Flutterby™! : You can't be Sirius 2009-06-08 14:04:12.937697+00

You can't be Sirius

2009-06-08 14:04:12.937697+00 by Dan Lyke 9 comments

Charlene and I took her car up to Grass Valley on Friday night for Roz's graduation, drove back on Saturday, then I ended up driving up to Santa Rosa before we headed back down to Larkspur for the ferry into Bookstock. A lot of driving the new car.

The car has a trial period for Sirius radio, so while we were tooling around we were also sampling the offerings. The thing that struck me was how bad the codec for that is. Music mostly sounds okay, but the voice quality was so bad that the only way it was listenable was with the treble knob turned all the way down, and even there there was clipping and hiss, so bad that I almost didn't recognize Peter Sagal's voice. Went back over to the FM version and it was fine. The free trial is not terribly enticing.

And while we were up in the foothills, both the TomTom and the iPhone tried to send us down Lost Lake Road. I'm not seeing how to turn off the road layer on Google Maps there, but if you look at the plain satellite images there's no road where the map line indicates there should be. I think it's that other cut over to the left, but at the second set of puddles of indeterminate depth we decided to back out and try another route.

Be wary of those GPS suggestions...

Every time I try to use my gps to get to or from my house, it directs me to drive through someone's horse field in my neighborhood. I can't possibly see that there was ever a road there, and have no idea why it would show up on any map. Google maps finally corrected it some time last year, but even my new Garmin streets still has it wrong.

Tom Tom's pretty good about user submitted updates.. I've only been routed into a berm of dirt once. My destination was on the other side of the berm, but the road was never finished. That route was not suggested the next time.

My parents have had XM for several years. They use it specifically for talk radio when they travel. Maybe it's better. Of course, my parents have always claimed they they could never hear the difference between a cheap stereo and a good one.

And always, always question nav systems. Prior to them becoming so wide spread, there was a lengthy, on-going debate in the academic literature about how to manage liability and disclosure. And it's not so much that their data is wrong, as it is the world changes - typically faster than the data updates. NavTeq spent about $330M last year just updating their data. And it's still wrong!

User generated data is generally better - but there has to be easy ways to integrate the information. TomTom MapShare is the leader among commercial vendors integrating user generated data. Open Street Map is an entirely user generated base map. Google relies on user generated data in areas where the commercial data is non-existent.

There are two significant problems with user generated data. First, it's the liability issue - you end up with even more disclaimers. If you get sent into a berm because of someone else's data edits, what do you do? Second, users don't understand the topology of road networks - so the routing algorithms get screwed up. This is a fundamental problem with Open Street Map (although people in Germany have gone through and cleaned the road network topology so it can be used for routing there).

Re: audio quality, my guess is that they're trying to cheat the talk-only stations out of bits and its failing horribly, because I didn't find the music nearly as hard to listen to. Of course with modern autotune technology it might be hard to differentiate between good and bad...

On user generated data, it sure seems that with enough people passively participating in something like Open Street Map the topology should largely fall out: Given enough samples you'll see pretty much everyone take every intersection. Come to think of it, this would probably be good for routing optimization, too: Get enough samples showing that even though the device recommends 4th Street in San Rafael people are consistently taking 2nd street and getting better results, you can weight that line in the graph appropriately.

If you could get some sort of vehicle information out of the contributors, you could probably do some decent correlation stuff there, too.

Dan: Automatically extracting topology and routing preferences from GPS data would amount to a nice PhD research project - and would probably work in places like Kansas where everything fits in 2 dimensions and more angles are 90 degrees. And that assumes a graph that connects properly (i.e., correct topology).

The roads in OSM aren't made from GPS tracks. OSM lets you overlay GPS tracks basically over a canvas onto which you draw the roads. Or, in places where the imagery is good enough, you just heads-up digitize (i.e., trace) the roads on the imagery. It's not nearly as sophisticated as one might think.

The problem with topology and routing is that you have to have close to 100% consistency in how your road segments connect - otherwise the routing algorithms send you around holes in the topology as though the roads didn't exist or were infinitely long. The only difference being that one may crash the algorithm and the other might not. NavTeq and TeleAtlas use very comprehensive training to get their data collectors to produce a consistent topology. NavTeq and TeleAtlas also capture lane-topology and handle the third dimension properly. Just because Petaluma Blvd S intersects 101 in 2 dimensions doesn't mean you can turn onto the 101 at that intersection.

As for a somewhat related topic of managed traffic patterns, why don't we just follow the ants?

#Comment Re: made: 2009-06-09 12:47:19.151281+00 by: Dan Lyke [edit history]

Hmmmm... Once the parts for my tracker get in I'm going to have to play with some of this. The issues I see with generating the topology automatically mostly revolve around filtering ±10M spatial samples and the relatively low temporal resolution. I'll have to see what happens when I can get a few tens of passes through the Petaluma Blvd S and 101 intersection, all ~~four~~ six ways.