate to see those vehicles forced off most major roads but that would seem to be the likely outcome.

* Those that do real work. Us country boys take this seriously.


  1. There are plenty of driverless car pieces moving forward all the time and there is no regulatory delay. More and more carmakers include rear cameras and more sensors that enable "driver assist" or other marketing words to describe the car doing something when you don't. It stops if it senses an object. It brakes if something gets too close.

    There's a cluster of these companies in Israel. Look up Mobileye, for example.

    1. Jonathan,

      Actually, I think those pieces are where the real story is. The press is fixated on both Google and driverless cars but that represents a small part of the total research going on out there. If I had to, I'd wager that while autonomous driving features will greatly affect our transportation over the next decade, the technology being developed by companies like Toyota and GM (and possibly the small players you mention) will play a larger role than what's coming out of Google.

  2. You're right to bring some perspective to this conversation. It's simply stunning that all these talking heads haven't bothered to look up the OR literature; OR researchers have studied driverless vehicles for many decades, and also studied lots of related transportation problems.
    The idea that having driverless lanes magically solves the problems you stated is simply wrong. If the lanes are on the left, the driverless cars have to cut through the lanes with human drivers in order to exit a highway, and this merging occurs at every exit. If the lanes are on the right, the human drivers have to maneuver their cars right through the driverless lanes, which disrupts the flow of the driverless traffic.
    I often wonder if those commentators spent more than 10 seconds thinking about what they are proposing.