wandership.ca

Driving with Google
written on 2005-02-14 at 12:00

After reading the Slashdot story about Google Maps, I experimented a bit with the service. I'm quite impressed with the responsiveness and interactivity possible by the combination of server-side and client-side code; I'm surprised it took someone this long to get it right.¹

As Google Maps is not advertised as complete, the world for the moment consists entirely of North America. All the U.S. states are labelled, but Canada and Mexico, at least from the top level, are vast expanses of white emptiness. (Of course there are some very obvious jokes that can be made about this, but I don't think that's quite fair to Google.)

For this reason, I expected that no information about Canada would be present, and that zooming in would reveal nothing outside the U.S. border: this is the case with Mexico. Canada, however, is actually there; it's not marked as well as the U.S. (province abbreviations, highway icons, etc.), but the streets are visible once you zoom in far enough.

After some playing with this, I decided to test Google'sg driving directions. At first I tried only in-city locations, which were all quite reasonable. Inter-city transit is a more difficult problem, and Google fared less well here.

My first attempt was the most dramatic and inexplicable failure. I requested a map from my current home of Waterloo, Ontario to downtown Hamilton, a distance which Mapquest reports as 43.98 miles with an expected time of 65 minutes, which is reasonable. For some reason, Google Maps suggests taking the 401 east to Toronto: this isn't ridiculous in itself, but rather than leaving the 401 for Highway 6 or even taking the 403 to Hamilton from Toronto, the instructions then suggest taking Highway 400 North through Barrie, almost to Bracebridge. Oddly, they don't then tell you to go south again, but they do tell you how to get to the 403 as though you were still in Toronto. From this point on they are reasonable, if you happened to be going to Hamilton from Toronto. The total distance is 300 miles and expected time 335 minutes, almost seven times the Maqquest distance.

A bit more experimentation seems to show that Google has a marked preference for the 400-series highways, perhaps indicating that their metric for assigning weights to highways does not yet take into account non-major highways. For example, a route from Cambridge to Hamilton suggests taking the 401 west all the way to the 403 junction (near Woodstock), then east to Hamilton, when Highway 8 directly links the two cities. More dramatically, the instructions from Milton to Hamilton suggest the same suboptimal route.

There's a peculiar feature of the service that I've also seen from Mapquest occasionally. An obvious thing I would desire from route-plotting software is this: if the computed optimal route from A to C passes through B, then I would expect this to be equivalent to the computed optimal route from A to B plus that from B to C.

In isolated cases Google Maps appears not to do this. For instance, if I request directions from my hometown of Dundas, Ontario to Winnipeg, the suggested route passes through Chicago, but first goes around the eastern end of Lake Ontario, through New York State. Yet the directions from Dundas to Chicago go the more reasonable way, i.e. west through Detroit.

If you look closely you'll see that the route to Winnipeg passes slightly west of Chicago (i.e. doesn't leave the highway), so perhaps that slight difference is enough to force the change.

In any case, this is just playing: it's in beta, it advertises itself as U.S.-only anyway, and there is already Mapquest. And even the suggested detour to Bracebridge is not as dramatic as this example from Microsoft's MapPoint service.

1. "someone" referring to web developers generally, rather than Google specifically.