by veen

I have a few feeds for autonomous vehicle related news. Sometimes I come across articles that I consider posting here, but ultimately don't because there are only a few interesting tidbits hidden behind useless blog filler.

This WSJ article is an archetypical example of that. But I think the few interesting points in it are worth telling / talking about, so instead of not sharing I'm going to walk through the article and annotate it with my input on the subject matter.

Why Cities Aren't Ready for the Driverless Car - Wall Street Journal

When it comes to adopting self-driving cars and trucks, the easiest part may well be building them. The far more difficult task will be maintaining our urban transportation infrastructures for autonomous vehicles to be functional, safe and practical.

Ah, finally someone asking the right questions, I naïvely thought. As someone from an urban planning background, I find this question one of the most interesting. Automated vehicles becoming popular in a few years can be a similar shift as from horses to cars. There's bound to be shifts in how people use it and interact with vehicles in an urban setting. If it's up to me I'll do my graduate thesis on something like this question next year:

What will cities have to do to get ready for the transition to the autonomous car?

As a basic assumption, we're talking about fully autonomous vehicles - so no driver's license needed, a possibly Uber-style service, providing mobility on demand. He doesn't make this explicit, but since he mentions later on that the cars can drive themselves to a parking spot, that's the assumption you need. I'm going to use the Google car for examples here.

For starters, they will have to maintain everything from complex intersections to lane markings to the specifications expected by vehicle software designers. Without a city’s commitment to certain standards, self-driving autos might freeze in place on streets lacking clear lane markings. Similarly, unmanned vehicles might proceed at speed through an intersection where a stop sign has been removed by college students or knocked down the night before by an impaired human.

Maintenance and complex intersections! Great points. Depending on how heavily autonomous vehicles rely on detailed logged maps or on-the-spot processing, this can be a minor or major issue respectively. A sidenote here is that AV's that only use lane markings and signs for their 'road logic' are probably not good enough for most real world information. Looking at what other drivers and people do in any intersection is also key at understanding a place. I saw a demo by Daimler a while ago that would use cameras to track pedestrian movement. The car would model the direction that pedestrians were turning their head, since people who intend to cross the street (including jaywalkers) turn their head to look for oncoming cars. Not using information like that and sticking to the rules means your car will run over that pedestrian.

But that kind of thinking through the situations is not very prominent in this article, because he follows it up with this:

Traffic rules may be writ in stone, but the autonomous car or truck should also understand local practice. If in a particular city it is customary for trucks to double-park while making deliveries, will the driverless vehicle coming up behind a stopped truck think (in software terms) that it is at a stop light and wait there for the unseen light to change?

The first sentence is, again, a good and serious concern. How do you develop AI that can distinguish between local quirks and global behaviours? The example is, to put it lightly, confusing. It suggests the vehicle doesn't recognize a truck (not a small vehicle I'll have you know) but does recognize its tail lights as traffic lights. That's what I think he's suggesting - a confused vehicle that doesn't understand the simplest and most easily recognizable blocked road situation.

Next up, we have the 'someone didn't do their homework' part of the article:

And in cities where it is customary for human drivers to anticipate the red light turning to green by inching into the intersection prematurely, will the driverless automobile allow for the custom? [...] Driverless cars can be programmed to be aggressive or patient. But who gets to choose? The software developer? The owner? Or perhaps the local police department?

Hitting two birds with one article.. (Relevant part is in the 'More Agressive' paragraph)

But upon sensing the yellow when almost into the intersection, will the driverless car stop suddenly or will it speed up?

smh

Will the car be programmed to respond differently when it is carrying passengers, comparing in milliseconds the risk of both actions and choosing the one less likely to cause injury?

Why, exactly, wouldn't it be programmed to choose whatever causes the least injury? Also, injury to whom? (By the way: the whole trolley problem thing was somewhat addressed by Chris Urmson in his recent SXSW talk. He said that the car avoids hitting the most vulnerable users first.)

But the ultimate testing will necessarily take place in real cities under real traffic conditions that test tracks can hardly be expected to replicate fully.

You mean like almost all of the serious tests that are being done right now? Gosh, what a surprise. Next, the author switches gears to briefly discuss vehicle-to-vehicle (V2V) communication:

It also means that increasing and maintaining the wireless-bandwidth capacity used by cars in traffic to communicate will become a responsibility as important as smooth roads and clear signage. The wireless network will have to be pervasive and fast enough to allow for adequate response times to, say, jaywalking pedestrians or erratic cyclists.

Again, it has a good part and a bad part. The bad part is that he completely fails to explain his assumptions - yes, if V2V becomes the real deal and if most vehicles actually actively use it in urban settings and if these V2V communication networks (that already have a part of the bandwith spectrum reserved) aren't cleared good enough and if these cars don't spot suprise jaywalkers in time, then maintaining that network is critical to save the jaywalker that regular cars would probably hit.

These are issues that must be resolved before autonomous vehicles are turned loose on the asphalt.

No, not really. V2V is only a solution, not the solution. Case in point, again: the Google car. I had to do a double-take, but yes, this article is written in 2016. Four days ago, actually.

Proponents of driverless vehicles focus—sometimes too much—on the benefits the vehicles offer to both individuals and the communities in which those individuals live.

Good point. The debate doesn't always go into discussing environmental issues, into the likelihood of induced demand or on any other externalities.

Whether cities will even allow self-driving vehicles will depend mainly on public policy. Despite its popularity as a touring aid in some cities, the Segway—a self-balancing electric scooter—has failed to find acceptance in the broad transportation market in part because there is little uniformity among communities regarding rules governing its use.

See? He does it again. He starts off with an interesting, debateable observation and then writes something inane. If you can't tell from my writing, I'm quite frustrated by articles like this. The interaction of autonomous vehicles with urban policy is very interesting. I can totally imagine some cities to heavily restrict or ban the vehicles in dense, complex areas, or countries limiting them to special roadways. That would've been a much more interesting point to discuss than to haul in the most useless transportation method ever invented:

In the postdigital age, drafters of municipal legislation may be looking at an extended legal battle to redefine what a driver is, and what that driver’s responsibilities and liabilities are.

Since when does municipal legislation define that? There are multiple national organizations working this one out, with the EU and the US working on legislation that allows driver-free vehicles on public roads. Note how that second, easily-Googled link is also from WSJ.

Better luck next time, professor.