Overview
It can be quite frustrating when you come home in the middle of the night only to be stopped by a red light when no one is in sight for a mile. It is even more frustrating when you go on a 55 mph road only to brake and stop every half mile at a traffic light. These are the functions of antiquated traffic signals designed several decades ago to manage peak traffic. The lights are not controlled in real-time, but rather are optimized (if at all) for conditions observed in the past.
The most basic type of traffic control systems are fixed-time systems that utilize either simple manual strategies or optimization packages to determine green splits, cycle lengths, and offsets using historical data. Variations on the number and speed of vehicles alone are quite large -making it impossible to account for all different combinations during optimization. This is only compounded with the incorporation of other key inputs such as time of the year and environmental conditions. So, while these systems can be adapted to respond to recurring traffic congestion, they are tillable to accommodate irregular congestion and other variable traffic conditions such as the weather, special events, and accidents. Even for recurring conditions, the performance of these types of systems is always suboptimal since it is not based on actual real-time traffic conditions.
In addition to fixed-time systems, there are intelligent light signals that react to the presence of vehicles at an intersection, While, they work very well for peak-traffic conditions, they can cause problems for management of non-peak traffic such as when an entire lane of fast-moving traffic comes to a halt at a red light that was triggered a second earlier by a single car arriving on a perpendicular road. In another case, a vehicle from a crossroad has to trigger a signal and wait for it to turn when no vehicle is in sight for miles. Would it not be better if signals were changed based on real-time intelligence of approaching traffic rather than relying on a traffic signal that triggers only on arrival- disrupting the smooth flow of traffic?
Traffic lights are primarily controlled individually, but there has been some work on synchronization of multiple traffic lights. This synchronization is usually done in one direction based on traffic volume, but this often leads to unneeded traffic delays in other directions. Several dynamic control systems have been proposed for traffic control based on traffic flows across multiple light signals; however, most of these are controlled at and communicate with a central location. Issues associated with such central control, include: scalability, resilience, and latency. As the number of traffic lights increases, it becomes exponentially difficult for them to be centrally controlled since increases in the number of variables lead to subsequent increases of computational complexity. Also, since the controls are not typically programmed to deal with exceptions, any traffic anomalies result in significant traffic disruptions. Finally, in a centralized system, there are often delays in acquiring, processing, and relaying information across multiple traffic signals. To deal with the issues associated with a centralized control system, we propose a distributed solution based on the concept of self-organization.
Self-Organization
Self-organization is the spontaneous emergence of globally coherent patterns out of local interactions among initially independent components (agents), without central control. These simple interactions between agents lead to emergence of sophisticated collective behavior. Throughout nature, evidence can be found of self-organization -in flight patterns of birds, schools of fish, light emissions of fireflies, hive behavior of bees and ants, formation of crystals, magnetic behavior, as well as in many other biological and ecological systems. This same behavior can be witnessed in man-made entities such as Internet hyperlinks and social networking. In a school offish, for example, fishes seamlessly organize into patterns that look more intimidating to predators than ally individual. Instead of having one "leader" fish that communicates with every fish in the school to create a pattern, each fish communicates to its closest neighbors and ascertains its position in tile pattern.
Such self-organized systems benefit from the ability to quickly adapt to changes in the environment and are endowed with attributes of self-healing and robustness, For instance, individual ants are primitive insects with limited memory and are capable of performing simple actions. However, ant colonies express sophisticated emergent behavior for tasks such as forming bridges, finding the shortest route to a food source, and carrying large objects. Any individual ant does not have global knowledge about the task it is performing; however, by coordinating with neighboring ants, a collective intelligent behavior emerges. If any ant in the formation is killed, another ant immediately takes its place so that the colony continues to function at the same level.
We believe that traffic control can benefit similarly from self-organization of traffic signals where individual traffic lights will be able to change their signals based on real-time comn1unication with their neighbors. Over time, as neighbors communicate and change their individual behavior, the entire traffic system will be synchronized. Any anomalies that occur will lead to chal1ges in behavior of traffic signals in close vicinity and via local interactions of the traffic signals, the entire traffic system adapts to this change.
Activities
In this work, we will define parameters that can be used to characterize traffic, identify constraints for traffic lights, and determine rules of self-organization for the traffic signals. As a part of this work, we will identify metrics to measure performance and scalability with respect to traditional dynamic traffic control approaches. A simulation for a self-organized traffic system will be developed to model this self-organizing behavior. Feasibility of such a transport system will be evaluated through the simulation and the performance of this system will be evaluated in context existing systems. Anomalies such as accidents, slow downs, large public events, and emergency vehicle passage, may be introduced to test the robustl1ess of the transportation model. Existing traffic and agent-based simulation tools will be considered for use in this simulation. The funding for this work will be used for student and faculty support as well as for procurement of any necessary software.