Lights Out for Traffic Woes?

Over the years, Dr. Dobb's has published a number of articles about traffic signals. Let's see, there was Extended State Diagrams and Reactive Systems and Visually Designing Embedded-Systems Applications, both by Doron Drusinsky. Then there was The Halting Problem by Ed Nisley. And Ala Qumsieh wrote Fuzzy Logic in Perl for The Perl Journal and touched on traffic signals.

Now the authors didn't write these articles, and we didn't publish them, because we're a bunch of caffine-stoked, Type A personalities who hate sitting in the car waiting for the traffic light to change. No, we're all laid back Type whatevers who like an intellectual challenge -- and traffic signal systems are great for this. Moreover, as we're learning these days, efficient traffic signal systems are really important in terms of energy consumption and the environment.

For instance, according to Dirk Helbing, a Professor of Sociology at the ETH Zurich Chair of Sociology, traffic flows account for up to one-third of global energy consumption -- so efficiently managing traffic flow can significantly reduce waste and lower harmful emissions. Consequently, Helbing and Stefan Lammer of the Institute for Transport and Economics at Dresden University of Technology, are proposing propose a self-organized control system for traffic lights to improve vehicular traffic flow.

The first problem, says Helbing, is that most systems are antiquated. Traffic lights are optimized for pre-established assumptions, and not for real time. Plus there was far less traffic when the systems were installed. To deal with this, the solution has been to simply install more lights. However, says Helbing, "even for normal-sized cities, supercomputers are just not fast enough to compute all of the different options that exist for controlling traffic lights. So the number of choices actually considered by the optimization program is significantly reduced." He goes on to say that "the variation in the number of vehicles that queue up at a traffic light at any minute of the day is huge, and you are optimizing for a situation that basically is true on average but that is never true for any single day or minute: essentially for a situation that never exists. Plus, even adaptive traffic lights in modern control schemes are usually restricted to a variation of cycle-based control."

What Helbing and Lammer propose is an optimized decentralized system that makes travel time more predictable, though traffic light sequence less so. Along with the optimization, they add a stabilizing strategy. Interestingly, by themselves each strategy -- optimization and stabilization -- underperformed. But when properly combined, the approaches performed great at controlling traffic at all volume levels.

Simulations they ran went well with non-periodic traffic lights releasing long traffic lines, making travel time more predictable. Fuel consumption and C02 emissions also went down.

Of course, all this would be a far less problem if we all were riding motor scooters :-)

-- Jonathan Erickson
jerickson@ddj.com

Posted by Jon Erickson at 11:29 AM  Permalink