ArticlePDF Available

Self-Organization in Traffic Lights: Evolution of Signal Control with Advances in Sensors and Communications


Abstract and Figures

Traffic signals are ubiquitous devices that first appeared in 1868. Recent advances in information and communications technology (ICT) have led to unprecedented improvements in such areas as mobile handheld devices (i.e., smartphones), the electric power industry (i.e., smart grids), transportation infrastructure, and vehicle area networks. Given the trend towards interconnectivity, it is only a matter of time before vehicles communicate with one another and with infrastructure. In fact, several pilots of such vehicle-to-vehicle and vehicle-to-infrastructure (e.g. traffic lights and parking spaces) communication systems are already operational. This survey of autonomous and self-organized traffic signaling control has been undertaken with these potential developments in mind. Our research results indicate that, while many sophisticated techniques have attempted to improve the scheduling of traffic signal control, either real-time sensing of traffic patterns or a priori knowledge of traffic flow is required to optimize traffic. Once this is achieved, communication between traffic signals will serve to vastly improve overall traffic efficiency.
Content may be subject to copyright.
Self-Organization in Traffic Lights:
Evolution of Signal Control with Advances in Sensors and
Sanjay Goel1, Stephen F. Bush2, Carlos Gershenson3
1 School of Business, University at Albany, SUNY
2 GE Global Research
3 Instituto de Investigaciones en Matemáticas Aplicadas y en Sistemas & Centro de Ciencias de la Complejidad,
Universidad Nacional Autónoma de México; SENSEable City Lab, Massachusetts Institute of Technology; ITMO
AbstractTraffic signals are ubiquitous devices that first appeared in 1868. Recent advances in information and
communications technology (ICT) have led to unprecedented improvements in such areas as mobile handheld
devices (i.e., smartphones), the electric power industry (i.e., smart grids), transportation infrastructure, and
vehicle area networks. Given the trend towards interconnectivity, it is only a matter of time before vehicles
communicate with one another and with infrastructure. In fact, several pilots of such vehicle-to-vehicle and
vehicle-to-infrastructure (e.g. traffic lights and parking spaces) communication systems are already operational.
This survey of autonomous and self-organized traffic signaling control has been undertaken with these potential
developments in mind. Our research results indicate that, while many sophisticated techniques have attempted to
improve the scheduling of traffic signal control, either real-time sensing of traffic patterns or a priori knowledge
of traffic flow is required to optimize traffic. Once this is achieved, communication between traffic signals will
serve to vastly improve overall traffic efficiency.
Index TermsTraffic signals, Self-organization, Information theory.
Traffic congestion is a perpetual source of frustration, lost productivity and greenhouse gas emissions. In 2011,
U.S. commuters are estimated to have wasted 2.9 billion gallons of fuel, equivalent to $121 billion, stuck in traffic [1].
Traffic signals regulate the flow of traffic through intersections and improve throughput. However, if configured
incorrectly, they can have a net negative impact on traffic flow. Consequently, optimization of traffic signals is essential
to the operation of the traffic grid. This is a complex multivariate optimization problem where road network topology,
intersection density, red signal probabilities, distance, speed, traffic density and other variables have to be considered.
Moreover, traffic light coordination is an EXP-complete problem [2], so large networks cannot be optimized in practice.
Since the precise traffic demand is changing constantly, any optimal scheme soon becomes obsolete.
Traffic control measures have long sought to address these issues and there have been many improvements
made over the course of the last century. More recently, however, these improvements have plateaued. Although
research on traffic control systems continues, as this review suggests, according to Zhao and Tian [3] most deployments
of adaptive traffic control systems in the U.S., for example, consist of the SCOOT or SCATS systems developed in the
1970s. In this paper we show that the concept of self-organization can leverage the advances in sensors and
communication to provide a new leap in traffic light coordination, as traffic lights can systematically adapt to changing
demands. This paper first provides a historical perspective on the development of a variety of traffic control systems and
techniques, leading to a review of self-organized traffic control. We provide a discussion of the basic components and
characteristics of self-organized traffic control systems, and their importance for future traffic optimization.
1.1 The Evolution of Traffic Control Systems
The first traffic signal was installed outside the British Houses of Parliament by J. P. Knight in 1868. The signal
consisted of a gas lantern that exploded a year later, severely injuring the operator [4]. The first electric signal, designed
by Lester Wire [5], was installed in Cleveland, Ohio in 1912 [6]. By 1917, interconnected traffic signal systems were in
use in Salt Lake City. In 1922, the first automatically controlled interconnected system of traffic signals was installed in
Houston, Texas [7].
Many traffic signals still function at a single intersection, and have pre-stored signal timing plans based on
historically determined flow patterns. Initial attempts to improve these signals were made by applying Erlang's (1909)
queuing theory, and F.V. Webster's (1958) queuing model for determining the optimal traffic signal cycle time.
Refinements to Webster’s model continue to be made [8]; [9]; [10]; [11]. Since their initial usage, several attempts have
been made to create a closed-form solution to address the drawbacks of this method [12]. Such systems are simple, but
despite all attempts to optimize them, they remain inefficient primarily because they are not based on real-time
conditions. Since traffic demand changes with each cycle, so too do the number of queuing vehicles and the idling green
times, and thus the optimal phases. Manual programming of each signal also makes frequent adjustments laborious and
cumbersome. To regulate traffic at intersections, roundabouts are efficient for low speed, low-density roads, but not for
high speed or high-density arteries [13] [14].
Modern traffic signals leverage sensors to adapt to prevalent conditions at the intersections. Such
systems that respond to current traffic conditions are known as adaptive traffic control systems (ATCS). ATCS make
signal timing decisions at short intervals based on real-time traffic conditions. These systems have been called third-
generation systems [3]. ATCS still leave a lot of room for efficiency improvement; by leveraging sensors and
communication traffic flow can be predicted better reducing wait times further. ATCS contrast with first-
generation systems, which use historical data about traffic patterns to determine signal timing plans. ATCS are also an
improvement to second-generation systems, which make use of signal timing plans based on surveillance data, but are
restricted as to how frequently the response is generated [15]. As summarized in Table 1, we foresee self-organized
systems as fourth-generation traffic control systems, providing a logical progression of improvement.
There are still a number of shortcomings with ATCS. Although ATCS are actuated signals, there is latency
associated with the response of the signal to the trigger, causing inefficiencies. Moreover, adaptive signals work well for
peak-traffic conditions; however, they can cause problems, for instance, when an entire lane of fast-moving traffic comes
to a halt at a red signal that was triggered by a single vehicle arriving at a side road. These signals are expensive, fragile,
and require constant maintenance. These and related problems are discussed in the authoritative traffic control handbook
prepared for the U.S. Federal Highway Administration [16] .
Table 1. System types and their evolution
System Type
Pre-stored signal timing plans
Uses historical data about traffic patterns to
determine signal timing plans.
Fixed-time signal timing plans
based on surveillance data
Signal timing reacts to traffic conditions, but
at fixed time intervals, and not in direct
response to real-time traffic flows.
Adaptive signals with
synchronized and centralized or
decentralized control
Signal timing reacts to real-time traffic
conditions, making constant changes based
on current traffic patterns. System is
operated under centralized or decentralized
Adaptive signals with
decentralized, and self-
organized control
Signal timing reacts to real-time traffic
conditions at multiple intersections in a
network through decentralized and
distributed control.
Attempts have been made to synchronize multiple signals together rather than manage them independently.
This involves offsetting the timing of subsequent signals, such that a car traveling through intersections continues to see
green light across multiple signals. This is usually done on the basis of traffic volume in one direction, and can often lead
to unnecessary delay in the other direction. Several dynamic systems have been proposed for control based on traffic
flows across multiple signals; however, most of these communicate with, and are controlled from, a central location.
Issues associated with central control include scalability, resilience, and latency. As the number of traffic signals
increases, so does the number of complicating variables; as the complexity increases exponentially, centrally controlled
systems do not cope well with anomalies [2]. A variety of approaches involving centralized control have been used to
optimize signal timings, such as: mixed-integer linear programming [17], [18], [19]; neural networks [20]; [21]; genetic
reinforcement learning [22]; and fuzzy logic [23]; [24]. He et al. [25] use mixed integer programming to prioritize
conflicting priority access to intersections of pedestrians and emergency vehicles, while optimizing signal coordination.
A model with independent controllers at each grid point that transmit data to a central controller for
optimization and receive feedback in real-time is proposed by Li and colleagues [26]. This model sacrifices latency for
computational power of a central control processor. Unfortunately, these models become more complex as the number of
signals increases. Traffic control could be enhanced by analyzing static and mobile vehicle sensor data, although this
would be extremely time-consuming. Other approaches to the problem involve graph theory and Petri nets [27]; [28],
heuristic algorithms [29] [30] and meta-heuristic search algorithms [31]; [32]; [33]; [34]; [35]. A schematic illustrating
different techniques and methods of traffic control is presented in Figure 1.
Figure 1. Traffic Signal Control Literature
1.2 Argument and Outline
Several innovative techniques can be applied to further improve traffic optimization, including queuing theory
and graph spectra. For instance, the time that a car is “blocked” is determined by the number of red signals encountered
en route to its destination; this can be determined analytically based on the density of intersections (the graph structure),
the probability of red signals (the control algorithm), as well as the distance, speed, and density of traffic. This again
becomes increasingly complex as the number of signals in the graph increases.
Ultimately, optimization of traffic systems, as in other mathematical applications, involves the use of an
objective function to determine the best set of inputs from a set of available alternatives, which maximizes (or
minimizes) some measure. In traffic studies, this measure usually includes average delays, stops or throughput. Although
many of the methods and models discussed here may provide optimum solutions within a defined space, as traffic is non-
stationary, the optimum changes constantly. Therefore, approaches based on objective functions do not always find the
best possible traffic performance [36].
Decentralized, distributed, adaptive, and machine learning techniques for traffic signal control have been used
with varying degrees of success. However, with recent advances in sensing and communication technologies, algorithms
and techniques that were not previously thought applicable may now serve to improve traffic signal optimization. Self-
organization is very effectively used in nature to manage complex systems with a large number of components [37].
Similarly, we believe that a distributed solution based on the concept of self-organization is possible in traffic control,
and could provide the best means to optimize traffic flow patterns.
The outline of this paper is as follows. Section II reviews adaptive traffic control models that have been utilized
in practice, as well as a number of traffic control models provided in the literature. We also review some aspects of urban
traffic models. Section III examines in more depth particular techniques employed as part of traffic control systems.
Section IV delves into self-organizing systems as an adaptive and efficient approach to traffic control. The general
concept and challenges of self-organization are reviewed, as well as analytical approaches involved in self-organized
traffic control systems. Section V looks at the driving forces for these principles: new capabilities in data collection and
communication. Section VI examines design elements of self-organizing systems. Section VII provides concluding
Traffic control systems use a variety of strategies to determine cycle lengths1, phase splits2 and offsets3. One
commonly used traffic control system is the Urban Traffic Control System (UTCS) developed by the U.S. Federal
Highway Administration (FHA) in the 1970’s. In UTCS, timing is optimized4 for each signal based on average traffic
conditions at a given intersection. Signal timing schedules are then developed and periodically downloaded to each
controller. The objective is to maximize the bandwidth of major arteries and minimize delays. Given the unpredictability
of traffic patterns, however, this technique is obviously suboptimal. In general, constant variations in the number and
speed of vehicles make it difficult to account for all possible combinations during optimization exercises.
2.1 Prevalent Adaptive Traffic Control Systems
Real-time adaptive traffic control systems that incrementally compute local conditions were developed to
address inefficiencies in the UTCS. The most notable adaptive systems are: Optimized Policies for Adaptive Control
(OPAC) [32]; [38]; the Split Cycle Offset Optimization Technique (SCOOT) [39]; [40]; [41]; the Sydney Coordinated
Adaptive Traffic System (SCATS) [42]; [40]; [43]; and the Real-time Hierarchical, Optimized, Distributed, Effective
System (RHODES) [44]; [45].
OPAC was first developed at the University of Lowell in the early 1980s, and has evolved over the years [46].
It is a real-time distributed signal timing system based on measured and predicted traffic demand. OPAC does not use
the concept of cycle; instead it optimizes control periods for particular intersections through phase switches at fixed time
intervals. At each switching sequence a delay function is defined based on the initial queue length times the interval of
time in which the cars are queued, with the goal of minimizing total vehicular delay [38]. For each stage (a sequence of
one and no more than three phases) the optimal switching decisions are calculated independently using on-line data from
upstream detectors as well as historical data. The optimization procedure is essentially a constrained search method [46].
SCOOT [39], developed in the UK, uses a steady-state model of traffic flow and platoon-dispersion equations
to minimize queue lengths and delays in the grid. SCOOT gathers data about approaching vehicle platoons from
upstream detectors in order to update the offset and cycle time at intersections downstream. Data is continuously
1 Cycle length is total time required for the entire phase set at an intersection where a phase set is the set of unique
combination of vehicle or pedestrian phases that occur at the intersection
2 Phase splits are the portion of the cycle length devoted to different phases during a complete cycle
3 Offsets are the time delay in start of signal cycle across two adjacent intersections typically used to coordinate
traffic flow across multiple signals.
4 We use the optimized as a measure of improvement of the fitness function. Given the complexity of the problem a
global optimum is not guaranteed or even achievable.
collected and stored in a processed format by ASTRID, the “Automatic SCOOT Traffic Information Database.”
Relevant data include: the number of vehicles per hour arriving at the intersection: total delay in vehicle hours (average
queue length in vehicles on a link); congestion; detector flow (recorded as vehicles cross a detector); and detector
occupancy (the total number of quarter-seconds that the detector is occupied as a percentage of the whole period). Data is
collected at a number of levels, including detector, link, node, region, route, and area.
Real-time data collected by SCOOT is used to generate traffic flow models, or “cyclic flow profiles. These
profiles are used to estimate the number of vehicles arriving at intersections downstream, providing estimation of queue
size for hypothetical changes in signal timing parameters. Prior to each phase change the flow model is used to determine
whether to delay or advance the time of the phase change by 4 seconds, or leave it unaltered. Similarly, once in each
cycle, a decision is made to either advance or delay the offset, or to leave it unaltered. Finally, once every few minutes, a
decision is made to determine if cycle time should be increased or decreased.
SCATS [47], developed in Australia, is a dynamic control system with a decentralized architecture that updates
intersection cycle length using detectors at a stopline. In this system, offsets between adjacent intersections are
predetermined and adjusted based on sensor information collected in real-time. The SCATS model groups intersections
into subsystems, each subsystem containing only one critical intersection. The critical intersection coordinates all other
intersections in the subsystem. Each intersection, at the subsystem level, can adjust the signal phase independently based
on traffic conditions. Moreover, time saved from any phase change is included in subsequent phases in order to maintain
a common cycle time for all intersections in the subsystem.
SCATS uses as its basic data the degree of saturation, or the ratio of the actually used green time to the total
available green time. The cycle time of a critical intersection is periodically adapted to preserve a high degree of
saturation, and the system also strives to maintain an equal degree of saturation on competing approaches by adjusting
phase splits at the critical intersection. Although degree of saturation is used to improve phase timing, the SCATS system
does not explicitly attempt to optimize any performance measure such as average delay or stops.
RHODES [44] is a traffic control system that predicts traffic flow in real-time, and optimizes phase timing
based on a chosen performance measure, such as average delays, stops and throughput. Its decision system employs a
hierarchical control structure. The top layer includes a dynamic network loading model that optimizes traffic based on
slowly changing characteristics. It calculates aggregated flow data of the entire network, enabling it to make initial
estimates about the load on each road segment. The middle layer, called network flow control, optimizes the flow across
individual arteries, altering the phase and cycle time to accommodate the movements of platoons. The lower layer
optimizes the flow based on individual vehicles at intersections.
2.2 Additional Traffic Control Models
A number of other models for traffic control systems have been proposed in the literature, employing a variety
of techniques. Dynamic programming and decomposition techniques were used with DYPIC [48] and in the 1980s with
the PRODYN algorithm [49]. PRODYN uses dynamic programming in a hierarchical fashion to make signal decisions
at the intersection level, while using a decomposition coordination technique at the upper level of the traffic network
[49]. Early work in field testing adaptive algorithms is exemplified by Vincent and Young’s Microprocessor Optimized
Vehicle Actuation (MOVA) strategy, which works by measuring the benefits of extending a green phase [50].
Mauro and Di Taranto [51] described a hierarchical system that applies priority assignment to public transport
and achieves global optimization. In their approach, called Urban Traffic Optimization by Integrated Automation
(UTOPIA), the decomposition algorithm considers an “intersection” and an “area” level, each with two main parts: an
observer and controller. At the intersection level, the observer estimates intersection state based on traffic counts and
state of traffic light; the controller determines the signal settings. At the area level, the observer obtains actual traffic
counts from the network based on a time discrete model. The controller optimizes the network through the rolling
horizon technique, and sends information to the intersection level. The system also gives absolute priority to selected
vehicles. In field experiments, the model showed good results for a number of criteria.
Yagar and Han [52] also proposed a real-time algorithm that takes into account the presence of transit vehicles
(i.e. vehicles that frequently need to load/unload passengers) in order to better allocate priority levels. In their approach,
detectors upstream capture information about vehicle type as well as time of detection, and create a priority list. Based on
this information, the algorithm attempts to minimize cost of stops and delays. Diakaki and colleagues [53] present an
extension to the Traffic-responsive Urban Control (TUC) model, a signal control strategy well-suited for saturated traffic
conditions. In this extension, the TUC is able to perform real-time cycle and offset control, also enabling priority
consideration for public transport.
ALLONS-D, presented by Porche and LaFortune [54], uses a decentralized architecture to provide real-time
adaptive signal plans at the level of the intersection. It uses a tree searching algorithm to minimize aggregate delay at all
approaches to an intersection [54]. Unlike the SCOOT and SCATS systems discussed above, ALONS-D is not
constrained to create a cyclical plan. This non-cyclic approach is also a feature of UTOPIA, OPAC and PRODYN, all
previously discussed. ALLONS-D uses a rolling horizon approach to determine vehicle arrivals, but unlike OPAC and
UTOPIA, for example, the horizon is defined by current traffic conditions and the time to clear all vehicles on the
intersection as identified by upstream detectors [54].
Gradinescu and colleagues [55] propose a system that can make signal control decisions that localize vehicles
in the traffic system through wireless devices that may be present inside the vehicles. The control system is supported by
a mobile ad-hoc network that is generated through the short range wireless signals around an intersection. With access to
this information, the controller can measure volume and demand of traffic and determine cycle length through Webster’s
formula based on the critical flow per capacity ratio.
Tian and Urbanik [56] developed a signal timing approach that optimizes bandwidth efficiency. The approach
adopts a partitioning technique that subdivides the arterial system into subsystems, and applies a bandwidth optimization
algorithm. Signal timings are optimized at the subsystem level, and then combined to propose a system progression band
for one direction, such as peak-traffic direction. The approach is able to maximize bandwidth efficiency in one direction,
but such maximization is not guaranteed in all directions or subsystems.
A number of other distinct approaches to optimization may be found in the literature. Bazzan [57], for example,
uses agent based modeling to propose a decentralized system that is able to coordinate and cooperate by using techniques
from evolutionary game theory. Zhu and colleagues [58] use reinforcement learning to optimize network traffic.
Sanchez, Galan and Rubio [34] [59] use a genetic algorithm to optimize the length of each cycle stage (the set of states of
traffic lights in an intersection) measured by the number of vehicles that ultimately leave the network. Sanchez and
colleagues [34] [59] also provide a modified version of a Cellular Automata traffic model that allows for the
identification of various vehicle properties and rules such as position, path, smooth braking, and others. There have been
numerous attempts to make real-time traffic controllers using techniques such as fuzzy logic, genetic algorithms, and
neural networks. These techniques will be further discussed in section 3.
2.3 Urban Traffic Models
Given the complexity of urban traffic, urban traffic simulations are required to test traffic signal control
methods. Several models for studying traffic behavior have their origins in physics, including: fluid dynamic models
[60]; [61]; vehicle-following models [62]; [63]; and coupled lattice models [64]. A cellular automaton for a self-
organized traffic grid in which signals communicate with immediate neighbors to coordinate actions is employed by both
Biham et al. [65] and Ohira [66].
The majority of traffic models are intended for studying highways. Some highway traffic models have been
extended to include intersections, while others have been proposed natively with intersections. Urban traffic models can
be categorized by their level of detail as macro-, micro- or mesoscopic [67]. Macroscopic models consider traffic as a
general, continuous fluid, such as the one presented by Lighthill and Whitham [60]. Microscopic models consider
individual traffic entities (e.g. vehicles) as discrete units, and with particular spatiotemporal behavior. Mesoscopic
models consider the behavior of individual entities but do not distinguish and trace their particular actions. As
Hoogendoom and Bovy [67] indicate, there are other criteria by which to classify traffic models, including the dimension
captured (e.g. velocity), scale (e.g. continuous, discrete), process representation (i.e. deterministic or stochastic),
operationalization (i.e. analytical or simulation) and area of application, which refer to different types of an urban traffic
network that may be examined (e.g. cross-section, single lane stretch). In this section, we review a selected number of
models that illustrate some of the different approaches taken in the literature.
Macroscopic models were the first urban traffic models to be developed, as more detailed models require digital
computers to handle their large number of variables. These models captured aggregate driver behavior and
conceptualized them as continuous flows, such as fluids or gases. Examples of these hydrodynamic approaches include
the Lighthill and Whitham [60] model and the Payne model [68] , as well as the models of Philips [69], Kuhne [70] and
Kerner et al. [71]. Examples of mesoscopic models include the headway distribution models, which use measures of the
difference in time between two successively passing vehicles, such as Branston’s generalized queuing model [72];
models, which identify groups of vehicles that share some properties, often using size and velocity as principal
characteristics [73]; and gas kinetic models, which use velocity distribution functions to model traffic flows (e.g. see
[74]). Hoogendoorn [75] unifies a number of gas-kinetic approaches to propose a generic model that describes the
dynamics of vehicles based on the distinction between free-flowing versus platooning.
A number of microscopic models have been proposed in the literature (for a more extensive review, see [67]).
Here we review a few microscopic models based on cellular automata. Cellular automata approaches have been
especially useful for self-organizing systems, and their relatively simple principles offer an opportunity to model large,
decentralized systems of cells, where the cells interact with their neighbors and adapt their behavior.
Nagel and Schreckenberg [76] present a cellular automata model of highway traffic flow (NaSch) in which a
lane is represented by a one-dimensional lattice (i.e. a row of connected points). Each of the lattice sites represents a cell
that can be either empty or occupied by at most one vehicle at a given time. This is a complex self-organizing system in
which competing forces are at work keeping the system dynamic, but also organized. The rules incorporate acceleration,
deceleration, randomization, as well as the actual movement of the vehicle.
In the NaSch model, vehicle motion is not deterministic; there is randomness due to unique driver reactions to
traffic conditions, which adds a layer of complexity to the dynamics. The model suggests that, when both critical density
and randomness are reached, spontaneous traffic jams will form. An interesting feature of this model is the flux density
curve. The flux, or flow rate, is simply the number of vehicles per second at a particular point. At low densities, vehicles
are free from interaction and can move at any speed. At high densities, vehicles must follow a strict order, as there are
velocity oscillations due to reduced space, in some cases with vehicles having to stop. At medium densities, the
interaction becomes most varied. Vehicles are close enough to interfere with one another, but not close enough to induce
a high degree of order.
Biham, Middleton and Levine [65] present a model (BML) based on CA and a two-dimensional plane. In this
model, a square lattice represents East-West and North-South streets. The spaces on the lattice represent intersections.
The rule in this model is very simple: a vehicle moves forward by one lattice if and only if the site in front is empty. At a
non-zero density, a phase transition can occur in which the speed of the vehicles discontinuously approaches zero. This is
due to a traffic jam so complete that it blocks all vehicles in the system. The BML model assumes random initial
distribution of vehicles, but subsequently provides a fully deterministic representation.
It is possible to combine the NaSch model, which describes traffic flow, with the BML model, which includes
traffic control representation at intersections. A transition from the freely flowing phase to the completely jammed phase
has been observed in such models at a vehicle density that depends on D (the number of points in the linear lattice for the
NaSch model) and T (the traffic light cycle length). The models are clearly simplifications, and purposely leave out such
details as the number of lanes, traffic light configurations, dedicated turn lanes, and left and right signals. Nagel and
Schreckenberg's demonstration that, at certain densities, traffic patterns change from laminar flow to start-stop waves has
also been used to improve traffic control systems. Krauss, Wagner and Gawron [77] (1997) present a generalization of
the NaSch model characterized by a family of models with one parameter r (the time steps that cars take to come to a
complete stop from maximum velocity). Their model properly accounts for the discontinuity and the different densities
that occur when traffic breaks and a jam occurs [77].
Chowdhury and colleagues [78] have examined the implications of both the NaSch and BML models, and note
that one cannot assume that traffic is uniformly distributed; if it were, traffic light optimization would be the same for all
lights. Instead, some roads dominate others in terms of traffic load, which influences local conditions.
Rickert, Nagel, Schreckenberg and Latour [79] provided an extension of the initial NaSch model to include
multi-lane traffic. Fouladvand and colleagues [80] modelled traffic flow on roundabouts. Here, traffic is controlled by a
self-organized scheme in which lights are absent. This controlling method incorporates a yield-at-entry strategy for the
approaching vehicles to the circulating traffic flow in the roundabout, also using the stochastic cellular automata
approach of the basic NaSch model.
An interesting experiment by Brockfeld and colleagues [81], examining traffic at a single intersection, reveals
that throughput is a function of cycle length and reveals an oscillatory pattern. Their research also shows that
experimenting with green wave synchronization of multiple lights is not likely to achieve any additional increase in
throughput. Additional experiments with green wave synchronization were conducted by allowing a phase offset
between light cycles, allowing minimal delay of traffic flow due to red lights. The outcome, as expected, is very similar
to the previous result for a single intersection in terms of flow rate versus traffic light cycle length. These experiments
also indicate that random traffic light cycle offsets, for low cycle times, outperform green wave traffic flow.
Burghout and Wahlstedt [82] investigate a hybrid traffic simulation model that integrates VISSIM (a
microscopic model) and MEZZO (a mesoscopic model). The hybrid model is applied to a network in which MEZZO
simulates the entire Stockholm area (6000 links), and VISSIM simulates a small area of interest containing three
intersections with adaptive signal control and bus-priority functions. The control and functions are simulated by a
separate signal controller that would take place in the field. They evaluate a fixed-time and adaptive control and show
clear improvement in terms of travel times, delays and stops. They see additional traffic in the local area but an overall
reduction in congestion, demonstrating the merit in simulating local and global traffic grid simultaneously. A self-
organized system will similarly model local control and evaluate global effect on the entire grid. Winters et al. [83]
examine the Python-based Lightweight Intelligent Traffic Simulator (LITS). The LITS is a vehicular network simulator
that models vehicle to traffic light communication with a goal of providing a test engine for other researchers to simulate
their algorithms.
Considering the number of factors that should be taken into account when modelling urban intersections, and
the difficulty in properly calibrating and validating models, Pretorious et al. [8] recommend evaluating intersections
simply in terms of volume/capacity ratios, or the saturation level, and carrying out improvements on the basis of
minimizing this ratio. Similarly, Akgungor and Bullen [84] suggest that saturation measures are the most important to
assess delay. Delay in turn serves as the most important measure to assess quality of service at an intersection. Akgungor
and Bullen [84] propose improvements in traffic modeling by presenting a methodology for determining the delay
parameter k as a function of the degree of saturation, rather than using a fixed value. In related work, Balmer, Axhausen
and Nagel [85] present a flexible agent-based modeling approach that allows for a variety of input data to create large-
scale transportation simulations. Suzuki et al. [224] introduced an Ising-like traffic model, where the state of traffic
signals interacts locally and leads to chaotic dynamics.
Rosenblueth and Gershenson [86] provide a CA model that is simple yet realistic, as it takes into consideration
the effect of traffic lights. Many of the basic approaches to urban traffic modelling, including the NaSch and BML
models, have seen updates to include additional realistic conditions of traffic (e.g. see [78]). Moreover, it should be noted
that the performance of the testing tools and platforms may vary depending on the topography, traffic density, data
available, and user expertise. They also vary in the interfaces used and the optimization techniques employed, but they
all tend to use standard traffic models available in existing literature.
This section reviews the large body of work on traffic signal synchronization. It focuses specifically on the
analytic techniques branch of the Traffic Signal Control Taxonomy presented in Figure 1, namely: dynamic
programming, control theory, fuzzy logic, machine learning, and queuing theory. Each of these elements will be
discussed in turn in this section.
3.1 Dynamic Programming
Dynamic programming was introduced by Bellman [87] as a way to solve multistage decision problems. It is a
way to reduce the problem into simpler sub-problems, and then combining the solutions to get the solution to the original
problem; the approach examines all possible ways to solve a problem and then selects the best solution. This is a time-
consuming and slow approach, but one that guarantees an optimum solution. Often, algorithms that do not require a
brute-force approach like a greedy algorithm are preferred to reduce the computational burden.
Dynamic programming has been applied at least since the 1970s with DYPIC [48], and the 1980s with the
PRODYN algorithm [49]. More recently, Heung et al. [27] introduced dynamic programming for signal optimization at
intersections with a fuzzy-logic controller. The assignment of green time to each phase of a traffic cycle is considered a
multistage control problem with a finite number of possible actions at each stage. A dynamic-programming technique is
used to derive the green time for each phase in a traffic cycle. Projected traffic flow from adjacent intersections is
incorporated into the analysis to ensure that control is extended beyond a single junction. The controller does not require
complex hardware, and the simulation results show that the delay per vehicle can be substantially reduced, even when
traffic demand reaches capacity. This bodes well for a self-organization approach, where information from adjacent
intersections is considered during the optimization process.
Yi et al. [88] use dynamic programming to optimize flow at intersections during periods of oversaturation.
They also address the problem of detection in oversaturated conditions when the queue spills past the detector. The
authors add a detector that counts vehicles leaving the intersection, allowing for a better estimate of queue size. Dynamic
programming leverages the fact that if a state-action is optimal, then removing the first state and action still leaves an
optimal state-action sequence [89]. This is known as the Bellman Optimality Principle.
In other research, Cai [90] and Cai, Wong and Heydecker [91] build on the work of Allsop [92], and provide an
Approximate Dynamic Programming (ADP) framework in which control parameters can slowly change over time
through learning techniques. ADP addresses the computational burden required from dynamic programming, as well as
the need for complete information by simplifying the cost-to-go function. A machine learning technique, such as
reinforcement or perturbation learning, can be employed to adjust the functional parameters.
3.2 Control Theory
Traffic signal optimization can be framed as a control problem. However, application of control theory can
vary based on the input parameters and the traffic parameters being controlled. Classical control theory involves a
closed-loop system with sensors and feedback. The feedback is used to control states or outputs; input parameters (e.g.,
the number of cars waiting at a red light) have a direct effect on control parameters (e.g., adjusting traffic light cycle
time). The input parameters are measured with sensors, and are processed by the controller. The result (the control
signal) is used as input to the process, thereby closing the loop. In an open-loop system, there is no sensing or feedback
from output, which makes it less robust than a closed-loop system. An example of an open-loop traffic light system is
one in which historical traffic patterns are used to set the timing parameters; as long as the traffic flow follows the
historical trend, the system performs well. However, variations from the norm, which can be expected, can lead to
significant performance deterioration. A closed-loop control system will typically use both historical traffic data and real-
time traffic conditions to make decisions.
Shimizu and colleagues [93] frame traffic signaling as a classic feedback control system to minimize
congestion at intersections. The input is the number of cars approaching the intersection, and the control variables are
cycle length, green split, and cycle offset. Oversaturation and queue spillback both adversely impact the performance of
the control algorithm. Wey [94] uses the Proportional Integral Derivative (PID) for signal control to minimize traffic
delay. Ball et al. [95] have developed the H method to determine optimal signaling control for an intersection. The
reference point keeps the queue length below a predefined maximum value. This technique assumes some means of
observing the vehicle flow rate. The effective green time and total cycle length are the control variables. The basic of the
H features are presented in an article by Kwakemaak [96]. Feedback control may also be used to order traffic flow.
Yang and colleagues [97] apply linear feedback control to signal setting, using reserve capacity and degree of saturation
as the control objectives.
Another nonlinear control optimization technique called adaptive fine-tuning addresses the problem of
efficiently determining an optimal value by evaluating the objective function on random perturbations around the current
control parameter vector [98]. The results are similar to gradient descent. This fine tuning algorithm has been applied to a
traffic signal control strategy called traffic-responsive urban control (TUC), which allows for control of green times,
cycle times and offsets; it also allows for the provision of public transport priority, and uses the store-and-forward
concept described below.
In the work of Aboudolas and colleagues [99], a linear-quadratic optimal control is applied to the store and
forward model (SFM)not to be confused with the notion of store and forward in computer networks. Aboudalas and
colleagues' model achieves a significant reduction in complexity by making the discrete time step equal to the traffic
light cycle time. This means that the model only samples the average output from a full traffic cycle, and ignores the
intra-cycle timing and queue variations, thus simplifying the model at the cost of some high-resolution variation.
Pecherkova and colleagues [100] meanwhile, focus on state and parameter estimation techniques for a traffic signal
control system, examining two techniques: the unscented Kalman filter and divided difference filters.
3.3 Fuzzy Logic
The concept of fuzzy logic was first introduced by Zadeh [101] in “Fuzzy Sets.” Fuzzy logic allows for
uncertainty in representation of information by allowing partial membership of sets. It mimics human perceptions in
control logic ,making it effective in representing human behavior. Pappis and Mamdani [102] applied fuzzy logic to the
analysis of traffic control. Their model assumes that vehicle loop detectors are placed upstream of the intersection to
measure approach flows and estimate queues. This information is used to determine whether to extend or terminate the
current signal phase. Each possible extension is assigned a degree of confidence, and the extension with maximum
confidence is implemented. Before the implementation ends, the rules are reapplied to see if further extensions are
The goal of fuzzy logic is to approximate expert perceptions. A fuzzy controller consists of a system that
contains fuzzy rules, a database that collects traffic data, and an inference engine that analyzes the rules. Initially, fuzzy
systems used a static set of rules, but more recent versions use evolutionary principles (e.g., genetic algorithms) to refine
the set of rules governing traffic flow [103]; [104]; [22]; [105]; [106].
Modern fuzzy systems assess the fitness of the initial rule set and select the most appropriate for propagation
into the next generation. The genetic operators of crossover and mutation are applied to those selected, and a new
generation of rules is obtained. With each generation, the average fitness of rules is expected to improve, resulting in
better overall performance. Niittymacki and Kikuchi [107] use fuzzy logic to control a pedestrian crossing signal, while a
number of other researchers have investigated decentralized approaches that incorporate self-organization concepts and
fuzzy rules [23]; [108]; [109]; [110].
3.4 Machine Learning
The simulation and optimization of traffic light controllers using an adaptive optimization algorithm based on
reinforcement learning is analyzed by Wiering and colleagues [111], as well as Cai et al. [91] as explored earlier.
Wiering and colleagues examine a traffic light simulator that enables experimentation with different infrastructures and
compares different controllers. Their experimental results indicate that the adaptive traffic light controllers outperform
other fixed controllers in all infrastructures.
Neural networksa type of machine learning techniquehas been used to improve the quality of fuzzy rules
for traffic signal optimization in a number of studies [112]; [113]; [114]. Researchers train the network to adapt to
different traffic flow conditions based on real-time data. Elements of dynamic programming have also been added to
form a neuro-fuzzy controller for improving the performance of the traffic optimization model across multiple networks
A hierarchical architecture that combines neural networks, fuzzy logic, and genetic algorithms to create a multi-
agent architecture system has been constructed by Choy and colleagues [116] [117]. The authors divide the primary
problem into sub-problems, each of which is managed by an intelligent agent using fuzzy neural decision-making rules.
A reinforcement learning process is implemented to improve the quality of decision making. Zhu et al., [118] use a radial
basis function neural network to forecast traffic volumes at intersections. The unique contribution that they have is that
they provide information about adjacent signals (15 min intervals) to the neural network for improved accuracy. The
question remains how this approach will work across different settings, especially those where volumes vary
significantly across multiple intersections.
Lu [119] has demonstrated the applicability of incremental multistep Q-Learning to adaptive traffic signal
control. Q-learning is unsupervised learning, and based on trial and error. Rather than being presented with a large set of
training examples, the Q-learning agent essentially generates its own training experiences from its environment. Ozan et
al. [120] have used Q-learning in conjunction with a TRANSYT traffic simulator and shown improvement compared to
Hill Climbing and Genetic Algorithms. Their benchmark using standard test cases demonstrates the feasibility of using
reinforcement learning in real-time traffic applications. Zhu et al. [58] develop the Junction Tree Algorithm (probabilistic
graph-based algorithm), which is another reinforcement learning algorithm that outperforms independent learning (Q-
Learning), real-time adaptive learning, and fixed time algorithms across multiple performance metrics. Their simulation
was done on 18 signalized algorithms in VISSIM.
3.5 Queuing Theory
Pioneering work in queuing theory started with Webster in 1956, and has since evolved into a widely accepted
analytical tool for traffic modeling. Queuing theory applies to the queue of cars that form at the intersections. The
analysis of traffic flow for signaling has taken two general approaches: a deterministic fluid flow component assuming
continuous variables and steady flow; and a stochastic component comprised of probabilistic events and their impact on
queuing. Queuing theory works well for steady state, but becomes problematic when the traffic flow rate exceeds
capacity. Another complicating factor is the lack of independence of traffic lights. While standard queuing input rate
assumptions can be assumed at an isolated signal, systems of interconnected signals impact one another in ways that
violate the Poisson assumptions [121].
Queuing formulae for traffic control can generally be separated into an under-saturated, or uniform, queuing
component and an oversaturated, or stochastic, queuing component [9]; [122]. The uniform component assumes that
traffic arrives and departs in a uniformly distributed manner. Yet, since real traffic is not uniform, a stochastic
component must be included in the analysis. When the load approaches capacity, the stochastic effects tend to dominate.
During transitional periods, the standard deviation can be of equal magnitude than the mean.
Beckmann and colleagues [123] provided the first queuing derivation for traffic signal delay under the
assumption of binomially distributed arrivals and uniform service rates [123]. Webster [124] applied a different formula
comprised of three components: (1) an analytical derivation of uniform delay; (2) a characterization of stochastic delay
analytically derived assuming Poisson arrivals and deterministic service rate; and (3) an additional term used to match
the analytical results with those observed from simulation data. Webster’s delay is still widely used and continues to
serve as the benchmark for traffic signal control.
Micro-simulation models are normally used to evaluate traffic-adaptive signal control systems. Lehmann [125]
developed an analytical approach to this evaluation based on queuing models. In particular, he investigates a queuing
model for a simplified adaptive control strategy based on a rolling horizon scheme, in which a signal serves two
movements alternatively. A numerical algorithm is developed in a stochastic context to compute steady-state
performance measures such as average delays and expected queue lengths. These results are compared with simulation-
based results; the analytically derived numerical method predicts the simulation results quite well. Foulandvand, et al.
[225] proposed an early decentralized approach for regulating traffic lights based on queues. However, their local
optimization failed to improve traffic globally. Still, a similar approach was later used by self-organizing methods
(detailed in Section IV), which successfully improved global traffic.
Tian [126] proposes two models that estimate the capacity of an intersection using actuated control: the
Minimum Delay Model and the Hybrid Model. In these models, the capacity of an approach to a lane group of the
intersection is a function of the saturation flow rate, the green time allocated to this approach or lane group, and the cycle
length of the intersection.
The Minimum Delay Model estimates the green times and cycle lengths from flow rates, minimizing total
delay. Parameters in this model include the ratio of green extension period to queue service time specific to each
approach or lane group. The parameters depend on the distribution of arrivals at the intersection. The Hybrid Model
combines the deterministic queuing model that estimates queue service time, and a theoretical model that estimates the
green extension period from the unit extension, the flow rate, the speed limit of the approach, and the detector length. A
method converting the left turn traffic volume to equivalent through volume is developed. The method is used to
estimate the capacity of intersections which permit left turn phases.
The Minimum Delay Model and the Hybrid Model are validated at the intersection level by comparing the
estimations of effective green ratios with those simulated by MITSIMLab. These two models have been validated at the
network level with real data. The results show that both the Minimum Delay Model and the Hybrid Model are
appropriate for estimating capacity of intersections with actuated control. The Minimum Delay Model is also suitable for
estimating capacity of intersections with adaptive control. The Emulation Model is applicable to off-line mesoscopic
dynamic traffic assignment.
A technique loosely related to queuing theory, known as shockwave analysis, uses the notion of vehicle arrival
rate, but differs in the notion of service rate. Instead of the paradigm of a server and service rate with an associated queue
for vehicles waiting for service, vehicle density and flow rate changes are monitored. The ratio of change in flow rate to
change in density yields a wave velocity that must have occurred to account for the changes in flow rate and density.
Shockwave analysis uses an intuitive, physics-based derivation, while queuing theory utilized a more computer
communication network or operations research-based derivation. Yeon and Ko [127] discuss the relationship between
shockwave analysis and queuing theory, and attempt to experimentally validate the comparison using a real traffic flow
study. Neither approach appears to be consistently more accurate than the other in the study. Le, et al. [226] propose a
decentralized BackPressure scheme, where queue length measurements “press” on downstream queues. In this way,
streets with a higher demand will obtain higher allocated green times within fixed cycles.
Freeman et al. [128] benchmark different tools available for traffic optimization, including Highway Capacity
Software, PASSER, SYNCHRO, SimTraffic, TRANSYT, and CORSIM. Their work focuses on accurately predicting
delays, queue lengths, spillback, and levels of service. Sensitivity analyses are performed to evaluate limitations. Their
aim is to identify the requirements for an “acceptable technique” to aid in the design and evaluation of signalized
intersection systems.
The fundamental limitation in previous work is both the choice of techniques that are used, and the formulation
of the problem. Formulating the traffic signal optimization as a centralized control problem does not lend itself to a
solution that is either robust or scalable. As the number of variables grows, the complexity of the problem increases,
making it infeasible to solve. Also, the techniques that rely on static data based on historical traffic patterns are unable to
capture the instantaneous variability of the traffic flow, making the solution suboptimal. In addition, the techniques are
not robust to traffic anomalies, such as sudden influx of traffic or the disruption of an artery. As an alternative to
overcome these limitations, we examine the use of self-organization as a viable option for traffic light coordination.
There are several instances in nature in which groups of organisms work together to collectively accomplish
complex tasks that they would otherwise be incapable of performing individually. In these instances, without the need of
central control, each organism has the capacity to coordinate with its neighbors, leading the group to function seamlessly
in a well-organized fashion. Consider, for instance, the ant. It is a simplebiological organism, in the sense that its
behavioral repertoire is limited to 10-40 elementary behaviors. In groups, however, ants exhibit sophisticated collective
behavior including the ability to build rafts and bridges by tangling their legs together. None of the insects grasps the big
picture, but collectively they can perform sophisticated tasks that contribute to the success of the overall colony. This
strong integration has led some biologists to consider the colony as an organism in itself.
The secret lies in self-organization: individuals follow simple rules based on local information, collectively
completing complex tasks. Such behavior is also called swarm intelligence [129]. This behavior is not only observed in
social organisms, but also in physical phenomena such as the crystallization of liquids [130] and the alignment of
magnetic spins [131]. Self-organizing techniques leverage the interaction of many similar components. Each component
follows a set of rules and adapts its behavior by sensing its neighbors and/or local conditions. Simple behavior at a local
level leads to complex collective outcomes; the neighbor-to-neighbor interaction leads to global coordination of the
entire group. Self-organization may thus be defined as an increase in order without outside control.
A rich literature has developed around the general concept of self-organization [132]; [133]; [134]; [135];
[136]; [37]; [137]. In studies of self-organized systems, entropy is used as the most common definition of disorder. The
usual approach is to look for conditions that appear to violate the second law of thermodynamics, which states that
systems in closed systems evolve to reach thermal equilibrium, or maximum disorder. Such a violation was posed by
Maxwell’s demon in relation to adiabatic computing. Self-organizing systems do not violate the second law of
thermodynamics, and thus must constantly ingest matter or energy with low entropy and expel the high entropy as heat.
The information-theoretic view of the second law may be stated as the fact that every system tends to its most probable
state [138]; [139].
4.1 Philosophical and Practical Challenges
Self-organization in multi-agent systems appears to contradict the second law of thermodynamics. This paradox
has been explained in terms of a coupling between the macro level that hosts self-organization (and an apparent
reduction in entropy) and the micro level (where random processes greatly increase entropy). Metaphorically, the micro
level serves as an entropy “sink,” permitting overall system entropy to increase while sequestering this increase from the
interactions where self-organization is desired. Wiering and colleagues [111] make this metaphor more precise by
constructing a simple example of pheromone-based coordination, defining a way to measure the Shannon entropy at the
macro (agent) and micro (pheromone) levels, exhibiting an entropy-based view of coordination
There are many philosophical open questions related to self-organization that restrict its use in engineering. For
example, it is the observer who ascribes properties, aspects, states, probabilities and, therefore, orders to the system. But
organization is more than just low entropy: it is structure that has a function or purpose. Thus, Stafford Beer (1966) made
a very important observation when he noted that what under some circumstances can be seen as organization, under
others can be seen as disorder, depending on the purpose of the system [138].
One of the most common problems associated with self-organizing systems is emergence. Self-organizing
systems typically have higher-level properties that cannot be observed at the level of the elements, and that can be seen
as a product of their interactions (more than the sum of the parts). Emergent properties are a hallmark of complex
systems. The problem here is ontological. According to classical thought, there is only one “true” description of reality.
In this case, a system cannot be a set of elements and a whole with emergent properties at the same time. Still, there are
alternatives to classical thought that allow an understanding of emergence [140].
One solution is to distinguish between an absolute “god-like” being that observes the system, and mere mortals
who possess only limited perception. While this kind of discussion is philosophically interesting, it is of limited use in
engineering. At some point, one must choose a practical definition and stick with it. The one that we have chosen is
expressed by Gershenson as follows: “A system described as self-organizing is one in which elements interact in order to
achieve dynamically a global function or behavior.” [36]
Traffic grids have the basic components needed in a self-organizing system: traffic lights are distributed
through the grid; there is connectivity between intersections via streets; and there can be local communication (at least in
terms of cars flowing from one intersection to the next). Consequently, traffic optimization can benefit from self-
organization models such that traffic signals communicate with their neighbors and adapt their signal timing accordingly.
Traffic signals adapt their behavior based on a universal set of rules using local information, and all the intersections can
be coordinated as changes cascade through the entire grid.
4.2 Techniques for Self-Organized Traffic Control Systems
There are several techniques on the basis of which self-organized traffic light control systems can be created.
Some of the most effective self-organization models are built using a set of heuristic rules developed from observation of
natural phenomena, that are then abstracted and adapted to the context of a specific problem (e.g. [141]). The goal is to
come up with a set of rules to govern the behavior of individual components of a self-organizing system that will result
in an overall coordination of the entire system. Ideally, we would like to invert the problem and derive the set of rules
based on the desired outcome [142]. However, there is no closed-form solution, and the heuristic rules are created based
on an initial intuition and then refined based on observation using trial-and-error.
There have been many formal ways to approach the problem. One formal technique that has been used to
investigate complex interaction networks is graph spectral analysis, which relates the network topology of the road
system to fundamental mathematical properties such as connectivity and maximum flow. Eigenvalues of the adjacency
matrix of the road network produced in this type of analysis have many important properties that have not yet been used
in traffic signal synchronization. Pulse-coupled oscillation (PCO) [143] and Markov Random Fields (MRF) [144] are
two other self-organizing techniques. PCO is the mechanism behind firefly synchronization; fireflies adjust the timing of
their flashes to the energy pulses of their nearest neighbors. Our own research has examined the possibility of using PCO
to provide synchronized, low-power time pulses for sensor networks. Similar techniques may be used to achieve new
types of green signal waves, as proposed by Faieta and Huberman [145]. The use of MRF, meanwhile, achieves global
coordination by minimizing local energy functions. This technique comes from the Ising model used in physics for
applications such as image processing. Spatial dependencies can be utilized to fill in missing or corrupt images. MRFs
are also used in forward error correction in communications. Similar techniques might be used to set traffic light timing
based upon spatial configurations.
Self-organized traffic-flow models have been investigated using cellular automata [76], swarm algorithms
[146], coupled phase oscillation [147], statistical physics [148], [149], fluid dynamics [150], queuing theory [151], multi-
agent simulations [36], and fuzzy logic [23]; [24]. Similar complex systems techniques have also been applied to
modeling pedestrian flows [152] and traffic jams [153]. We discuss these and other works in greater detail in the
following subsections.
4.2.1 Heuristics-Based Algorithms
Current advanced traffic management systems (ATMS) use learning methods to adapt phases of traffic signals,
normally using a central computer. This self-organizing approach is inherently distributed, as the global synchronization
is adaptively achieved by local interactions between vehicles and traffic signals, generating flexible green waves on
demand. The notion of green wave [154] is popular with traffic engineers: phases and periods of light signals are chosen
such that a series of lights is coordinated to allow continuous traffic flow through multiple intersections, typically in one
or two directions, as there are mathematical constraints to achieving full coordination in all directions. Green waves can
be effective if traffic velocity is relatively constant, and most traffic goes in the same direction at the same hour.
However, the optimal synchronization of traffic lights changes with vehicle density: as traffic density increases, vehicle
speeds tend to decrease. In this way, different coordination schemes are required for different densities. To resolve these
inefficiencies, numerous attempts have been made to synchronize traffic lights by using a central computer to collect and
analyze data and adapt to real-time traffic conditions [39]; [155].
There are several problems with this approach, including latency and poor scalability. Current ATMS use
learning methods to adapt traffic light phases with the help of a central computer. Still, this adaptation tends to be much
slower (at the scale of minutes or hours) than the change in demand (at the scale of seconds). The self-organizing
approach discussed here does not need a central computer, as global synchronization is adaptively achieved by local
interactions between cars and traffic lights, generating flexible green waves on demand. Cools and colleagues [156]
dispute the efficacy of green wave methodsincluding ones that have been tested in a number of simulationssuch as
the Green Light District/iAtracos project. Simulation results from Belgium indicate that a self-organizing controller
performs better than the green wave controller. Further work has confirmed this conclusion [157]; [158]; [159].
Gershenson [36] and Gershenson and colleagues [156], have written extensively about self-organizing systems,
specifically self-organizing traffic control systems. One aspect of optimal control that features prominently in their work
is wavesa sequence of traffic lights in a single direction that allows for continuous flow through multiple intersections.
Appropriate phases and periods that optimize traffic flow can be readily selected if a constant traffic velocity is assumed.
However, as traffic density increases, speed tends to be reduced.
Gershenson also addresses self-organization and complexity from fundamental principles [36]; [160]. The
algorithm he proposes is very simple and does not require direct communication, just the ability of each traffic signal to
count the number of vehicles approaching the light while it is red [161]; when this count reaches a threshold value, the
green signal cycles to yellow then red. The red signal resets the count to zero and proceeds as before. This tends to
encourage larger groups of vehicles (convoys) to flow together. In a sense, this acts like congestion control in a
communication network in which the maximum delay-bandwidth product of a link is utilized. There also tends to be
larger spaces between the convoys, thus allowing for cross traffic, and there is a time threshold so that the light does not
change too quickly during periods of high traffic density.
Two further conditions are taken into account to regulate the size of convoys. Before changing a red signal to
green, the controller checks if a convoy is crossing through, in order to avoid breaking it. More precisely, a red light does
not change to green if, on the crossing street, there is at least one vehicle approaching within a pre-specified distance of
the intersection. This rule keeps crossing convoys together.
For high densities, this condition alone would cause havoc, since large convoys would block the traffic flow of
intersecting streets. To avoid this, a second condition is introduced. Condition one is not taken into account if there are
more than a specified number of vehicles approaching the green light. With this rule, long convoys can be broken up if a
convoy will soon be through an intersection. The metric used in this approach is average trip waiting time, which is the
travel time minus the minimum possible travel time averaged over all vehicles. Note that the algorithm parameters
mentioned previously had to be determined manually. Vincent and Young [50] adopted a similar self-organizing
technique decades before for isolated intersections. This algorithm has recently been refined with further rules which
increase the performance for very low and very high densities, bringing them close to the theoretical optimum for all
densities [158]. Burguillo, et al. [227] extended [160] to propose history-based self-organizing traffic lights. In this
method, the duration of the next green cycle of every light is directly proportional to the number of cars which crossed
the intersection in the last green cycle. Results showed slight improvements over [161].
Particle Swarm Optimization (PSO) is a meta-heuristic algorithm that does not require computation of gradients
[162]. It changes the solution stochastically and uses an optimization function to evaluate the objective function. The
algorithm has the capability of navigating local optima in the search space. However, there is no guarantee of a global
optimum. Chen and Xu [33] apply PSO to traffic signal timing by deploying a local fuzzy-logic controller (FLC)
installed at each intersection. These controllers provide some initial solutions for the algorithm; coordination parameters
from adjacent intersections are incorporated, and the membership functions and rules of the FLC are optimized. They
optimize the average delay and average number of stops for adjacent intersections. Simulation results show that the delay
per vehicle can be substantially reduced under both constant and time-varying traffic demands, particularly when
upstream demand exceeds downstream demand. The implementation of this method does not require complicated
hardware; its simplicity makes it a useful tool for off-line studies and real-time control purposes. The rules, however,
need to be made universal to work under different scenarios. Kammoun et al. [163] present an optimization algorithm for
route guidance for vehicles based on GPS information of vehicles that incorporates fuzzy logic in decision analysis. In
their formulation the road network is hierarchically organized with multi-level decision analysis.
Tatomir and Rothkrantz [164] propose a dynamic traffic routing algorithm based on Ant Colony Optimization
(ACO) model. They use a routing table with potential split rates of traffic flow; at each network node the traffic is split
into different arteries based on the split rates defeined in the table. Using a central table to define the splitrates results in
poor scalability as the network size grows. Alves et al. [165] use ACO for traffic optimization for single-origin single-
destination networks in which they use a static traffic model with time-invariant traffic conditions. Cong et al. [166]
present their ant colony routing algorithm for redirecting traffic at intersections, and address the limitations that they
perceive in the classic ant colony algorithm, i.e. inability to distinguish flows from different origins, inability of
controlling distribution of vehicles along routes, inability to constrain flows or add link costs to routes. They incorporate
pheromones and multi-colored ants to control distribution of traffic along links, and to distinguish between different
traffic flows. Cesme and Furth [167] use self-organization to adapt traffic lights, including rules for extending green
phases for arriving platoons (useful for long corridors) and a dynamic coordination to synchronize signals that are close
to each other. They tested their methods in realistic scenarios with heterogeneous demands, showing that self-
organization can reduce transit delay.
Wang, et al. [228] propose a control method that decomposes the network optimization problem into
overlapped and interactive problems between “basic coordination units.Intersections use a forecast model, where not
only the local performance is considered, but also that of the downstream intersections. Placzek [229] considers
heterogeneous traffic flows and proposes a system where agents at intersections use models to predict the cost of control
actions using the interval microscopic model, which combines cellular automata and interval arithmetic. After control
actions are evaluated, the best is executed. Scenarios with road-side vehicle detectors and vehicular sensor networks are
evaluated, considering heterogeneous traffic. The proposed mechanisms are shown to reduce delays compared with [160
and [185]. The advantage of Placzek method lies in the fact that it can adjust parameters in a heterogeneous fashion,
while other methods use global parameters.
4.2.2 Cellular Automata
Cellular automata (CA) is another approach to model self-organizing systems that has been used to study traffic
signal behavior and control since the 1950's [76]. Cellular automata models are described in greater detail in section 2.3
in the context of urban traffic modeling. Studies that have used CA with a focus on traffic signaling include [86]; [168].
Gershenson and Rosenblueth [157] use a CA model to examine traffic signaling and self-organization in relation to
the problem of static-optimization of constantly changing traffic behavior. A macroscopic two-dimensional cellular
automata model of an urban traffic signal control system, in which each intersection is regarded as a cell, and the flow
pressure is treated as the state, is analyzed by Wei and colleagues [169], and Fouladvand and colleagues [80]. Wei and
colleagues, using the BML model, simulate and analyze a two-dimensional traffic system controlled by traffic lights; Wu
and colleagues [170] take a similar approach. Yang and colleagues [171], and Shoufeng and Ximin [172] examine travel
and route guidance.
Shoufeng and Ximin combine a Hybrid Algorithm with a cellular automata simulation to calculate travel time
and optimize signal settings. Paz et al. [173] integrate a genetic algorithm to search for a global optimum, and couple it
with simulated annealing to get the best local optimum. While the results were quite similar to Simultaneous
Perturbation, Stochastic Algorithm combination of a global and local optimum approach reduces the computational
effort significantly. Kumar and Mitra [174], meanwhile, use a cellular automata model to examine what happens at an
intersection when traffic signals malfunction. By modeling individual vehicles as agents, they were able to replicate the
surprisingly well-organized traffic flow observed at a malfunctioning intersection in India; counter-intuitively, the
behavior that normally causes jams induced traffic to flow smoothly.
In the Cell Transmission Model (CTM), a highway is partitioned into a series of cells, the densities of which are
determined by vehicle concentration [150]. Flotterod [175] and Tampere and Immers [176] modify this idea by applying
a Kalman filter to the model. The problem of multi-intersection control using the CTM and hierarchical control is
addressed by Chen and colleagues [177], [178], who distinguish between feeding delay (the delay of vehicles entering a
saturated cell), and nonfeeding delay (the delay of all other vehicles). Their proposed solution is formulated as a
conflicted multi-objective optimization problem. In this type of problem, a single optimal solution may not exist. Instead,
there may be a family of non-dominated or non-inferior solutions. These are the best solutions that can be obtained
among the alternatives, such that increasing any one objective would degrade another. This is also known as a Pareto
front, or Pareto optimal solution.
In the context of urban traffic modeling, it is interesting to note that bottlenecks create a situation similar to that
described by self-organized criticality. Self-organized criticality (SoC) is a property of dynamic systems that have a
critical point as an attractor. In other words, the system state naturally tends to move towards the attractor state, and that
state, being at a critical point in the system, is near a phase transition, thus keeping the system in a dynamic, but self-
organized situation. Thus, there is no need to precisely tune the system parameters to maintain this situation [179]. A
famous example is the sand pile, in which, as sand is added to the pile, it will maintain a minimally stable state until it
eventually collapses.
In summary, cellular automata is one of many techniques from statistical mechanics used to assess the micro-
and macroscopic behavior of vehicles in traffic signaling. Many useful fundamental results have been obtained from
these analyses. Current traffic engineering techniques rely on a combination of queuing theory and empirical techniques,
and statistical mechanics approaches have not been fully integrated with these. As intelligent transportation becomes
more mature, the need to incorporate these techniques will become increasingly significant.
4.2.3 Neural Networks
Neural networks are designed to observe and learn patterns using an architecture similar to that of brains, where
multiple neurons are interconnected and learn to fire at different strengths in response to different environmental or
internal stimuli, creating different impacts on the chemical and physical responses in the body. Similarly, an artificial
neural network contains multiple interconnected neurons that learn from prior experience using algorithms, allowing the
network to respond to previously unknown inputs. In effect, an artificial neural network builds its own regression model
based on data used for its training, and then uses the model to predict outcome based on new input. In case of traffic, the
model learns from the past traffic behavior and suggests an optimal timing sequence for traffic lights.
Ledoux [21] has used a back-propagation neural network to create a cooperation-based flow across multiple
intersections. Such cooperation has many of the hallmarks of self-organization. The basic characteristic of such a neural
network is that it can use data from past behavior to predict future behavior. Neural networks can be used to model
traffic flow either at a single intersection or through multiple intersections. This enables observers to predict queue
lengths and output flows one minute ahead of the decision point.
Nakatsuji & Kaku have also developed a self-organizing system using neural networks [20]. They trained a
back-propagation neural network to predict traffic behavior. Input measures of cycle length and signal phase splits were
used as inputs, with measures of effectiveness such as queue lengths or performance indexes as output. The authors
make claims for self-organization in their model; however, it is more adaptive in nature insofar as each signal is able to
adapt its behavior based on immediate conditions predicted by the neural network. There does not seem to be
cooperation among the nodes in terms of information exchange.
4.2.4 Utopian Solution
Given the rapid improvements in sensing and communication that have allowed autonomous driving, in the
future, knowledge of the location of cars, their routes, and velocities can be assumed. The cars are then akin to particles
flowing through a network structure, and the goal is to avoid collision. For any given set of cars and complete knowledge
of their paths, it should be possible to compute the optimum trajectory so that no collisions occur. For example, assume
no traffic signals exist, and allow the cars to proceed until they reach their destinations. Then, note how many collisions
would have occurred. These configuration estimates are typically based upon the requirement of maintaining a safe
distance between cars, below which a collision is assumed. It is only at these collision points that traffic signals are
required. Studies examining this idea include [180]; [181]; [182]; [230]
The question then becomes how to optimally apply signals. Should the cycle time of green to red be
proportional to the traffic that actually arrives? Or should it be proportional to the traffic that is predicted to arrive?
Prediction is feasible given that the shortest routes are assumed. One can further assume that vehicles will continue on a
course that is as linear as possible with respect to the road topology. With these assumptions, the use of self-organizing
traffic lights is not only feasible, but also very possible. Goel et al. [183] have developed an algorithm based on the
predicted arrival of vehicles through communication among neighboring signals. Sensors located at the entrance to an
artery detect a vehicle entering the artery, and relay the information to the next signal. Based on the congestion
(computed by total number of cars in the artery) and posted speed limits, they can compute the arrival time to the next
intersection. Based on the arrival times of vehicles at the intersection, an optimum signal time is computed. Goel et al.
[184] have also developed a process for retrofitting existing legacy controllers with a microprocessor to support the self-
organization algorithm. In contrast, Gershenson [36] relies on counting vehicles as they arrive at a traffic light, and does
not require communication across traffic lights. Helbing and Lämmer [185] suggest a self-organizing algorithm where
traffic conditions from the recent past are used to make future predictions to optimize the traffic signal timing, making
them adaptive in real-time. This information is collected via sensors at individual intersections and fed into a
microprocessor on the traffic signal to make optimal future signal timing decisions. Data collection and sensing are
obviously keys to improving the efficiency and design of self-organizing algorithms. To identify the amount of data that
needs to be collected, the impact of latency and precision of information on the overall performance needs to be
measured. There are issues of cost, reliability, and privacy that also need to be addressed. We discuss data issues in depth
in the next section.
5.1 Data Collection
Traffic data collection is critical to signal timing and is often performed manually. Intelligent transport systems,
on the other hand, continuously optimize signal timing and require high-quality real-time traffic data. A variety of
sensors can be used for real-time data collection, the most common of which is the inductive loop detector [186]; [187].
The detector consists of one or more loops of wire embedded in the pavement and connected to a control box, activated
by a signal ranging in frequency from 10 KHz to 200 KHz. When a vehicle passes over or rests on the loop, inductance
is reduced, thus indicating the presence of a vehicle. These detectors work in most weather conditions, but can be
unreliable and expensive. Ultrasonic sensors [188], microwave sensors [189], and infrared sensors [190] are also used to
monitor traffic, though these have limitations as well. Ultrasound sensors are vulnerable to pedestrian and vehicular
impact and have limited range (12m); infrared sensors perform poorly in noisy environments; and microwave sensors are
unable to detect slow or stationary vehicles. Consequently, loop detectors remain the most popular technology in vehicle
data monitoring.
Other new techniques for measuring traffic data include: radar [191]; [192]; cellular phones that record
motorists’ positions at check points [193]; Global Positioning Systems [194]; [195]; and RFID (radio-frequency
identification) receivers which can be installed at strategic locations and are already used for toll collection [196]. Several
other techniques are voluntary that require motorist cooperation. There is also a concern that some of these technologies
violate personal privacy, and allow government agencies to monitor the behavior of private citizens.
The use of video surveillance is considered particularly pernicious by privacy advocates [197]; [198]; [199].
Nevertheless, vision-based systems are flexible and versatile in traffic monitoring applications; improvements in
technology will likely continue to make these increasingly reliable and robust. Laser sensors used in autonomous
vehicles are very effective for detecting vehicles, but their cost is still too high to be installed massively.
A robust tracking algorithm for crowded intersections has been developed by Kamijo [200]. This algorithm,
based on the Spatio-Temporal Markov Random Field model, is robust against occlusion and clutter issues and can be
modified to deal with illumination variation. This system is capable of generating traffic event statistics such as vehicle
count, travel direction, velocity, and frequency paths.
With the advent of vehicular sensors there is a great potential for improving traffic efficiency. Sensor data can
enable real-time control of traffic and improve the performance of existing adaptive systems by allowing a more precise
adjustment of signal timing sequences.
5.2 Communication
There are two aspects of communication that are significant for self-organizing traffic systems: (1) sensor to
traffic light, and (2) traffic light to traffic light. Sensor to traffic light has existed for decades, and is fairly robust. The key
to self-organization is communication between traffic signals, which can be done using several different media such as
fiber optic, radio signals, microwave signals, etc. depending on the topography, weather, and other physical constraints.
While the technologies exist, there is a considerable cost in deploying such communication systems.
The vehicle to traffic light communication is rapidly evolving with considerable implications for traffic
efficiency improvement. Audi already has pilot studies in several cities, including all of Berlin, where equipped cars get
information from traffic lights and accelerate or reduce their speed depending on whether they will get a green light at
the next intersection or not [201]. Okada and colleagues [202] propose the use of visible light wireless communication
for vehicle-to-road communication at intersections. They recommend using a LED traffic light as the transmitter and an
on-vehicle high-speed camera as the receiver. They also present an algorithm for detecting the transmission. The authors
discuss a number of related factors, such as positioning of road signals and cameras, and how to achieve uninterrupted
communication between traffic signals and moving vehicles.
Lee and colleagues [203] have suggested using a modified blackbody radiation model to capture the effect of
ambient-light noise in the context of visible light communication. They study the impact of daylight on system
performance and demonstrate that it is possible to obtain relatively accurate results with a relatively modest effort. The
authors also introduce a new receiver structure with a selective combining technique that reduces the effect of
background noise by 5 dBs.
Vehicle-to-vehicle communication is also novel, with even greater implications for traffic optimization using
self-organization. Belinova [204] proposes the use of self-organization based on vehicle-to-vehicle communication. In
her model, each vehicle is considered an independent intelligent entity, and a driver makes decisions based on
information uniquely available to him/her from other vehicles and traffic signals. When the various reactions of
individual drivers are combined, self-organizing processes may arise and contribute to the improvement of traffic,
though the opposite is possible as well.
The Cluster-based Self-organizing Protocol, or CSP, was developed to organize and cluster vehicles [205]. This
protocol assumes that each vehicle has the ability to communicate with a central server and exchange messages at
intersections. Vehicles are clustered based on the information exchanged between them. This is somewhat analogous to a
Mobile ad hoc network (MANET), in which one node acts as the base station and performs additional overhead tasks. In
one MANET there are multiple service areas, with one sub-service area holding multiple mobile stations and one base
station monitoring these.
An example of vehicle-to-vehicle communication for self-organizing highway traffic was developed by Kesting
et al. [206]. A similar example was presented by Yang and Recker [207]. It is likely that this technology can also be
extended to coordinate vehicles through intersections, as explored by Ilgin-Guler et al. [208]. Pehofer and Bettstetter
[209] have explored self-organization based on ubiquitous wireless communication. They pose the fundamental
question: What are the design paradigms for developing a self-organized network function? We attempt to answer that in
the next section.
Even though self-organized systems fall under the general rubric of complex systems, their most desirable
attribute is that simple rules applied in a local neighborhood lead to sophisticated behavior globally. The key to this
outcome, however, is finding the right set of rules. As previously noted, in nature many complex systems operate
efficiently through the use of simple self-organization rules. While natural systems such as ant colonies, have precise,
preordained, and predictable global outcomes, most man-made systems exhibit imprecise, unpredictable and chaotic
behavior, and the difference is self-organization vs. central control.
Ashby introduced the concept of “requisite variety” which, loosely, posited that a variety of response
mechanisms are necessary to deal with the variety of states a system can be in [210]. Considering urban systems as
complex suggests that we cannot manage the complexity of large urban systems centrally, as a single central system does
not have the variety of response mechanisms required to control the variety of all possible traffic states in a city.
Solutions to traffic signal problems must distribute the coordination mechanism so that the complexity is allocated
among and across the (self-organizing) controller modules.
Precisely controlled communication at the local level leads to desired behavior outcomes at the global level in
natural systems. The desire to institute central control in artificial complex systems is inhibitive. In our efforts to control
complex systems, we need to embrace their inherent qualities and behaviors in order to align them with their natural
states, instead of forcing them under centralized control. Several artificial systems have effectively embraced self-
organization, the most notable of which is the Internet, which uses self-organization by coordinating millions of
relatively simple components to collectively create a virtual world exhibiting sophisticated behavior and properties.
Social networking applications (e.g., Facebook, Twitter, and LinkedIn) also exhibit such behavior. Spontaneous eruption
and scalability are hallmarks of these self-organized systems, which make them so potent.
A self-organized traffic system can similarly be designed by exploiting the natural order to optimize flow
through the entire traffic grid by using simple rules that govern a single signal’s behavior while enabling global
synchronization. Four paradigms are germane to the development of self-organized systems: (1) designing local
interactions that achieve global properties; (2) exploiting implicit coordination; (3) minimizing the maintained state; and
(4) designing protocols that adapt to change.
Associating complex systems with traffic control and developing the proper rules for interaction and behavior
can be a difficult undertaking. Traditional optimization techniques have a role in the design of complex systems. We
propose implementing a self-organizing traffic model using classical optimization techniques for local optimization at
each intersection, and coupling with adjacent signals through offsets that can be varied through simulated annealing or
some other stochastic method. Over time, the offsets can be synchronized pairwise across the entire grid, resulting in
global synchronization. The amount of information (e.g. sensors, statistical data flows) used in the optimization also
influences the performance of such algorithms. Consequently, local optimization algorithms, rules of coupling across
intersections, and the information available for optimization, all need to be investigated while developing self-organized
traffic models.
There are fundamental questions that need to be answered based on any model that we develop. For instance,
organisms that self-organize successfully use a limited number of attributes in decision-making. What is the minimal set
of rules that can be applied to efficient systems? This leads us to the notion of complexity. Our goal should be not only to
determine the minimal subset of rules, but also to understand the limits of self-organization. As the number of attributes
increases, will self-organization still be feasible? For instance, the free market economy is a self-organizing system with
a basic set of rules that all participants must follow. It generally works efficiently; however, its rules occasionally need to
be tweaked as participants capitalize on systematic inefficiencies and disrupt its natural harmony. Such possible flaws in
the rules of self-organization for traffic control could lead to dangerous traffic conditions and need to be identified. All of
these are relevant research problems that need to be addressed.
Self-organizing traffic signals promise to revolutionize current traffic control mechanisms and have the
potential for results beyond our expectations. For instance, Audi is already testing vehicular speed controllers that will
adjust speed based on traffic congestion and signal timing, such that the car gets a green light when it reaches an
intersection [201]. There are broader implications to this work; urban planning may need to be reviewed and traffic rules
There are other innovative ways to think about traffic optimization using techniques that become feasible as
sensing and communication technologies develop further. For instance, the network structure formed by the physical and
logical connections of traffic intersection signals can also be studied using graphs, such that each node is a signal and
each link is an artery. Traffic behavior can also be examined in a manner akin to communication network analysis.
Scheduling in traffic networks differs insofar as cars do not have destination addresses; their precise routes cannot be
known with certainty. In addition, there is a wavelike synchronization issue in which traffic that starts at the right time
and flows at the right velocity can “catch” all the green lights. In fact, if individual cars could be scheduled to leave at a
scheduled time and maintain a constant speed, it would be possible to minimize both delay and the very need for traffic
lights by having the flows become “orthogonal” to one another, in other words, able to flow through one another without
collision. This seems utopian right now, but with collision avoidance systems and vehicular sensors this may become a
reality sooner than we realize.
An intersection scheduling model based on individual vehicle arrivals has been developed by Wu and
colleagues [211]. They schedule the use of the intersection by different vehicles based on expected arrival time. This
approach assumes that there are no traffic lights at the intersection, but rather an intelligent device that determines the
schedule of intersection use for vehicles based on location, velocity, urgency, etc. Decomposition and backtracking logic
(at each time unit) are used to estimate which lane to release to minimize waiting time. Vehicles negotiate their access
time through an intelligent device embedded in the intersection. The scheduling mechanism is based on the space
between two successive vehicles in the conflicting traffic lane, and minimum space between successive vehicles. The
authors show that negotiation between vehicles by means of well-adapted approaches can be effective in improving
traffic control at a simple intersection. A central server manages and monitors the traffic in each intersection. This of
course assumes that all vehicles follow the command of the central controller, reducing considerably the complexity of
the traffic flow and facilitating its prediction. Thus, while we still have independently interacting vehicles, we must build
traffic light systems that match the complexity of the traffic flow. Self-organization offers one way of achieving this.
There are many factors which contribute to traffic flow, including driver behavior, urban topology, transit rules,
driving conventions, infrastructure, and demand. Here we have focused on one factor only: traffic signals. As we have
argued in the preceding analysis, traffic signal design is a multivariate optimization problem involving numerous
decision variables such as cycle lengths, split times and offsets. It is also a distributed optimization problem in which the
behavior of each traffic light can be controlled independently. Finally, it is a coupled optimization problem in which
changes in the timing patterns of one signal will influence the performance of other signals in the grid. Ideally, we would
like a closed-form solution to the problem such that all the traffic lights are optimized simultaneously (cycle lengths and
timing splits are coordinated) so as to minimize the aggregate delay across the entire grid. However, the problem is far
too complex to obtain a closed-form solution. Moreover, the specific traffic conditions change constantly, i.e. traffic is a
non-stationary problem, and so even if we were able to find an optimal solution, this would be obsolete in seconds. In
general, the problem can be approached iteratively, whereby solutions are created and evaluated successively to improve
the design progressively. Stochastic, gradient-based, and heuristic techniques have all been used for traffic signal
optimization. Both real-time and a priori approaches can be used with their own unique pros and cons. The
unpredictability of traffic patterns undermines the effectiveness of a priori models, while high latency and lack of
scalability have a negative impact on real-time models.
We can optimize complex systems by decoupling the problem naturally using the principles of self-
organization. This not only makes the solutions efficient, but also adaptable, as discussed in previous sections. The traffic
signal optimization problem is a classic case of such a complex system in which a network of heterogeneous components
interacts nonlinearly, giving rise to emergent behavior. Agents interact with each other and make independent local
decisions based on input from the environment.
While the optimality of results cannot be gauranteed, they are more responsive to local conditions, and are
closer to the theoretical optimum compared to existing techniques. Preliminary attempts have been made to address the
problem of traffic signal synchronization using self-organizing approaches. We believe that this is a promising area of
research that remains largely unexplored. A variety of optimization algorithms and traffic condition data sets could be
tested at the local level to investigate the impact of various datasets, and to test the robustness of models in a variety of
weather conditions. Benchmarks and optimality measures are required for comparing different proposals in simulations.
In this direction, Zhang et al. [231] have used macroscopic fundamental diagrams to compare different traffic light
coordination mechanisms, showing that self-organizing methods achieve overall better performance and a higher
network capacity than centralized or adaptive control mechanims.
In practice, the suitability of such techniques depends on the quality of real-time traffic data in the local signal
grid, which comes at a cost. With advances in sensor and communication technology, the availability and quality of data
will continue to improve. However, there is no doubt that communications and sensors have advanced sufficiently to
allow us to move such a system from conception to reality.
There are several unanswered questions about the design of such systems that will need to be addressed, and are
potent topics of research. The research on complex systems in this area is still rudimentary, and needs to be examined
further. Also, with advances in vehicular sensors, we may be able to create more innovative models for traffic
optimization by controlling the behavior of vehicles, rather than just traffic lights. The work of Berger and Rumpe [212],
for example, shows that the development of autonomous vehicles is already moving in this direction. Given the trend
towards interconnectivity, it is only a matter of time before vehicles communicate with one another and with
infrastructure. With greater research, lessons from natural self organized systems could inform complex transportation
systems towards greater precision of controlled communication at the local level leading to desired behavior outcomes at
the global level.
“The author(s) declare(s) that there is no conflict of interest regarding the publication of this paper.”
D. Schrank, B. Eisele e T. Lomax, “Urban Mobility Report,” Texas A&M Transportation Institute,
Texas, 2012.
C. H. Papadimitriou e J. N. Tsitsiklis, Mathematics of Operations Research, vol. 24, nº 2, p. 293
305, 1999.
Y. Zhao e Z. Tian, “An overview of the usage of adpative signal control system in the United States
of America,” Applied Mechanics and Materials , Vols. %1 de %2178-181, pp. 2591-2598, 2012.
L. Day e I. McNeil, Biographical Dictionary of the History of Technology, Taylor & Francis, 1996.
G. M. Sessions, Traffic Devices: Historical aspects thereof, Washington, D.C.: Institute of Traffic
Engineers, 1971.
T.O.S.A. Association, New Traffic Signal Installed, 1914, pp. 28-9.
P. Robert, L. Gordon e P. W. Tighs, Traffic Control Systems Handbook, Sunn Engineering
Associates in Association with Siemens Intelligent Transportation Systems, 2005.
P. Pretorious, S. van As e R. Troutbeck, “Issues with the operational analysis of urban
intersections,” em 23rd Annual Southern African Transport Conference, 2004.
H. Van Zuylen e F. Viti, “Delay at Controlled Intersections: Theo Old Theory Revised,” em IEEE
Intelligent Transportation Systems, 2006.
J. Ben-edigbe, “Emperical Delays from Actuated and Optimised Static Signal Compared,” ARPN
Journal of Engineering and Applied Sciences, vol. 5, nº 1, pp. 42-8, 2010.
M. J. Panaggio, B. J. Ottino-Löffler, P. Hu e D. M. Abrams, “Symmetry breaking in optimal timing
of traffic signals on an idealized two-way street,” Phys. Rev. E 88: 032801, 2013.
B. Ramanathan, M. G. McNally e R. Jayakrishnan, “Formulation of modern signal control
operations as a non-linear mixed integer program,” em VNISL: Vehicle Navigation and Information
Systems Conference, 6th, Seattle, WA, 1995.
R. M. Kimber, “The traffic capacity of roundabouts,” Transport and Road Research Laboratory
(No. TRRL LR942 Monograph), 1980.
M. E. Fouladvand, Z. Sadjadi e M. R. Shaebani, “Characteristics of vehicular traffic flow at a
roundabout,” Physical Review E, vol. 70, nº 4, p. 046132, 2004.
P. T. Martin, J. Perrin, B. R. Chilukuri, C. Jhaveri e Y. Feng, “Adaptive Signal Control II,” Utah.
Department of Transportation. Research Division, 2003.
D.E. Associates e S.I.T. Systems, “Traffic Control Systems Handbook,” Federal Highway
Administration, 2005.
J. D. Little, “Synchronization of Traffic Signals by Mixed-Integer Linear Programming,”
Operations Research, vol. 14, nº 4, pp. 568-94, July-August 1966.
Q. He, K. L. Head e J. Ding, “PAMSCOD: Platoon-based arterial multi-modal signal control with
online data,” Transportation Research Part C: Emerging Technologies, vol. 20, nº 1, pp. 164-184,
K. Han, Y. Sun, H. Liu, T. L. Friesz e T. Yao, “A bi-level model of dynamic traffic signal control
with continuum approximation,” Transportation Research Part C: Emerging Technologies, vol. 55,
pp. 409-431, 2015.
T. Nakatsuji e T. Kaku, “Development of a self-organizing traffic control system using neural
network models,” Transportation Research Board, vol. 1324, pp. 137-45, 1991.
C. Ledoux, “An urban traffic flow model intergrating neural networks,” Transportation Research,
Part C: Emerging Technologies, vol. 5, pp. 287-300, 1997.
S. Mikami e Y. Kakazu, “Genetic reinforcement learning for cooperative traffic signal control,” em
IEEE World Congress on Evolutionary Computation, 1994.
S. Chiu e S. Chand, “Adaptive Traffic Signal Control Using Fuzzy Logic,” em IEEE 2nd
International Conference on Fuzzy Systems, 1993.
M. Trabia, M. Kaseko e M. Ande, “A two-stage fuzzy logic controller for traffic signals,”
Transportation Research, Part C: Emerging Technologies, vol. 7, nº 6, pp. 353-67, 1999.
Q. He, K. L. Head e J. Ding, “Multi-modal traffic signal control with priority, signal actuation and
coordination,” Transportation Research Part C: Emerging Technologies, vol. 46, pp. 65-82, 2014.
W. Li, R. Li, D. He e F.-Y. Wang, “Intelligent traffic signal system based on network control,” em
IEEE Networking, Sensing, and Control, 2005.
T. Heung, T. Ho e Y. Fung, “Coordinated Road-Junction Traffic Control by Dynamic
Programming,” IEEE Transactions on Intelligent Transportation Systems, vol. 6, nº 3, pp. 341-50,
M. Dotoli, M. P. Fanti e C. Meloni, “A signal timing plan formulation for urban traffic control,”
Control Engineering Practice, vol. 14, nº 11, pp. 1297-1311, 2006.
H. Ceylan, “Developing combined genetic algorithm - hill-climbing optimization method for area
traffic control,” Journal of Transportation Engineering, vol. 132, nº 8, pp. 663-71, 2006.
M. Patriksson, “Sensitivity analysis of traffic equilibria,Transportation Science, vol. 3, nº 258-81,
p. 38, 2004.
M. Shenoda e R. Machemehl, “Development of a phase-by-phase, arrival-based, delay-optimized
adaptive traffic signal control methodology with metaheuristic search,” College Station, TX, 2006.
D. Sun, R. F. Benekohal e S. T. Waller, “Multiobjective traffic signal timing optimization using
non-dominated sorting genetic algorithm,” em IEEE Intelligent Vehicles Symposium, Columbs, OH,
J. Chen e L. Xu, “Road-Junction Traffic Signal Timing Optimization by and Adaptive Particle
Swarm Algorithm,” em 9th International Conference on Control, Automation, Robotics, and
Vision, 2006.
J. J. Sanchez, M. Galan e E. Rubio, “Genetic algorithm and cellular automata: a new architecture
for traffic light cycles optimization,” em CEC: Congress on Evolutionary Computation, 2nd,
Portland, OR, 2004.
S. Takahashi, H. Nakamura, H. Kazama e T. Fujikura, “Genetic algorithm approach for adaptive
offset optimization for the fluctuation of traffic flow,” em IEEE Internation Conference on
Intelligent Transportation Systems, Singapore, China, 2002.
C. Gershenson, Design and Control of Self-organizing Systems, Brussels: Vrije Universiteit, 2007.
S. Camazine, J.-L. Deneubourg, N. R. Franks, J. Sneyd, G. Theraulaz e E. Bonabeau, Self-
Organization in Biological Systems, Princeton University Press, 2003.
N. H. Gartner, “OPAC: A demand-responsive strategy for traffic signal control,” Transportation
Research Record, nº 906, pp. 75-81, 1983.
P. B. Hunt, D. I. Robertson, R. D. Bretherton e R. I. Winton, “SCOOT - A traffic responsive
method for coordinating signals,” Berkshire, England, 1981.
Y. K. Luk, “Two Traffic-Responsive Area Traffic Control Methods: SCATS and SCOOT,” Traffic
Engineering and Control, vol. 25, nº 1, pp. 14-20, 1984.
D. Robertson e R. Bretharton, “Optimizing networks of traffic signals in ,real-time- the SCOOT
method,” IEEE Transactions on Vehicular Technology, vol. 25, nº 1, pp. 11-5, 1991.
A. G. Sims, “The Sydney Coordinated Adaptive Traffic Systems,” em Engineering Foundation
Conference on Research Priorities in Computer Control of Urban Traffic Systems, New York, NY,
P. R. Lowrie, SCATS, Sydney Co-ordinated Adaptive Traffic System : A Traffic Responsive
Method of Controlling Urban Traffic, Sydney, Australlia: Roads and Traffic Authority of New
South Wales. Traffic Control Section, 2002.
P. Mirchandani e F.-Y. Wang, “RHODES to Intelligent Transportation Systems,” IEEE Intelligent
Transportation Systems, vol. 20, nº 1, pp. 10-15, January/February 2005.
P. Mirchandani e L. Head, “RHODES: A real-time traffic signal control system: architecture,
algorithms, and analysis,” Transportation Research Part C: Emerging Technologies, vol. 9, nº 6,
pp. 415-32, 2001.
L. Liao, “A review of the Optimized Policies for Adaptive Control Strategy (OPAC). California
PATH Working Paper.,” Institute of Transportation Studies, University of California, Berkeley,
P. R. Lowrie, “SCATS: A Traffic Responsive Method of Controlling Urban Traffic Control,”
Sydney, Australia, 1992.
D. I. Robertson e R. D. Bertherton, “Optimum control of an intersection for any known sequence of
vehicular arrivals,” em Proceedings of the 2nd IFAC-IFIP-IFORS Symposium on Traffic Control
and Transportation system, Monte Carlo, 1974.
J. J. Henry, J. L. Farges e J. Tuffal, “The Prodyn real-time traffic algorithm,” em IFAC Control in
Transportation Systems, Baden-Baden, Germany, 1983.
R. A. Vincent e C. P. Young, “Self Optimising Traffic Signal Control Using Microprocessors - the
TRRL "MOVA" Strategy for Isolated Intersections,” Traffic Engineering and Control, vol. 27, nº 7-
8, pp. 385-7, July 1986.
V. Mauro e D. Di Taranto, “UTOPIA,” em Proceedings of the 6th IFAC / IFIP / IFORS Symposium
on Control Computers and communication in Transportation, Paris, France, 1990.
S. Yagar e B. Han, “A procedure for real-time signal control that considers transit interference and
priority,” Transportation Research B, vol. 28B, nº 4, pp. 315-331, 1994.
C. Diakaki, V. Dinopoulou, K. Aboudolas, M. Papageorgiou, E. Ben-Shabat, E. Seider e A. Leibov,
“Extensions and new applications of the traffic signal control strategy TUC,” Transportation
Research Record, vol. 1856, p. 202211, 2003.
I. Porche e S. Lafortune, “Adaptive Look-Ahead Optimization of Traffic Signals,” Intelligent
Transportation Systems Journal, vol. 4, nº 3-4, pp. 209-54, 1999.
V. Gradinescu, C. Gorgorin, R. Diaconescu, V. Cristea e L. Iftode, “Adaptive traffic lights using
car-to-car communication,” em Vehicular Technology Conference, 2007. VTC2007-Spring. IEEE
65th , Dublin, 2007.
Z. Tian e T. Urbanik, “System partition technique to improve signal coordination and traffic
progression,” Journal of Transportation Engineering, vol. 133, nº 2, pp. 119-128, 2007.
A. L. C. Bazzan, “A distributed approach for coordination of traffic signal agents,” Autonomous
Agents and Multiagent Systems, vol. 10, nº 1, pp. 131-164, 2005.
F. Zhu, H. M. A. Aziz, X. Qian e S. V. Ukkusuri, “A junction-tree based learning algorithm to
optimize network wide traffic control: A coordinated multi-agent framework,” Transportation
Research Part C: Emerging Technologies, p. In Press, 2015.
J. Sanchez, M. Galan e E. Rubio, “Applying a traffic lights evolutionary optimization technique to a
real case: “Las Ramblas” area in Santa Cruz de Tenerife,” IEEE Transactions on Evolutionary
Computation, vol. 12, nº 1, pp. 25-40, 2008.
M. Lighthill e G. Whitham, “On kinematic waves, II. A theory of traffic flow on long crowded
roads,” em Proceedings from the Royal Society of London, Series A, Mathematical and Physical
Sciences, London, 1955.
W. Leutzbach, Introduction to the Theory of Traffic Flow, Berlin, Germany: SpringerVerlag, 1988.
R. Herman, E. Montroll, R. Potts e R. Rothery, “Traffic Dynamics: Analysis of Stability in Car
Following,” Operations Research, vol. 7, nº 1, pp. 86-106, 1959.
M. Bando, K. Hasbe, A. Nakayama, A. Shibata e Y. Sugiyama, “Dynamical Model of Traffic
Congestion and Numerical Simulation,” Physical Review E., vol. 51, nº 2, pp. 1035-42, 1995.
S. Yukawa e M. Kikuchi, “Coupled-Map Modeling of One-Dimensional Traffic Flow,” Physics and
Society, Osaka, Japan, 1995.
O. Biham, A. Middleton e D. Levine, “Self-organization and a Dynamical Transition in Traffic-
flow Models,” Physical Review A, vol. 46, nº 10, pp. 46-55, 1992.
T. Ohira, “Autonomous Traffic Signal Control Model with Neural Network Analogy,” Sony
Computer Science Laboratory, Osaka, Japan, 1997.
S. Hoogendoom e P. Bovy, “State-of-the-art of vehicular traffic flow modelling,” Journal of
Systems and Control Engineering, vol. 215, nº 4, pp. 283-303, 2001.
H. J. Payne, “Models of freeway traffic and control,” Simulation Councils Proc. Series:
Mathematical Models of Public Systems, vol. 1, nº 1, pp. 51-60, 1971.
W. F. Philips, “Kinetic Model for Traffic Flow with Continuum Implications,” Transportation
Research Planning and Technology, vol. 5, pp. 131-138, 1979.
R. D. Kuhne, “Traffic patterns in unstable traffic flow on freeways,” em Proceedings of the
International Symposium on Highway Capacity, Balkema, Rotterdam, 1991.
B. Kerner, P. Konhäuser e M. Schilke, “A new approach to problems of traffic flow theory,” em
Proceedings of the 13th International Symposium of Transportation and, Lyon, 1996.
D. Branston, “Models of Single Lane Time Headway Distributions,” Transportation Science, vol.
10, pp. 125-148, 1976.
X.-F. Xie, S. F. Smith, L. Lu e G. J. Barlow, “Schedule-driven intersection control,” Transportation
Research Part C: Emerging Technologies, vol. 24, pp. 168-189, 2012.
I. Prigogine e R. Herman, Kinetic Theory of Vehicular Traffic, New York: Elsevier, 1971.
S. P. Hoogendoorn, Multiclass Continuum Modelling of Multiclass Traffic Flow, PhD. Thesis
T99/5. Delft University Press, 1999.
K. Nagel e M. Schreckenberg, “A cellular automaton model for freeway traffic,” J. Phys. I France,
pp. 2221-9, 1992.
S. Krauss, P. Wagner e C. Gawron, “Metastable States in a Microscopic Model of Traffic Flow,”
Physical Review E, vol. 55, nº 304, pp. 55-97, 1997.
D. Chowdhury, L. Santen e A. Schadschneifer, “Simulation of Vehicular Traffic: a Statistical
Physics Perspective,” Computing in Science & Engineering, vol. 2, nº 5, pp. 80-7, 2000.
M. Rickert, K. Nagel, M. Schreckenberg e A. Latour, “Two lane traffic simulations using cellular
automata,” Physica A: Statistical Mechanics and its Applications, vol. 231, nº 4, pp. 534-550, 1996.
M. E. Fouladvand, Z. Sadjadi e M. R. Shaebani, “Characteristics of Vehicular Traffic Flow at a
Roundabout,” Physical Review E, vol. 70, nº 4, pp. 1-8, 2004.
E. Brockfeld, A. Barlovic, A. Schadschneider e M. Schreckenberg, “Optimizing Traffic Lights in a
Cellular Automaton Model for City Traffic,” Physical Review E, vol. 64, nº 5, p. 56132, 2001.
W. Burghout e J. Wahlstedt, “Hybrid traffic simulation with adaptive signal control,”
Transportation Research Record: Journal of the Transportation Research Board, pp. 191-7, 2007.
T. Winters, M. Johnson e V. Paruchuri, “LITS: Lightweight Intelligent Traffic Simulator,” em
IEEE Internation Conference on Network-Based Information Systems, 2009.
A. P. Akgungor e A. G. R. Bullen, “A New Delay Parameter for Variable Traffic Flows at
Signalized Intersections,” Turkish Journal of Engineering and Environmental Sciences, vol. 7, nº 1,
pp. 61-70, 2007.
M. Balmer, K. Axhausen e K. Nagel, An Agent Based Demand Modeling Framework for Large
Scale Micro-Simulations, Zurich, Switzerland: Working paper, 329, Institute for Transport Planning
and Systems (IVT), ETH, 2005.
D. A. Rosenblueth e C. Gershenson, “A model of city traffic based on elementary cellular
automata,” Complex Systems, vol. 19, nº 4, pp. 305-322, 2011.
R. Bellman, “The theory of dynamic programming,” Bulletin of the American Mathematical
Society, vol. 60, pp. 503-516, 1954.
P. Yi, C. Shao e L. Sheng, “Improved Signal Control for Oversaturated Intersection,” em IEEE
Intelligent Transportation Systems, 2006.
E. Todorov, “Optimal control theory and grants in aid,” em Bayesian Brain, Cambridge, MA, MIT
Press, 2006, pp. 1-28.
C. Cai, “An Approximate Dynamic Programming Strategy for Responsive Traffic Signal Control,”
em International Symposium on Approximate Dynamic Programming and Reinforcement Learning,
C. Cai, C. K. Wong e B. G. Heydecker, “Adaptive traffic signal control using approximate dynamic
programming,” Transportation Research Part C: Emerging Technologies, vol. 17, nº 5, pp. 456-
474, 2009.
R. E. Allsop, “Delay-Minimizing Settings for Fixed-Time Traffic Signals at a Single Road
Junction,” IMA Journal of Applied Mathematics, vol. 8, nº 2, pp. 164-85, 1971.
H. Shimizu, M. Kobayashi, H. Fujii e S. Katagiri, “A development of deterministic signal control
system in urban road networks,” em SICE Annual Conference, Fukuyama, 2008.
W. Wey, “Applications of linear systems controller to a cycle-based traffic signal control,” em
IEEE Intelligent Transportation Systems, 2001.
J. Ball, M. Day, T. Yu e P. Kachroo, “An isolated traffic intersection feedback control using H,” em
IEEE Intelligent Transportation Systems, 1997.
H. Kwakernaak, “Robust Control and H Optimization - Tutorial Paper,Automatica, pp. 255-73,
X. Yang, C. Yuan, F. Shen e W. Ma, “Feedback Control Strategy for Isolated Intersections,” em
IEEE Intelligent Transportation Systems, 2006.
E. B. Kosmatopolous, M. Papageorgiou, A. Vakouli e A. Kouvelas, “Adaptive Fine-Tuning of
Nonlinear Control Systems with Application to the Urban Traffic Control Strategy TUC,” IEEE
Transactions on Control Systems Technology, vol. 15, nº 6, pp. 991-1002, 2007.
K. Aboudolas, M. Papageorgiou e E. Kosmatopoulos, “Store-and-Forward Based Methods for teh
Signal Control Problem in Large-scale Congested Urban Road Networks,” Transportation
Research Part C: Emerging Technologies, vol. 17, nº 2, pp. 163-74, 2009.
P. Pecherkova, M. Flidr e J. Dunik, “Application of estimation techniques on queue lengths
estimation in traffic network,” em IEEE 7th International Conference of Cybernetic Intelligent
Systems, Prague, 2008.
L. A. Zadeh, “Fuzzy Sets,” Information and Control, vol. 8, pp. 338-53, 1965.
C. P. Pappis e E. H. Mamdani, “A Fuzzy Logic Controller for a Traffic Junction,” IEEE
Transactions on Systems, Man, And Cybernetics, vol. 7, nº 10, pp. 707-17, 1977.
I. Nainar, M. Mohammadian e J. Millar, “Adaptive Control of Traffic Signals Using Fuzzy Logic
and Genetic Algorithms,” em 4th International Conferene on Automatic Control, Robotics, and
Computer Vision, Singapore, 1996.
O. Cordon, F. Herrera, F. Hoffman e L. Magdalena, Genetic Fuzzy Systems: Evolutionary Tuning
and Learning of Fuzzy Knowledge Bases, Hackensack, NJ: World Scientific Publishing Company,
Y. C. Chiou e L. Lan, “Adaptive traffic signal control with iterative genetic fuzzy logic and
evolutionary algorithms (GFLC),” em IEEE International Conference on Networking, Sensing, and
Control, 2004.
Y. Hu, P. Thomas e R. J. Stonier, “Traffic signal control using fuzzy logic and evolutionary
algorithms,” em IEEE Congress on Evolutionary Computation, 2007.
J. Niittymaki e S. Kikuchi, “Application of Fuzzy Logic to the Control of a Pedestrian Crossing
Signal,” Transportation Research Record, vol. 1651, nº 1, pp. 30-8, 1998.
F. Kobayashi, K. Takishita e T. Fukuda, “Self-organization of non-signal urban traffic flow with
fuzzy model,” em IEEE Intelligent Transportation Systems, 2000.
G. Nakamiti, R. C. Demac e F. Gomide, “Fuzzy sets in distributed traffic control,” em IEEE 5th
International Conference on Fuzzy Systems, 1996.
J. Lee e H. Lee-Kwang, “Distributed and cooperative fuzzy controllers for traffic intersections,”
IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications on Reviews, vol. 29, nº
2, pp. 263-71, 1999.
M. Wiering, J. Van Veenen, J. Vreeeken e A. Koopman, “Intelligent traffic light control,”
European Research Consotrium for Informatics and Mathematics, vol. 53, pp. 40-1, 2003.
K. Bogenberger, H. Keller e S. Vukanovic, “A neuro-fuzzy algorithm for coordinated traffic
responsive ramp metering,” IEEE Intelligent Transportation Systems, vol. 49, pp. 94-9, 2001.
L. Conglin, W. Wu, I. Member e T. Yuejin, “Traffic variable estimation and traffic signal control
based on soft computation,” em IEEE 7th International Conference on Intelligent Transportation
Systems, 2004.
L. Zang, L. Jia e Y. Luo, “An Intelligent Control Method for Urban Traffic Signal Based on Fuzzy
Neural Network,6th World Congress on Intelligent Control and Automation, pp. 3430-4, 2002.
T. Li, D. Zhao e J. Yi, “Adaptive dynamic neuro-fuzzy system for traffic signal control,” em IEEE
International Joint Conference on Neural Networks, 2005.
M. Choy, D. Srinivasan e R. Cheu, “Cooperative hybrid agent architecture for real-time traffic
signal control,” IEEE Transactions on Systems, Man, and Cybernetics, Part A, vol. 33, nº 5, pp.
597-607, 2003a.
M. Choy, D. Srinivasan e R. Cheu, “Neural networks for continuous online learning and control,”
IEEE Transactions on Neural Networks, vol. 17, nº 6, pp. 1511-31, 2003b.
J. Z. Zhu, J. X. Cao e Y. Zhu, “Traffic volume forecasting based on radial basis function neural
network with the consideration of traffic flows at the adjacent intersections,” Transportation
Research Part C: Emerging Technologies, vol. 47, pp. 139-154, 2014.
S. Lu, “Incremental multistep Q-learning for adaptive traffic signal control based on delay
minimization strategy,” em IEEE 7th World Congress on Intelligent Control and Automation, 2008.
C. Ozan, O. Baskan, S. Haldenbilen e H. Ceylan, “A modified reinforcement learning algorithm for
solving coordinated signalized networks,” Transportation Research Part C: Emerging
Technologies, vol. 54, pp. 40-55, 2015.
N. Rouphail, A. Tarko e J. Li, “Traffic flow at signalized intersections,” em Traffic Flow Theory,
Monograph, Revised, Transportation Research Board, 1975.
F. Viti, The Dynamics and the Uncertainty of Delays at Signals, Delft, Netherlands: Delft
University of Technology, 2006.
M. J. Beckmann, C. B. McGuire e C. B. Winsten, Studies in the Economics of Transportation, Yale
University Press, 1956.
F. V. Webster, Traffic Signal Settings, London: Road Research Lab, 1958.
H. Lehmann, “Direced and Diffusive Contributions to Urban Traffic Flow Patterns,” Europhysics
Letters, vol. 64, nº 2, pp. 288-94, 2003.
Z. Tian, “Capacity Analysis of Traffic-Actuated Intersections,” 2002.
J. Yeon e B. Ko, “Comparison of Travel Time Estimation Using Analysis and Queuing Theory to
Field Data Along Freeways,” em Internation Conference on Multimedia and Ubiquitous
Engineering, 2007.
W. J. Freeman, K. Y. Ho e E. A. McChesney, “Evaluation of Signalized Intersection System
Analysis Techniques,” ITE, 2000.
E. Bonabeau, M. Dorigo e G. Theraulaz, Swarm Intelligence: From Natural to Artificial Systems.
Santa Fe Institute Studies in the Sciences of Complexity, New York: Oxford University Press,
P. Boolchand, G. Lucovsky, J. C. Phillips e M. F. Thorpe, “Self-organization and the physics of
glassy networks,” Philosophical Magazine, vol. 85, nº 32, pp. 3823-3838, 2005.
M. Bartolozzi, D. B. Leinweber, T. Surungan, A. W. Thomas e A. G. Williams, “Scale-free
networks in complex systems,” em SPIE International Symposium Microelectronics, MEMS, and
Nanotechnology, Brisbane, 2005.
W. R. Ashby, “Principles of the self-organizing dynamic system,” Journal of General Psychology,
vol. 37, pp. 125-128, 1947.
H. von Foerster, “On self-organizing systems and their environments,” em Self-Organizing Systems,
M. C. Yovitts e S. Cameron, Edits., New York, Pergamon, 1960, pp. 31-50.
G. G. Lendaris, “On the definition of self-organizing systems,” Proceedings of the IEEE, vol. 52, nº
3, pp. 324-325, 1964.
H. Haken, Information and Self-organization: A Macroscopic Approach to Complex Systems,
Berlin: Springer-Verlag, 1988.
F. Heylighen, “The science of self-organization and adaptivity,” em The Encyclopedia of Life
Support Systems, L. D. Kiel, Ed., Oxford, EOLSS Publishers, 2003.
C. Gershenson e F. Heylighen, “When can we call a system self- organizing?,” em Advances in
Artificial Life, 7th European Conference, ECAL 2003 LNAI 2801, W. Banzhaf, T. Christaller, P.
Dittrich, J. T. Kim e J. and Ziegler, Edits., Berlin, Springer, 2003, p. 606614.
S. Beer, Decision and Control: The Meaning of Operational Research and Management
Cybernetics, West Sussex: John Wiley and Sons, Ltd., 1966.
H. Van Dyke Parunak e S. Brueckner, “Entropy and self-organization in multi-agent systems,” em
5th International Conference on Autonomous Agents, New York, NY, 2001.
Heylighen, F., Cilliers, P. & Gershenson, G., “Complexity and Philosophy,” em In Bogg, J. and R.
Geyer (eds.) Complexity, Science and Society., Oxford, Radcliffe Publishing, 2007, p. 117134.
M. Dorigo e T. Stützle, Ant Colony Optimization, Cambridge, MA: MIT Press, 2004.
Werfel, J., Petersen, K. & Nagpal, R, “Designing collective behavior in a termite-inspired robot
construction team,” Science 343 (6172), p. 754758, 2014.
R. E. Mirillo e S. H. Strogatz, “Synchronization of Pulse-Coupled Bilogical Oscillators,SIAM
journal of Applied Mathematics, vol. 50, nº 6, pp. 1645-62, December 1990.
R. Kindermann e J. L. Snell, “Markov Random Fields and Their Applications,” Contemporary
Mathematics, vol. 1, nº 1, p. 142, 1980.
B. Faieta e B. A. Huberman, “Firefly: A synchronization strategy for urban traffic control,”
Technical Report SSL-42, Xerox PARC, Palo Alto, CA, 1993.
R. Hoar, J. Penner e C. Jacob, “Evolutionary swarm traffic: if ant roads had traffic lights,” em
Congress on Evolutionary Computation, 2002.
S. Lammer, H. Kori, K. Peters e D. Helbing, “Decentralised control of material or traffic flows in
networks using phase-synchronisation,” Physica A, vol. 363, nº 1, pp. 39-47, 2006.
A. Schadscneider, “Traffic Flow: A Statistical Physics Point of View,” Physica A: Statistical
Mechanics and its Applications, vol. 313, nº 1-2, pp. 153-87, 2002.
M. Sasaki e T. Nagatani, “Transition and saturation of traffic flow controlled by traffic lights,
Physica A: Statistical Mechanics and its Applications, vol. 325, nº 3-4, pp. 531-46, 2003.
C. Daganzo, “The cell transmission model: A dynamic representation of highway traffic consistent
with the hydrodynamic theory,” Transportation Research Part B: Methodological, vol. 28, nº 4, pp.
269-87, 1994.
D. Helbing, “A Section-based Queuing-theoretical Traffic Model for Congestion and Travel Time
Anaysis in Networks,” Journal of Physics A: Mathematical and General, vol. 36, nº 46, pp. 593-8,
D. Helbing e P. Molnár, “Social Force Model for Pedestrian Dynamics,” Physical Review E, vol.
65, nº 9, pp. 1331-86, 2002.
T. Nagatani, “The Physics of Traffic Jams,” Reports on Progress in Physics, vol. 65, nº 9, pp. 1331-
86, 2002.
Török & Kertész, “The green wave model of two-dimensional traffic: Transitions in the flow
properties and in the geometry of the traffic jam,” Physica A: Statistical Mechanics and its
Applications, pp. 515-533, 1996.
P. Koonce, L. Rodegerdt, K. Lee, S. B. Quayle, C. Braud, J. Bonneson, P. Tarnoff e T. F. Urbanic,
“Traffic Signal Timing Manual,” 2008.
S. B. Cools, C. Gershenson e B. D'Hooghe, “Self-organizing traffic lights: A realistic simulation,”
em Advances in Applied Self-Organizing Systems, M. Prokopenko, Ed., London, Springer, 2008,
pp. 41-50.
C. Gershenson e D. A. Rosenblueth, “Adaptive self-organization vs. static optimization: A
qualitative comparison in traffic light coordination,” Kybernetes, vol. 41, nº 3, pp. 386-403, 2012.
C. Gershenson e D. A. Rosenblueth, “Self-organizing traffic lights at multiple-street intersections,”
Complexity, vol. 17, nº 4, pp. 23-29, 2012.
D. Zubillaga, G. Cruz, L. D. Aguilar, J. Zapotécatl, N. Fernández, J. Aguilar, D. A. Rosenblueth e
C. Gershenson, “Measuring the Complexity of Self-Organizing Traffic Lights,” Entropy, vol. 16,
pp. 2384-2407, 2014.
C. Gershenson, “Self-organizing Traffic Lights,” Complex Systems, vol. 16, nº 1, pp. 29-53, 2005.
C. Gershenson, “A General Methodology for Designing Self-Organizing Systems,” 2005.
J. Kennedy e R. Eberhart, “ Particle swarm optimization,” Proceedings of IEEE international
conference on neural networks, vol. 4, nº 2, pp. 1942-1948, 1995.
H. M. Kammoun, I. Kallel, J. Casillas, A. Abraham e A. M. Alimi, “2014,” Transportation
Research Part C: Emerging Technologies, vol. 42, pp. 147-167, 2014.
B. Tatomir e L. Rothkrantz, “Ant based mechanism for crisis response coordination,” Berlin
Heidelberg, Springer, 2006, pp. 380-387.
D. Alves, J. v. Ast, Z. Cong, B. D. Schutter e R. Babuska, “Ant colony optimization for traffic
dispersion routing,” em Proceedings of the 13th International IEEE Conference on Intelligent
Transportation Systems (ITSC 2010), Madeira Island, Portugal, 2010.
Z. Cong, B. De Schutter e R. Babuška, “Ant Colony Routing algorithm for freeway networks,”
Transportation Research Part C: Emerging Technologies, vol. 37, pp. 1-19, 2013.
B. Cesme e P. G. Furth, “Self-organizing traffic signals using secondary extension and dynamic
coordination,” Transportation Research Part C: Emerging Technologies, vol. 48, pp. 1-15, 2014.
J. de Gier, T. M. Garoni e O. Rojas, “Traffic flow on realistic road networks with adaptive traffic
lights,” Journal of Statistical Mechanics, vol. P04008, 2011.
J. Wei, A. Wang e N. Du, “Study of Self-Organizing Control of Traffic Signals in an Urban
Network Based on Cellular Automata,” IEEE Transactions on Vehicular Technology, vol. 54, nº 2,
pp. 744-8, 2005.
X. Wu, G. Sun, Z. Piao, Z. Liu e P. Yang, “Application of the Cellular Automata in Simulating and
Analyzing Two-Dimaneional Traffic Model,” em IEEE Intelligent Control and Automation, 6th
World Conference, Shenyang, China, 2006.
Z. Yang, S. Lu e X. Liu, “Combined Traffic Signal Control and Route Guidance: Multiple Uesr
Class Traffic Assignment Model versus Discrete Choice Model,” em Multiconference on
Computational Engineering in Systems Applications, 2006.
L. Shoufeng e L. Ximin, “Based on Hybrid Genetic Algorithm and Cellular Automata Combined
Traffic Signal Control and Route Guidance,” em Chinese Control Conference, Changsa, 2007.
A. Paz, V. Molano, E. Martinez, C. Gaviria e C. Arteaga, “Calibration of traffic flow models using
a memetic algorithm,” Transportation Research Part C: Emerging Technologies, vol. 55, pp. 432-
443, 2015.
S. Kumar e S. Mitra, “Self-organizing Traffic at a Malfunctioning Intersection,” Journal of
Artificial Societies and Social Simulation, vol. 9, nº 4, 2006.
G. Flotterod, “Some practical extensions to the cell transmission model,” em IEEE Intelligent
Transportation Systems, 2005.
C. M. Tampere e L. H. Immers, “An Extended Kalman Filter Application for Traffic State
Estimation Using CTM with Implicit Mode Switching and Dynamic Parameters,” em IEEE
Intelligent Transportation Systems Conference, 2007.
J. Chen, L. Xu, X. Yang e C. Yuan, “A hierarchy control algorithm and its application in urban
arterial control,” em IEEE Intelligent Transportation Systems, 2007a.
J. Chen, L. Xu, X. Yang e C. Yuan, “Oversaturated adjacent intersections control based on multi-
objective compatible control algorithm,” em IEEE Intelligent Transportation Systems, 2007b.
P. Bak, C. Tang e K. Wiesenfeld, “Self-organizaed Criticality: An Explanation of the 1/f Noise,”
Physical Review Letters, vol. 59, nº 4, pp. 381-4, 1987.
C. Gershenson, “Control de tráfico con agentes: CRASH,” em Memorias XI Congreso Nacional
ANIEI, Xalapa, 1998.
K. Dresner e P. Stone, “Multiagent Traffic Management: A Reservation-Based Intersection Control
Mechanism,” em The Third International Joint Conference on Autonomous Agents and Multiagent
Systems, 2004.
M. Vasirani e S. Ossowski, “A computational market for distributed control of urban road traffic
systems,” IEEE Transactions on Intelligent Transportation Systems , vol. 12, nº 2, pp. 313-321,
S. Goel, S. F. Bush e K. Ravindranathan, “Self-organization of traffic lights for minimizing vehicle
delay,” em Proceedings of the 3rd International Conference on Connected Vehicles and Expo
(ICCVE), IEEE, Vienna, Austria, 2014.
S. Goel, E. Dincelli, A. Parker e E. Sprissler, “Enabling a self-organized traffic system in existing
legacy hardware,” em Proceedings of the 3rd International Conference on Connected Vehicles and
Expo (ICCVE), IEEE, Vienna, Austria, 2014.
S. Lammer e D. Helbing, “Self-control of traffic lights and vehicle flows in urban road networks,”
Journal of Statistical Mechanics: Theory and Experiment, vol. P04019, 2008.
J. Gajda, R. Sroka, M. Stencel, A. Wajda e T. Zeglen, “A vehicle classification based on inductive
loop detectors,” em IEEE 18th Instrumentation and Measurement Technology Conference:,
Budapest, 2001.
B. Coifman e S. Kim, “Speed Estimation and Length Based Vehicle Classification from Freeway
Single Loop Detectors,” Transportation Research Part C: Emerging Technologies, vol. 17, nº 4,
pp. 349-64, 2009.
A. Carullo e M. Parvis, “An Ultrasonic Sensor for Distance Measurement in Automotive
Applications,IEEE Sensors Journal, vol. 1, nº 2, p. 143, 2001.
J. Wang, E. Case e D. Manor, “The Road Traffic Microwave Sensor (RTMS),” Nivigation and
Information Systems, 1992.
L. Andreone, P. C. Antonello, M. Bertozzi, A. Broggi, A. Fascioli e D. Ranzato, “Vehicle detection
and localization in infra-red images,” em IEEE 5th International Conference Intelligent
Transportation Systems, 2002.
S. Turner, “Advanced techniques for travel time data collection,” em 6th International VNIS -
Vehicle Navigation and Information Systems, 1995.
J. Fang, H. Meng, H. Zhang e X. Wang, “A Low-cost Vehicle Detection and Classifiation System
Based on Unmodulated Continuous-Wave Radar,” em IEEE Intelligent Transportation Systems
Conference, 2007.
D. B. Work, O.-P. Tossavainen, Q. Jacobson e A. Bayen, “Lagrangian Sensing: Traffic Estimation
with Mobile Devices,” em American Control Conference, 2009.
S.-H. Lee, M. Viswanathan e Y.-K. Yang, “A Hybrid Soft Computing Approach to Link trael Speed
Estimation,” em Lecture Notes in Computer Science - Third International Conference, Xi'an, 2006.
S. Amin, S. Andrews, S. Apte, J. Arnold e J. Ban, “Mobile Century Using GPS Mobile Phones as
Traffic Sensors: A Field Experiment,” em ITS World Congress, New York, NY, 2008.
G. Seo, A. Yazici, U. Ozguner e J. Cho, “An approach for data collection and Traffic Signal
Control in the futuristic city,” em IEEE 10th International Conference on Advanced
Communication Technology, 2008.
K. Wang, Z. Li, Q. Yao, W. Huang e F. Wang, “An automated vehicle counting system for traffic
surveillance,” em IEEE International Conferene on Vehicular Electronics and Safety, 2007.
V. Kastrinaki, “A survey of video processing techniques for traffic applications,Image and Video
Computation, vol. 21, nº 4, pp. 359-81, 2003.
W. Wei, Q. Zhang, M. Wang e Z. Huang, “Detection of Traffic Parameters Based on Computer
Vision and Image Processing,” Information and Control, vol. 30, pp. 257-61, 2001.
S. Kamijo, “Spatio-Temporal MRF model and its Application to Traffic Flow Analyses,” em 21st
International Data Engineering Workshops, Tokyo, 2005.
M. Zweck, I. G. Audi AG e M. Schuch, “Traffic light assistant: Applying cooperative ITS in
European cities and vehicles,” em Connected Vehicles and Expo (ICCVE), 2013 International
Conference on , Las Vegas, NV , 2013.
S. Okada, T. Yendo, T. Yamazato, T. Fujii, M. Tanimoto e Y. Kimura, “On-vehicle receiver for
distant visible light road-to-vehicle communication,” em IEEE Intelligent Vehicles Symposium,
I. Lee, M. Sim e F. Kung, “Performance enhancement of outdoor visible-light communication
system using selective combining receiver,” IET Optoelectronics, vol. 3, nº 1, February 2009.
Z. Belinova, “Self-organizing processes in transportation - can we use them to improve traffic
control?,” Transport Problems, vol. 2, nº 2, pp. 49-54, 2007.
M. O. Cherif, M. Senouci e B. Ducourthial, “A new framework of self-orgnization of vehicular
networks,” em Global Information Infrastructure Symposium, 2009.
A. Kesting, M. Treiber e D. Helbing, “Enhanced Intelligent Driver Model to Access the Impact of
Driving Strategies on Traffic Capacity,” Philosophical Transactions of the Royal Society, vol. A
368, pp. 4585-4605, 2010.
X. Yang e W. Recker, “Simulation studies of information propagation in a self-organizing
distributed traffic information system,” Transportation Research Part C: Emerging Technologies,
vol. 13, pp. 370-390, 2005.
S. Ilgin Guler, M. Menendez e L. Meier, “Using connected vehicle technology to improve the
efficiency of intersections,” Transportation Research Part C: Emerging Technologies, vol. 46, pp.
121-131, 2014.
C. Prehofer e C. Bettstetter, “Self-organization in communication networks: principles and design
paradigms,” IEEE Communications Magazine, vol. 43, nº 7, pp. 78-85, 2005.
W. R. Ashby, “Requisite variety and its implications for the control of complex systems,”
Cybernetic, vol. 1, nº 2, pp. 83-99, 1958.
J. Wu, A. Abbas-Turki e A. E. Moudni, “Intersection traffic control by a novel scheduling model,”
em IEEE/INFORMS International Conference on Service Operations, Logistics, and Information,
C. Berger e B. Rumpe, “Autonomous Driving-5 Years after the Urban Challenge: The Anticipatory
Vehicle as a Cyber-Physical System,” em GI-Jahrestagung, 2012.
A. K. Erlang, “The Theory of Probabilities and Telephone Conversations,” Nyt Tidsskrift for
Matematik B, vol. 20, 1909.
M. Papageorgiou, C. Kiakaki, V. Dinopoulou e A. Kotsialos, “Review of road traffic control
strategies,” em IEEE, 2003.
B. G. Heydecjer e I. W. Dudgeon, “Calculation of Signal settings to minimize delay at a junction,”
em Transportation and Traffic Theory, Elsevier, NY, 1987, 1987, pp. 159-78.
E. Rowe, “The Los Angeles Automated Traffic Surveillance and Control (ATSA) System,” IEEE
Transactions on Vehicular Technology, vol. 40, nº 1, pp. 16-20, 1991.
X.-H. Yu, “An integrated model for signalized traffic intersection control,” em IEEE Conference on
Control Applications, 2005.
J. Wu, A. Abbas-Turki e A. El Moudni, “Discrete Methods for Urban Intersection Traffic
Controlling,” em IEEE Vehicular Technology Conference, 2009.
S. Moyer, “Mr. Trafficlight,” Motor News, vol. 27, pp. 14-5, 1947.
S. F. Bush e S. Goel, “Graph Spectra of Carbon Nanotube Networks,” em 1st International
Conference on Nano-Networks and Workshops, 2006.
S. F. Bush e Y. Li, “Network Characteristics of Carbon Nanotubes: A Graph Eigenspectrum
Approach and Tool Using Mathematica,” GE Global Research, 2006.
S. F. Bush e S. Goel, “An Active Model-Based Prototype for Predictive Network Management,”
IEEE Journal on Selected Areas in Communications, vol. 23, nº 10, pp. 2040-57, 2005.
S. F. Bush e A. B. Kulkami, “Genetically Induced Communication Network Fault Tolerence,”
Complexity Special Issue: Resilient & Adaptive Defense of Computing Networks, vol. 9, nº 2, pp.
19-33, 2003.
H. Suzuki, I. Jun-ichi, and A. Kazuyuki, "Chaotic Ising-like dynamics in traffic signals," Scientific
reports, vol. 3, pp. 1127-1133, 2013.
[225] M. E. Fouladvand, M. R. Shaebani, and Z. Sadjadi, "Intelligent controlling simulation of
traffic flow in a small city network," Journal of the Physical Society of Japan, vol. 73, pp. 3209-
3214, 2004.
[226] T. Le, P. Kovács, N. Walton, H. L. Vu, L. L. Andrew, and S. S. Hoogendoorn, Decentralized
signal control for urban road networks,” Transportation Research Part C: Emerging Technologies,
vol. 58, pp. 431-450, 2015.
[227] J. C. Burguillo-Rial, P. S. Rodríguez-Hernández, E. C. Montenegro, and F. G. Castiñeira,
History-based self-organizing traffic lights,” Computing and Informatics, vol. 28, pp. 157-168,
[228] H. Wang, Y. X. Liu, "Self-Organized Traffic Signal Coordinated Control Based on
Interactive and Distributed Subarea," Applied Mechanics and Materials, vols. 241-244, pp. 2031-
2037, 2013.
[229] B. Płaczek, “A self-organizing system for urban traffic control based on predictive interval
microscopic model,” Engineering Applications of Artificial Intelligence, vol. 34, pp. 75-84, 2014.
[230] R. Tachet, P. Santi, S. Sobolevsky, L. Reyes-Castro, E. Frazzoli, D. Helbing, et al.,
“Revisiting Street Intersections Using Slot-Based Systems,” PLoS ONE, vol. 11, no. 3, 2016.
[231] L. Zhang, T. M. Garoni, and J. Gier, “A comparative study of Macroscopic Fundamental
Diagrams of arterial road networks governed by adaptive traffic signal systems,” Transportation
Research Part B, vol. 49, pp. 1-23, 2013.
Sanjay Goel is an Associate Professor in the School of Business and the Director of
Research at the NYS Center for Information Forensics and Assurance at UAlbany. He
represents UAlbany in the Capital Region Cyber Crime Partnership. Dr. Goel received his
Ph.D. in Mechanical Engineering from RPI. His research includes information security, risk
analysis, security policies, information classification, cyber warfare and self-organization in
complex systems. His latest research on self-organizing systems includes traffic light
coordination, nano-bio computing and social networks. He and his team have worked with CSCIC in developing the
information classification policy for New York. He is currently leading an effort launched by IEEE Communications
Society and the IEEE Standards Association to create a vision for the Smart Grid future 15 years ahead. He won the
promising Inventor’s Award in 2005 from the SUNY Research Foundation. In 2006, he was awarded the SUNY
Chancellor’s Award for Excellence in Teaching, the UAlbany Excellence in Teaching Award, and the Graduate Student
Organization Award for Faculty Mentoring. He was named an AT&T Industrial Ecology Faculty Fellow for 2009-2010.
He has received grants and funding from NIJ, NSF, UTRC, NYSERDA, US Department of Education, and CSCIC.
Stephen F. Bush is a researcher in Algorithmic Communications Network Theory at the GE
Global Research Center. Dr. Bush was presented with a Gold Cup Trophy Award from
DARPA for his work in fault tolerant networking. Stephen F. Bush received the B.S. degree in
electrical and computer engineering from Carnegie Mellon University, M.S. degree in
computer science from Cleveland State University, and Ph.D. degree from the University of Kansas. He is currently a
researcher at General Electric Global Research, Niskayuna, NY. He is the author of Nanoscale Communication
Networks (Norwood, MA: Artech House, 2010). He coauthored a book on active network management, titled Active
Networks and Active Network Management: A Proactive Management Framework (New York, NY: Kluwer
Academic/Plenum Publishers, 2001). He is an internationally recognized researcher in Active Networking and
Algorithmic Communications Networking Theory with over 75 peer-reviewed publications. Dr. Bush is the past chair of
the IEEE Emerging Technical Subcommittee on Nanoscale, Molecular, and Quantum Networking and currently chair
for the IEEE 1906.1 standards working group on nanoscale communication networks. Dr. Bush is an IEEE
Distinguished Lecturer on the smart grid and nanoscale communication networks. He is also on the steering committee
for the IEEE Smart Grid Vision Project.
Carlos Gershenson is a full time research professor at the Instituto de Investigaciones en
Matemáticas Aplicadas y en Sistemas at the Universidad Nacional Autónoma de México
(UNAM), where he leads the Self-organizing Systems Lab. He is also affiliated
researcher and member of the directive council at the Center for Complexity Sciences at
UNAM. He is Visiting Professor at the Massachussetts Institute of Technology and at
Northeastern University. He has a wide variety of academic interests, including self-organizing systems, artificial
life, evolution, complexity, cognition, artificial societies, and philosophy. He is Editor-in-Chief of Complexity
Digest ( and Complexity-at-large editor for the journal Complexity.
... Existing studies on fixed-schedule reversible lanes have direction changing periods of 1-6 hours (Claes et al., 2011;Gershenson & Rosenblueth, 2012;Goel, Bush, & Gershenson, 2017). As noted by Hausknecht et al. (2011), without consideration of vehicle safety, more frequent lane direction switches leads to performance improvements. ...
... The control frequency of these schemes is low and these approaches are unable to cope with traffic flow changes satisfactorily (Gershenson & Rosenblueth, 2012). The frequencies of these systems are greatly lower than that of real-time adaptive control systems widely used at signalized intersections (Goel et al., 2017). ...
Full-text available
In recent years, large-scale testing has begun for connected and autonomous driving, making it possible to implement the concept of Dynamic Lane Reversal (DLR), which can quickly shift lane directions to reflect instantaneous flow dynamics. DLR is to make full use of road space and avoid waste of road capacity, and potentially alleviate congestion. However, the impact and feasibility of DLR remain unclear. In order to investigate the effectiveness, the feasibility, and the applicability of DLR, we utilize a direction-changeable, lane-based cell transmission model to find an optimal DLR scheme for a roadway segment with stochastic traffic flow in both directions. A proxy model is also designed to realize DLR in VISSIM. A regression analysis was carried out to find the impacts of directional flow rate and the number of lanes on delay reductions due to DLR. Results indicate that implementing DLR can reduce the total queuing delay considerably compared to traditional reversible lane strategies. We found that DLR achieved superior performance on segments with more lanes and when the flows from each direction were close to one another. A zigzag frontier of the delay reduction was discovered. Findings from this research shed light on the feasibility and effectiveness of DLR on various types of road segments.
... The cycles are not considered as tools for adaptive real-time control of the traffic behavior. For the evolution of the traffic signal control, one can refer to [4]. The simplest form of control is applied to isolated intersections under slow fluctuations of the traffic intensity [5]. ...
Full-text available
A bi-level model for traffic signal optimization is developed. The model predictive framework is applied for traffic control in an urban traffic network. The potential of the bi-level formalization is used to increase the space of control influences with simultaneous evaluation of the green light and cycle durations. Thus, the increased control space allows more traffic parameters to be considered, such as vehicles queues and traffic flows. A particular modification of the bi-level control is applied for the synchronization of the traffic lights in the network. The model predictive approach is used for the real-time management of the traffic in the network. The control implementations are constrained by the shortest evaluated cycle. Thus, a synchronization of the traffic lights is applied for the minimization of the queues and maximization of the outgoing flows from the network. The bi-level model has been numerically tested on a set of intensive crossroads in Sofia. The numerical simulations prove the superiority of the developed bi-level control in comparison with the classical optimization of queue lengths.
... Application of bi-level optimization in the domain of transportation policies one can find also in [23]; for network transportation -in [24]; for special kind of goods and their transportation in [25], [26]; for intermodal transport - [27] , [28] ; for locating of logistics in [29]. Timing considerations in transportation are addressed in [4], [7], [30], [31]. ...
... The timing of traffic signal lights has attracted many researchers to study the problems involving traffic light management and looking for an inexpensive and effective solution that requires inexpensive changes in the infrastructures of city areas (Babicheva, 2015;Covell et al., 2015;Goel et al., 2017;Pop, 2018). A simple traffic light system uses a fixed control setting based on the latest traffic data, and the setting could be manually changed. ...
Full-text available
Traffic signal lights system is a signalling device located an intersection or pedestrian crossing to control the movement of traffic. The timing of traffic signal lights has attracted many researchers to study the problems involving traffic light management and looking for an inexpensive and effective solution that requires inexpensive changes in the infrastructures. A simple traffic lights system uses a pre-timed control setting based on the latest traffic data, and the setting could be manually changed. It is a common type of signal control and sometimes the setting was not correctly configured with the traffic data, thus leading to congestion at an intersection. Many mathematical strategies were applied to get an optimal setting. This study aims to model the traffic flow at Persiaran Kayangan and Persiaran Permai Intersection, Section 7, Shah Alam, as the case study, by using AnyLogic simulation software. The model was used to determine the best timings of traffic green lights that minimise the average time at the intersection and reduce traffic congestion. The findings showed that the best timings of traffic green lights for four directions at the intersection are 120 seconds, 75 seconds, 130 seconds and 100 seconds, respectively. These timings of green lights produced the lowest average time at the intersection (55.65 seconds).
... This is usually better than not having adaptation, where the best possible option would be to take average measurements, set fixed phases, and perhaps change the programs a few times per day. However, if traffic lights can adapt at the same timescale as the traffic demand does, i.e. every cycle, then the performance would be much improved (Goel et al., 2017). ...
Full-text available
Self-organization offers a promising approach for designing adaptive systems. Given the inherent complexity of most cyber-physical systems, adaptivity is desired, as predictability is limited. Here I summarize different concepts and approaches that can facilitate self-organization in cyber-physical systems, and thus be exploited for design. Then I mention real-world examples of systems where self-organization has managed to provide solutions that outperform classical approaches, in particular related to urban mobility. Finally, I identify when a centralized, distributed, or self-organizing control is more appropriate.
Full-text available
Self-organization offers a promising approach for designing adaptive systems. Given the inherent complexity of most cyber-physical systems, adaptivity is desired, as predictability is limited. Here I summarize different concepts and approaches that can facilitate self-organization in cyber-physical systems, and thus be exploited for design. Then I mention real-world examples of systems where self-organization has managed to provide solutions that outperform classical approaches, in particular related to urban mobility. Finally, I identify when a centralized, distributed, or self-organizing control is more appropriate.
Full-text available
with the increasing vehicle ratio on our roads, a lot of factors such as pollution, time constraints and environmental factors need to be addressed. One main issue to be addressed is traffic congestion during the peak hours. This issue affects drivers in numerous ways including loss of productive working hours by lining up in traffic queue. It also leads to loss of natural resource such as fossil fuel used by the vehicle engine while the car is running but lining up in traffic. In this paper, we propose an intelligent sensor-based traffic light control system. To analyze and compare performance, the system was implemented in MATLAB/SIMULINK and simulation results demonstrate the feasibility and high accuracy of the proposed intelligent traffic light system.
Conference Paper
Full-text available
The objective of this paper is to present a detailed description of using DSP board and image processing techniques to construct an automated vehicle counting system. Such a system has many potential applications, such as traffic signal control and district traffic abduction. We use TI TMS320DM642 DSP as the computational unit to avoid heavy investment in industrial control computer while obtaining improved computational power and optimized system structure. The overall software is comprised of two parts: embedded DSP software and host PC software. The embedded DSP software acquires the video image from stationary cameras, detects and counts moving vehicles, and transmits the processing results and real-time images after compression to PC software through network. The host PC software works as a graphic user interface through which the end user can configure the DSP board parameters and access the video processing results. The vehicle detection and counting algorithm is carefully devised to keep robust and efficient in traffic scenes for longtime span and with changeful illumination. Experimental results show that the proposed system performs well in actual traffic scenes, and the processing speed and accuracy of the system can meet the requirement of practical applications.
Full-text available
Many intersections have varying mechanism for vehicles right of way as they approach the intersection. With actuated signal, induction loops buried in the roadway stop-line, video, infrared or microwave detection system automatically adjusts timings relative to prevailing degree of saturation. An intersection without such detection system operates on fixed times (static). Signal settings are based on fixed proportional distribution of effective green per cycle time. In the paper, daylight and dry weather traffic performances at standalone signalised 4-way intersection were investigated under actuated and optimised signal timing conditions. Based on the hypotheses that peak traffic performance at standalone between optimised static and actuated signal settings are insignificant; discharge rates, delays and effective green timings for both were estimated compared and contrasted. Given that an optimised static signal assigns predetermined time irrespective of traffic demand; saturation flows were fixed at 1900 for straight, 1800 left turning and 1700 right turning vehicles per hour per lane respectively. Results show marginal differences in peak period effective green, discharge rates and delays. The paper concluded that optimised static signal can produce good results and should also be considered especially at standalone intersections where traffic operations are at peak regularly.
Full-text available
Estimation of delay at signalized intersections is a complex process and depends on a number of parameters, among which the degree of saturation (x = v/c) is the most important. This paper presents a new methodology for estimating delay parameter κ, and proposes an analytical delay model for signalized intersections that considers the variation in traffic flow. Unlike the existing delay models in the literature, the delay parameter k is not a fixed value and is expressed as a function of the degree of saturation. The proposed model is applicable to the entire range of expected operation including highly oversaturated traffic conditions. A comparative study of the proposed model against the existing models was performed to verify the improvements by the proposed model. Later, the TRAF-NETSIM (Traffic Network Simulation) microscopic simulation model was used in the calibration and verification of the new delay model for oversaturated conditions. The simulation was conducted in 3 consecutive time periods to reflect fluctuation in traffic demand. The relationship between the simulated and proposed delay model results were analyzed for 48 different scenarios.
Conference Paper
Problems of delay and resultant driver frustration continue to get worse as congestion increases in large cities. The delays can be minimized by optimization of traffic signals. Traffic signal optimization is a complex problem making improvements difficult. Recent advances in sensors and communication have enabled the use of novel algorithms for improving traffic signal efficiency. This paper investigates the use of a distributed traffic signal control algorithm based on the concepts self-organization. The algorithm is benchmarked against traditional traffic signal algorithms. The self-organization based algorithms perform significantly better compared to the existing algorithms. A simulation model was created based on an abstraction of city road with multiple intersections using data on actual traffic counts. The results are presented in this paper along with the details of the algorithm.