Saturday, September 18, 2010

On GPS navigators, traffic, and correlations

Our old TomTom GPS navigation device recently bit the dust, so we bought a new one, which came with free lifetime traffic information.  We don’t get traffic information around home—nor do we need it—but it’s a nice bonus when we travel to the big cities.

Generally speaking, traffic data is available for freeways, but not surface streets.  The TomTom unit has statistic knowledge of speeds on the surface streets, called “IQ Routes” (e.g. on Monday mornings, traffic on Foo Street averages 25 mph).

When there’s traffic on the freeway, TomTom will announce the delays (“12 minutes”) and in some cases suggest an alternate route (“5 minutes faster”).  The alternate route usually involves surface streets in lieu of the freeway.  It seems to me that TomTom must be comparing the actual freeway speed with the typical surface street speed (because it has no real-time data for the surface street). But does it understand correlations?  If the freeways are unusually slow, then presumably more people than normal would have already opted for the surface streets on the basis of traffic reports.

Second, what happens when most people have GPS navigators with real-time traffic, and follow the units’ advice?  Thousands of drivers may swamp a side street, trying to avoid a jam on the highway.  When that happens, we have a scenario more like finance than physics: the predictions of the model influence reality.  If you think you have a profit-making (time-saving) opportunity, it disappears in response to the advice.

The solution may be for the GPS navigators to introduce randomness into the suggested routes, dispersing their drivers onto several alternate routes.

I’ve also thought that TomTom should recommend randomized routes as a matter of course (when there are multiple almost-equally-good options) as a means of improving the IQ Routes database.

No comments:

Post a Comment