Smarter Traffic Lights

By Roland Piquepaille

If you're like me, I bet you hate moments when you're in a hurry and all the traffic lights seem to intentionally switch to red just in front of your car. Now, according to Nature, a Belgian traffic researcher thinks that traffic lights that respond to local conditions could ease congestion and reduce your frustration. His method would not give you the individual power to switch the light to green. But if you were part of a group of cars approaching a red light, inexpensive traffic-flow sensors would detect your group in advance and turn the light to green. His simulations show that such adaptive traffic control is 30% more efficient than traditional ways of regulating traffic. However, his system has not been adopted by any large city. So you'll continue to be frustrated by these ?%&$!§ traffic lights for a while. Read more...

Here is a description of the problem.

Carlos Gershenson at the Free University of Brussels, Belgium, says his system of traffic lights would be able to adapt to changing traffic conditions, allowing it to find a better switching sequence than one imposed rigidly on all situations.

There have been some attempts to make traffic signals more flexible, responding to the state of the traffic. These intelligent 'advanced traffic management systems' generally connect the lights to a centralized computer that is constantly seeking an optimal switching sequence.

Such approaches are costly to implement, however, and can be computationally very challenging. In Gershenson's method, by contrast, traffic lights at a junction act on their own, responding simply to the local conditions.

Gershenson used computer simulations to estimate traffic conditions under four systems, traditional ones and his proposed adaptive ones.

Gershenson also tries two adaptive schemes. In the first, called 'request' control, a traffic light switches from red to green if the number of vehicles approaching it, or the time vehicles have spent waiting, exceeds a certain threshold. Under such conditions, a large convoy of cars can force a red light to go green as it approaches a junction, opening up a 'green corridor' as the convoy progresses across the grid.

Alternatively, in so-called 'phase' control, the same rules apply except that there is a minimum time for switching from stop to go or vice versa. Gershenson finds that, in his simulations, adaptive request control is the most efficient for low traffic densities, but works poorly for dense traffic. Adaptive phase control also works well at low densities, and doesn't clog up at high densities either; so on average, it is the best method overall. Both schemes are typically around 30% more efficient than the non-adaptive ones.

Gershenson admits that the benefits wouldn't be as large in a big city where the situation is much more complex than in his simulations. But as the method involves only low costs for its implementation, maybe it will be used one of these days.

The research work has been published by arXiv. Here is the abstract of the paper named "Self-Organizing Traffic Lights."

Steering traffic in cities is a very complex task, since improving efficiency involves the coordination of many actors. Traditional approaches attempt to optimize traffic lights for a particular configuration. of traffic and density. The disadvantage of this lies in the fact that traffic configurations change constantly. Traffic seems to be an adaptation problem rather than an optimization problem. We propose a simple and feasible alternative, in which traffic lights self-organize to improve traffic flow. We use a multi-agent simulation to study two self-organizing methods, which are able to outperform two traditional rigid methods. Using simple rules, traffic lights are able to self-organize and adapt to changing traffic conditions, reducing waiting times, stopped cars, and increasing average speeds. Even when the scenario simplifies real traffic, results are very promising, and encourage further research in more realistic environments.

And here is a direct link to a draft of the full paper (PDF format, 16 pages, 426 KB).

And if you want to know more about traffic lights, here is an article from HowStuffWorks, "How does a traffic light detect that a car has pulled up and is waiting for the light to change?".

Sources: Philip Ball, Nature, December 3, 2004; arXiv, November 30, 2004; HowStuffWorks website

Related stories can be found in the following categories.

• Artificial Intelligence
  
• Sensors
  
• Transportation