The Soapbox

This page written circa 3 January, 1999.

Come with me Traffic Light, I Think I Can Help You

Traffic lights come more and less clever. The simplest approach is to time the interval spent green in each direction. I do not know of any traffic lights that still work this way. Virtually all traffic lights have sensors in the road surface, sort of land mine detectors in reverse, that sense the metal of a vehicle overhead. Thus they can tell when something is above the sensor, and adjust their light timings. They contain a simple microprocessor, and are programmed with instructions to alter their behaviour in response to the sensor reports.

Of course, the programs can be better and worse. What is possible? I find the best way to explain what to expect of a traffic light is to ask you to imagine that you are in a dark room. Imagine that you are controlling a simple crossroad intersection. Ignore pedestrians for a moment. You have, say, 8 indicator lights, each one tells you if there is a vehicle above the spot where a car would stop if the light was red. (You have four roadways of four lanes each, say, that is two lanes of approaching traffic in each of the four directions.) You must switch the lights to red one way or red the other, based on the eight lights. Imagine in this first instance you have one switch to do this. (The yellow-red sequence is automatic, you just decide when to change.)

In quiet traffic times, it is easy. Imagine all of your lights are off. When one comes on, if the lights are red against that vehicle you change them. You will see only a quick flash on the light if the set is already green to that direction, of course. As traffic gets heavier, more lights come on and eventually go off in more lanes. You can count the blinks, get an idea of which route is more busy, spend longer green that way.

If there are turning lanes and turning arrows, it is more complicated, because there are more choices than just red-green or green-red, but the problem is essentially the same, you have to give each possibility its turn, duration regulated by usage, if there is a car there wanting that possibility. Same with pedestrians, more choices, but the same game.

The next addition is that you know the time of day, maybe the day of the week, and you know which road is the main road. You know to expect peak hours, you know to leave the lights green to the main route when there is no traffic, you know which road to favour at various peak hours. (Remember, you can't see how big a queue is at you light, and in peak hour you might be giving equal time to each direction, with 50 cars queueing in one direction and 500 in the other! Knowing the peak can help a lot.) This is about as clever as Sydney's lights were when I was first a driver, in the early 1970s.

Now, let us consider something interesting. If one light were to stay on for more than one change of the lights, you would get suspicious. Either something is parked over the sensor, or broken down, or it is broken. After a couple of changes, with no activity, you would ignore that light, and if you have the facility, you call your host system and report a problem requiring human intervention. If you have a line to a host, you probably also check your own light globes, so you put in a service call if a lamp blows. This is what lights did in Sydney by the late 1970s. They were getting networked.

Of course, this occasionally goes wrong. I had to get out of my car en route to UWS one day and ask the blockhead in front of me to pull close enough to the lights to trip the sensor... but then cretins in the path is no new phenomenon.

Another innovation, briefly tried, was to add extra sensors in each lane, so that the traffic light controller (this is you in the dark room) knows (1) if the queue is getting long, (2) if a car is approaching a red light that could go green before it gets to a complete halt. You can also deduce if a "stuck" lamp is a stuck car or a stuck sensor. (Exercise: How do you do this?)

The next step is to realise that the "advanced" set of sensors can be the sensors in the next intersection, provided the different intersections can talk to each other. Once networked, however, the most useful advance is for adjacent traffic lights to cooperate. Thus is born the idea of coordinating lights on a main road so that they all go green to the main stream at once. In peak hour, the changes propagate as "waves" travelling in the direction appropriate to facilitate the peak flow at the cost of disrupting the contraflow. Now we are realy cooking. You in the dark room have voice contact with all the other people in dark rooms, comtrolling their own intersections. You get and give advice, you arrange to move in concert. The job is now really one for a computer... such concentrated, continuous counting and communicating is more than humans can get right for any duration.

In Sydney, traffic lights do this. The system is called SCATS, and it is a wel-known engineering success; in addition, it is human-overseen in control rooms with many videocameras, so that people can deal with things that no amount of intelligence in the dark room can handle, because the intellignce is essentially blind.

In California, freeways are relatively good. Traffic lights, compensatingly, seem to be pretty stupid. I saw a car broken down on a sensor the other night, and the lights were all but stuck on the route with that car, and no other traffic. Several cars patiently waited for 60 seconds for not one moving vehicle, just the broken down car, pulled over on the sensor. This was on an intersection in Santa Rosa, near the business district, with other sets of lights not far away in all directions. Still, the freeways are great.

Traffic congestion can suck away a lot of your life. Only an hour of your time each day, if you commute by car, amounts to about 6% of your waking life. The difference between a commute in peak hour and on a quiet night is easily 30 minutes, if you are but a few city miles away. In the last set of referenda, Californians heartily approved the large-scale widening of 101 (the main N-S freeway), though I am glad to say that they simultaniously rejected a sales-tax hike to pay for it. However, it is also about time they poached Australia's Traffic-light designers.

The title, incidentally, comes from an Arthur Brown track from an album that dates, from memory, from the 1970s.

| Home | Back |