Yes, I am taking the pro side here on autonomous vehicles

In a comment on my recent relatively positive post on autonomous vehicles, Joseph points us to a Megan McArdle article that takes a different view.

While it is good to see that the conventional wisdom is starting to acknowledge some of limitations with driverless cars, I still have quite a few problems with the piece. McArdle makes some good points about the labor implications, but she does not seem to have a strong grasp of the technological or the implementation issues involved with using autonomous vehicles for long-haul trucking. We can get back to implementation later; for now let's talk about tech.

Here's McArdle:

You hear a lot about how Google cars have driven an amazing number of miles without accidents. You hear less, however, about how they have achieved this feat: by 3-D mapping every inch of those roads so that the car has a database of every stationary object, from traffic lights to guardrails. That allows the car to devote its processing power to analyzing the movement of objects that aren't in its database.

Such mapping is incredibly labor intensive, which is why, according to Lee Gomes, those amazing mile counts that Google's driverless cars are racking up "are the same few thousand mapped miles, driven over and over again." Most of them are near Google's headquarters in Mountain View, a place that gets only 15 inches of rain a year and never has snow or ice -- three common weather hazards that long-haul truckers must frequently contend with.

Just getting Google's technology to a point where we could have self-driving trucks would require mapping every inch of the nation's more than 164,000 miles worth of highways. But then what do you do with the truck? You're probably going to have to map some of the roads that connect to those highways too. And then constantly remap them, because things change all the time. You'll also have to teach the computer system what to do in a blinding snowstorm on Wolf Creek Pass. As we wrote in January, "The technology giant doesn’t intend to offer a self-driving car to areas where it snows in the near term."

McArdle makes a couple of common mistakes: assuming that, because Google dominates the coverage of driverless cars, it also dominates the research (which we'll get to later); and assuming that what is difficult for humans is difficult for robots and vice versa.

Rain and snow are problematic for us humans both because they can limit visibility and because they tend to create very complex physics problems that have to be solved in a fraction of a second. Bad weather visibility is much less of an issue with autonomous vehicles* than it is with human drivers while classical physics problems are the sort of thing that roboticists are very good at.

Along similar lines, McArdle observes [emphasis added] "But it seems like getting from there to fully automated trucks--necessarily huge, heavy, and  capable of horrific damage, with handling capabilities that change depending on the load, and a stopping distance almost twice that of a car at high speeds, will probably take a while." Yes, this will take a while, but not for the reasons McArdle imagines. Load effects and long stopping distance do make truck driving much more difficult for humans, but for computers they just represent simply another set of parameters. Furthermore, the biggest factor in real-life stopping distance is often reaction time, an area where computers have a distinct advantage.

Nor does the fair-weather testing complaint hold up. It is true that Google has largely limited its testing to clement conditions, but you certainly can't say the same for the upcoming Volvo test in, you know, Sweden.

Google's PR department has done a masterful job identifying the company with autonomous vehicles. This is not simply a matter of corporate ego. As I said earlier:

Google has a lot of reasons to want to be seen as a diversified, broadly innovative technology company, rather than as a very good one-trick pony cashing in on a monopoly (possibly two monopolies depending on how you want to count YouTube). A shiny reputation helps to keep stock prices high and regulators at bay.

It is enormously telling that McArdle cites Google ten times in her article while she doesn't mention Daimler by name and she never refers to Volvo at all.

* As far as I can tell, Daimler's prototype is doing its mapping independently in real time. While impressive, I'm sure the production models will share data and will also rely on existing maps.