Sunday, June 13, 2010

New GPS, or So lifelike!

We went on a road trip a couple of weeks ago (New York and Washington, D.C.) and, before we left, bought a new GPS. There is a point in here about technical writing but you'll have to persevere to get to it.

Our old GPS worked fine but the maps were out of date. On our road trip to Texas last January we found ourselves navigating through turnpikes that didn't appear on the map and taking off-ramps that didn't appear on the device (while driving past "phantom" ramps that were on the map but no longer existed on our plane of existence). We spent 15 or 20 miles driving along a road that didn't exist (the device showed empty field) while it constantly urged us to go to a much poorer road about 500 yards to our right.

We could have upgraded the maps, of course, but it would have cost $100 so considering buying a new device was also an option. A new device was attractive for a bunch of reasons: The old device was big and clunky so it could only be used in the car; newer devices, being slimmer, can be used when walking around. If we got a new device, we could also pass our old device onto some friends who  needed to go to places they hadn't been to before. Looking at newer devices I realized that I could get a bigger screen (I now have a separate pair of reading glasses) and "lane guidance" (I was interested in seeing how that works).

And, besides, who doesn't like a new toy?

I also assumed that new devices would just be generally "better" (our existing device was about 6 years old). For instance, the old GPS had virtually no 'intertial navigation'--when it lost contact with the satellites, it lost its mind. We first noticed this when driving into New York at 10:00 at night: In the canyons formed by the building, the device lost satellite contact and alternated between spitting out random instructions and just giving up. We got into the habit of reviewing the lists of turns before we drove into an environment where we thought we would lose our signal.


Taking advantage of a sale we got a new device for about $170.00. The inertial navigation is much better: The new device guided us through the Holland tunnel, for instance, long after we went underground. The bigger screen contains more information while being easier to read (for instance, it tells me whether my next turn is a left or right turn much further in advance than the old device did). The lane guidance is useful but, sadly, only available for US locations. The reaction time isn't as good as the old device: For instance, when we missed a turn with the old device, it started figuring out a new route almost immediately; the new device takes about 5 seconds to recognize the problem. On the other hand, the new device takes into account more information than the old device did. The old device would give you the same route between two places every time you requested the route; the new device gives you different routes at different times of the day because it takes into account shifting traffic patterns: during rush hour, a route that's longer may take less time because travel times are shorter.

But what really impressed me was the improvement in the text to speech feature (which is why this discussion turns up in this blog). The new device does a much better job of communicating meaning rather than direction and it does that by being both more specific and more colloquial. For instance, when coming up an on-ramp to a highway, the new device doesn't use the generic "bear right" or "get into the right lane"; instead, it says what a person would say "Get onto the freeway" or "Merge onto the freeway". When you're approaching a T intersection where your road terminates, it doesn't say "Turn right" or "Turn left" (which are perfectly accurate)--it says what a person would say "When the road ends, turn left." One of the wonderful things about this last instruction is that it allows you to ignore any other turns between you and the T intersection: because the instruction is more specific, you know that the only intersection that matters is the one at the end of the road. Obviously, a lot more of the context ("the scenario") is being taken into account when generating text which results in better directions. They probably had a technical writer involved.

Addendum: The subtitle for this column, by the way, comes from an episode of Criminal Minds. One of the running gags on the show deals with the team's resident genius, Spencer Reid. In one particular episode, after commenting that the usual procedure in working out the result for a problem would have been to create a program, Spencer comments that he found it easier just to plug in his own values. One of the other team members reaches out, strokes his face, and comments "So lifelike." Part of the joke (for me) is that, rather than process all possible values for a problem, it is often easier to just plug in a couple of exploratory values, see what result you get, and then narrow your search to the most likely inputs. Often, in fact, you know enough about the problem to know that you don't need to look at "all possible values"--that the real world constraints mean that there are only a few values that are worth testing.

Reading or read

2 comments:

Anonymous said...

I see that you're reading The T.E. Lawrence Poems by Gwendolyn MacEwen. You'll probably also be interested in Mighty Oaks, available on Amazon.com, which deals, in part, with Gwen's brief marriage to Milton Acorn.
http://www.amazon.com/Mighty-Oaks-Lorne-S-Jones/dp/145054004X

Peter Vogel said...

You're right, I bet I would. Didn't even know the book existed.