Driving through any of the world’s large cities is frustrating. It always seems that the traffic lights are conspiring against you and stopping you from traversing from point A to B in the most efficient way. One’s mind always wonders if there isn’t some way to make the lights intelligent. Sensing when to change to reduce the average idle time while allowing more cars to pass without stopping.

Again, KSX is an idea tool to implement just such a system. Traffic control is a hierarchical optimization problem that has time of day dependencies.

The first objective is to make each of the traffic lights or semaphores smart. That is, aware of the time of day, basic turn red, green or yellow rules, and perhaps what traffic looks like in all directions based upon locally mounted cameras. The camera images can provide the basis for adaptive rules based on traffic demands.

Once individual lights are made smart through local analysis and rules then the next step is to combine individual light/KSX systems together through our distributed portals to create a higher level of optimization and knowledge to oversee multiple lights and direct them in a coordinated way. Combining these sub-systems together in ever-larger systems, only to be combined again is the blue print to intelligent traffic control.

A problem, or more appropriately, an opportunity as large as intelligent traffic control, is a daunting task. However, breaking it up into small pieces, embedding intelligence at a local level is a great way to approach the opportunity. By compartmentalizing this way insurmountable tasks can be defined and solved.

Here are a few samples of the types of rules that might be employed at a couple levels of an intelligent system.

Individual traffic light:

1. If the time of day is between A and B then red on time is X, yellow on time is Y and green on time is Z.
2. If direction A is green and direction A number of cars visible is very low and direction B is red and direction B number of cars visible is very high then trigger direction A to red and trigger direction B to green.
3. If direction A weighted average wait time is greater than direction B weighted average wait time then incrementally increase within limits the direction A green on time and decrease, within limits, direction B green on time.

As you can see from these rules that use both crisp and fuzzy terms that it is possible to establish rules that indeed transform the average light from dumb to smart, or one that is aware of current conditions and then responds intelligently to them.

Grouping Multiple Lights Together From Improved Control:

A traffic “system” starts to emerge when individual lights are grouped together to work together. Rules written at this level are designed to coordinate variable parameters, at the individual lights, taking into account multiple lights as well as over arching strategies and objectives. At this level we are starting to “supervise” individual lights to optimize the whole by using individual lights intelligently.

Here, the actual distances between the lights are known and it is now possible to take into account average travel rates given the current light sequences of each individual light and how they are phased together. Again, rules at this higher level might be more concerned with both subjective issues like driver comfort as measured by combined stop time through the series of lights or economic measures like the combined idle time at each light through the series of lights. Each of the individual lights, through their cameras can track the current traffic loads in all directions and report that to the coordinating KSX node.

In my next blog post I’ll discuss multi-agent systems in general and how KSX can be used as a agent-based simulator leading to a third post suggesting how KSX can be used to perform multi-agent simulation of an intelligent traffic system.