Tuesday, May 3, 2011

Traffic Congestion as an Information Problem

I've mentioned IBM's Smart Traveler program before, but the New Scientist has a short article about it that deserves comment. From the article:
The Smarter Traveler Research Initiative blends real-time traffic data with past traffic patterns to predict congestion up to 40 minutes into the future. Drivers are then automatically sent an email or text message of conditions on their regular commute before their trip begins.


"If you are already on the road and a sign says 'congestion next 5 miles' you may have very few options," says Naveen Lamba of IBM. "But if you get that information prior to starting your journey, you can choose to stay at home, work late or take a different route."

What this explanation suggests is that IBM views traffic congestion as an information problem. If you know your route will be congested you can alter your behavior.

Yet don't most commuters already know this? If you leave work at rush hour, the roads will be jammed and anyone who has commuted more than one day will not be surprised by this. Commuting is often a regularly occurring phenomenon, where drivers learn about their commutes simply by repeating them a couple of times a day. I expect that drivers already decide when to travel and what route to take as best they can. It isn't obvious to me (at this point, but that's why we do research) that specific knowledge about traffic congestion is necessarily better than general knowledge about traffic congestion in terms of changing travel choices. I don't view congestion as an information problem, where congested conditions can be avoided simply by knowing more. Congestion occurs because too many people want to use scarce roadspace at the same time, even though the motorists know lots of other people want to move around as well.

I do see potential for these Smart Traveler types of applications for autonomous vehicles, though, and projects such as IBM's may be useful in that regard.


David said...

I agree that for recurring congestion this would be close to useless. For non-recurring congestion (incidents/crashes/events) this might have some use. I suspect it cannot be sufficiently accurate to be very helpful, and if it is wrong it may be detrimental, but on average it might be able to be a net positive. See




for some model results with good and bad information.

Unknown said...

I do see the potential is some cases. I'm more curious about how much of congestion (whether recurring or not) can be avoided though additional information. Specifically, what is the value of these technologies above and beyond those already available using radio alerts, traffic maps and highway signage. For each congesting event the value of information depends in part on available substitutes.
As you point out in your ATIS paper, the value of information may have meaningful effects on the edges of rush hours where congestion can be minimized through small changes in traffic. The IBM and other similar technologies can also cover more areas than radio and other conventional ways which is also beneficial. As I mentioned, this is why we do research. I see the potential benefits but it isn't obvious that they will be substantial, and certainly not as substantial as the news stories proclaim. I suspect the researchers are more measured with their claims. However, the more our cars choose the routes to take and the less drivers make these choices, the more valuable these services. This may very well be the norm soon, in part because of things like this technology.