network20q

Waze: Social Navigation

Looks like the blogging and commenting is slowing down a little, so here’s a fun one.

I’m trying out iPhone app Waze.

It’s free navigation. That’s cool, but no big deal.

The unique part is ‘social navigation’. (I think) They use your iPhone’s GPS and clock to find out driving speed, thus congestion condition, on the streets, and then through some algorithms, guide drivers to take longer but faster routes.

There’s a whole bunch of fun stuffs that one could think about here (some probably already done in the app):

1. Inferring congestion from noisy, partial view of delay information.

2. Fusing information from distributed sources in real time.

3. Algorithms that turn fused information into recommendations.

4. User profiling based on your commute habit.

All of these touch on what we talk about in the class. But the following two dimensions stand out:

5. Networking effect: Waze is only useful if many people uses it (otherwise it’s just another iPhone GPS based navigation app). If too few around your home and work use it, you may decide not to bother to click it when you start driving. The nice point is that when the road is congested, by definition, there’re a lot of cars on the road (and more accurate the delay signals). So when you really need it, there’s at least many cars out there (and maybe a significant enough fraction of these drivers have Waze activated).

6. Braess Paradox: sometimes the more information a driver is given, less efficient is the collective behavior of route selection. If it’s a distributed decision making by each driver based on congestion information, you can’t help but wonder: if I switch to that route, other Waze drivers may be thinking the same, so that route won’t be fast by the time I get there. But if all others think like what I just thought, they won’t take it, so the route is still good. You can report congestion condition at time $t$, but you can’t easily predict intelligent drivers’ game theoretic reaction and the consequence at $t+1$. We’ll later talk about Wardrop equilibrium and the distribution of traffic at equilibrium. Now, if it’s a centralized suggestion instead, there’s the issue of fairness: which fraction of the drivers would Waze recommend switching to a faster route?

5 and 6 above are kind of at odds: you have a lot of intelligent drivers precisely when you have congestion.

Finally, feels like the current version of Waze is targeting commuters (if you’re on a long road trip, you probably are less delay sensitive and also would rather stick to a standard navigator’s suggestion without the uncertainty of social navigation mentioned above). But for most commuters, there’re likely 2, at most 3, paths that actually make sense and are sufficient different from each other. So the decision is actually very simple: should I go path A or B? But unless there’s an accident, I usually know which path is better at which hour anyway.

The biggest hit of Waze might be social collaboration to report where the police cars are hiding.