DECISION SUPPORT FOR DYNAMIC CITY TRAFFIC
MANAGEMENT USING VEHICULAR COMMUNICATION
Jana G¨ormer
1
, Jan Fabian Ehmke
2
, Maksims Fiosins
1
, Daniel Schmidt
3
,
Henrik Schumacher
4
and Hugues Tchouankem
4∗
1
Department of Informatics, Clausthal University of Technology, Clausthal-Zellerfeld, Germany
2
Decision Support Group, Technische Universit¨at Braunschweig, Braunschweig, Germany
3
Institute of Transportation and Urban Engineering, Technische Universit¨at Braunschweig, Braunschweig, Germany
4
Institute of Communications Technology, Leibniz Universit¨at Hannover, Hannover, Germany
Keywords:
Traffic simulation, Communication simulation, Vehicular communication, Multiagent systems, Agent organi-
zation, Decentralized decision making, Dynamic city traffic management.
Abstract:
In this paper, we present an integrated simulation approach featuring centralized and decentralized traffic
management in urban areas. Our aim is to improve traffic flows by dynamic traffic management which is
supported by vehicular communication interlinking centralized and decentralized decision making. We focus
on traffic state estimation and the optimization of traffic lights as a central component to influence local traffic
states, while individual traffic participants’ behavior is modeled by multiagent systems. Traffic participants
achieve their individual goals by formation of groups and improving their knowledge about the road network
by means of learning. Modeling of vehicular communication takes into account specific characteristics of
urban areas, ensuring the realistic collection and dissemination of (de)centralized information.
We provide a comprehensive microscopic traffic simulation framework featuring innovative functionality re-
garding dynamic traffic management, decentralized decision making as well as realistic communication mod-
eling. To illustrate and validate our approach, we present a use case in a city scenario. Simulations are
implemented based on the microscopic traffic simulator AIMSUN, which is significantly extended using the
AIMSUN API.
1 INTRODUCTION
The 21st century will be a century of urbanization,
since cities provide more attractive opportunities for
employment, education, cultural and sports activities.
However, increasing traffic flows within limited city
space lead to negative effects in terms of emissions
and traffic jams.
Dynamic traffic management in cities may be the
key for efficientcontrol of urban traffic flows. It refers
to a broad concept which aims at the collection of traf-
fic data and the real-time control of traffic flows by
dissemination of traffic information. Most approaches
follow a centralized perspective, which is based on
data collection and information processing in central-
∗
The authors are grateful to the Lower Saxony Univer-
sity of Technology (NTH) for funding our work within the
project ”Planning and Decision Making in Networks of Au-
tonomous Actors in Traffic (PLANETS)”.
ized traffic management centers. Here, traffic control
strategies are usually designed area-wide.
The centralized approach requires a large amount
of information to be transmitted to control centers in
the form of network data and back in the form of con-
trol actions. This requires a very powerful and re-
liable communication network infrastructure. In the
case of a network failure the centralized management
becomes unavailable.
The alternative approach to centralized traffic
management is the decentralized one. Here, traffic
participants act according to local information, i.e.
they optimize their behavior based on local data.
This approach can be more cost-effective and ro-
bust against communication network failures.
In our work, we focus on the combination and
alignment of centralized and decentralized traffic con-
trol methods in order to improve the efficiency of
traffic management. Vehicular communication serves
327
Görmer J., Fabian Ehmke J., Fiosins M., Schmidt D., Schumacher H. and Tchouankem H..
DECISION SUPPORT FOR DYNAMIC CITY TRAFFIC MANAGEMENT USING VEHICULAR COMMUNICATION.
DOI: 10.5220/0003598503270332
In Proceedings of 1st International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2011), pages
327-332
ISBN: 978-989-8425-78-2
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

as a fundament for the linking of centralized con-
trol strategies and traffic participants. Decentralized
decision making is investigated in terms of multi-
agent systems. We examine the alignment of central-
ized and decentralized traffic control by an integrated
simulation of city traffic, communication and multi-
agent modeling. This paper focuses on the required
methodology and the enhancement of a traffic simu-
lation environment that corresponds to these require-
ments. First results illustrate the necessity of an inte-
grated approach.
This paper is organized as follows. Section 2
describes role and tasks of dynamic traffic manage-
ment. Fundamentals of decentralized decision mak-
ing with respect to multiagent systems are discussed
in Section 3. Basics of modeling and simulation of
vehicular communication are presented in Section 4.
The integration of methods in a simulation environ-
ment, the description of a use case and preliminary
results are shown in Section 5. Finally, Section 6 con-
cludes this paper.
2 DYNAMIC TRAFFIC
MANAGEMENT
In urban areas, infrastructure often presents a bottle-
neck due to traffic demands exceeding available road
capacities. Here, dynamic traffic management is nec-
essary in order to provide strategies for efficient con-
trol, e.g. by applying time-dependent signal plans
considering the respective day of the week or hour of
the day to incorporate the actual traffic situation.
The traffic load typically varies strongly through-
out the day. Therefore an adaptation of the traffic
management according to the actual demand is es-
sential. Well-known dynamic traffic management so-
lutions include variable traffic signs and variable di-
rection signs at motorways. To enable dynamic ap-
proaches within urban areas, a detailed knowledge of
the actual traffic state is essential (see Section 2.1).
Furthermore, an optimization of traffic signal plans is
required (see Section 2.2).
2.1 Traffic State Estimation
A fundamental requirement for every kind of traffic-
dependent control is the availability of information
about the actual traffic state. In the following, we fo-
cus on agent-based methods for traffic coordination
and management to meet this challenge.
Within the work presented here we use the ca-
pabilities of communicating vehicles for traffic state
estimation in terms of level of service, LOS (FGSV,
2001). By collecting messages which are created by
vehicles equipped with communication systems at the
traffic management center (TMC), we measure the
mean travel time on a road segment. In particular,
we compute the traffic state at every road (section)
for each time interval (e.g. a cycle, every 5 minutes,
etc.) according to the following algorithm:
1. Collect the timestamps t
in,i
and t
out,i
for the vehi-
cle i entering and leaving a given road section.
2. Compute the mean real travel time t
real
t
real
=
1
n
·
n
∑
i=1
(t
out,i
− t
in,i
),
where n is the number of vehicles on a section.
3. Compute the ideal travel time t
ideal
=
ℓ
v
opt
, which
is the time a vehicle would require to traverse the
section during free, uncongested traffic (where ℓ
is the section length and v
opt
the optimal speed).
4. Calculate the mean delay time t
del
= t
real
− t
ideal
.
5. Identify the LOS by using the classification of t
del
according to the following table (FGSV, 2001):
t
del
[sec] <20 <35 <50 <70 <100 >100
LOS A B C D E F
2.2 Optimization of Traffic Lights
Dynamic traffic management can be realized by
means of centralized and decentralized control. Us-
ing centralized control, data is collected in a traffic
management center (TMC) in order to provide a con-
sistent picture of the actual traffic state. Thus, goals
like overall traffic quality and incident management
can be pursued by implementing strategies via traffic
infrastructure components. Using decentralized con-
trol, infrastructure elements like single traffic lights
make simple local decisions which are usually not
balanced with decisions made in the neighborhood.
The objective of traffic management systems is
the use of information infrastructure to detect dif-
ferent states of traffic flow and to react accordingly
in order to preserve or improve the supervised net-
work’s overall performance. In our work we com-
bine a new dynamic traffic control approach, reduc-
tion of intergreen times by elimination of phases and
phase changes, with well known procedures of traffic
planning (signal plan adjustment (FGSV, 2010)) and
traffic control (signal plan transition (Shelby et al.,
2006)).
SIMULTECH 2011 - 1st International Conference on Simulation and Modeling Methodologies, Technologies and
Applications
328

2.2.1 Determination of Nonessential Phases
A signal plan consists of phases and phase changes.
The duration of the phase change is determined by the
decisive intergreen time, i.e. the longest intergreen
time among all combinations of ending and starting
signal groups.
A recent study shows that there is a potential
of a capacity increase up to 7 % by reducing inter-
green times and thus phase change times (Boltze and
Wolfermann, 2011). To reach this potential, a precise
computation method for intergreen times compared
to the state of the art (FGSV, 2010) is needed. An-
other way to reduce intergreen times is the elimina-
tion of one or more phases and corresponding time-
consuming phase changes. Using this approach, the
validity of the remaining signal plan has to be assured:
There has to be either a green time or a close alterna-
tive route for every traffic participant.
We test every possible phase to beeliminated from
the signal plan. For this purpose we
• eliminate a phase from the actual signal plan and
in doing so eliminate some corresponding turn-
ings,
• check whether the new signal plan is valid (see
above),
• determine an optimal signal plan and
• calculate the benefit.
The benefit of a new signal plan can be measured
in terms of capacity increase or performance improve-
ment (e.g. LOS improvement). At the end of this pro-
cedure, the best valid signal plan is chosen.
3 MULTIAGENT
ORGANIZATION FOR
DECISION SUPPORT
In this Section we describe the model of the vehicles’
behavour on the individual level. Multiagent systems
(MAS) constitute an appropriate and most commonly
used model for decentralized systems modeling, in-
cluding traffic systems. In (Fiosins et al., 2011) we
presented a two-stage multiagent planning approach
for vehicle agents in urban traffic: During strategic
planning the vehicles plan their optimal routes, dur-
ing tactical planning they make decisions about speed
regulation and lane changes. The amount of commu-
nication between agents and the TMC should also be
taken into account in order to make the model realis-
tic.
3.1 Individual Learning
We begin with a description of the strategic planning
(optimal route selection): The environment is repre-
sented as a sear graph G = (V,E) (Bergenthal et al.,
2004), where nodes V correspond to the streets, the
edges E to the turnings, which connect these streets.
The strategic planning process of the agent j is
based on its individual weights c
ind
j
(t), which corre-
spond to the edges of G. We suppose that there are k
classes of the agents, which differ by the route prefer-
ences (initial individual weights).
Each vehicle agent is equipped by a communica-
tion system, which allows it to receive the information
from the TMC. It receives two types of the informa-
tion: strategic weights c
str
(t) and actual travel times
c
act
(t). The agent updates its individual weights as
c
ind
j
(t + 1) = U
j
(c
ind
j
(t), c
tmc
(t)).
The function U
j
can be a linear combination of a
vehicle and TMC weights or act similar to Q-learning
approach. According to the individual weights, the
shortest path in the search graph is constructed.
In the following, we describe the tactical planning
approach. According to its strategic plan, a vehicle
agent enters a street and then learns how to use it.
The state s
j
(t) of the agent j is described
by a tuple, consisting of a distance to the
end of the street x
j
(t), lane l
j
(t), speed v
j
(t)
and the time from the last traffic light re-
peat u
j
(t). Possible actions consist of pairs
a = h∆v, ∆li ∈ A, which correspond to speed and lane
change. We assume that for each state s
j
(t) ∈ S a cor-
responding reward r(s
j
) is available.
We introduce a state-action value function
Q
j
(s
j
,a
j
) (Sutton and Barto, 1998), which represents
an average reward for the j-th agent, which starts
from the state s
j
∈ S and performs the action a
j
∈ A.
Then for the state-action pair (s
j
,a
j
) and the next
state s
′ j
, there exists the following recurrent equation:
Q
j
(s
j
,a
j
) = Q
j
(s
j
,a
j
)+
+α
j

r(s
′ j
) + γ
j
max
a
′ j
Q
j
(s
′ j
,a
′ j
) − Q
j
(s
j
,a
j
)

,
where α
j
and γ
j
are learning parameters.
3.2 Grouping
Agents join groups (Song et al., 2007) in order to
reach individual goals in a more efficient way by ad-
justing speed and possible lanes, like getting from A
to B with a low travel time. Groups need to be moti-
vated and are usually formed by common goals. For
motivation of vehicle agents we use reward functions
and the group rewards are defined to be higher than re-
wards for fulfilment of individual goals. Once a goal
DECISION SUPPORT FOR DYNAMIC CITY TRAFFIC MANAGEMENT USING VEHICULAR COMMUNICATION
329

is adopted, an agent cannot drop it freely; the agent
must keep the goal until it is fulfilled, unfulfillable or
irrelevant.
Let R(S, A) be the joint reward func-
tion for the joint state S and joint action A,
which include all agents in the system. Let
τ = {τ
1
,τ
2
,. .. ,τ
k
} be the set of groups and
R
τ
i
(S
τ
i
,A
τ
i
) be the reward of the group τ
i
, where the
arguments S
τ
i
and A
τ
i
are the states and the actions of
the group τ
i
. Note that sometimes we write R
τ
i
(S, A);
in this case, the group reward function depends on
the group agent states and actions.
The Difference Utility (Reward) concept
(Agogino and Tumer, 2004) is usable in the
case of partially observable domains.
We call the system factored, if the group rewards
have the following property:
R
τ
i
(S, A) ≥ R
τ
i
(S, A
′
) ⇔ R(S, A) ≥ R(S, A
′
). (1)
In order to estimate the group formation effi-
ciency, the difference reward function DR
τ
i
for the
group τ
i
may be used. It is defined as
DR
τ
i
(S, A) = R(S, A) − R(S, A
−τ
i
+ A
c
τ
i
), (2)
where A
−τ
i
is the joint action of the agents, not af-
fected by the group τ
i
, but A
c
τ
i
is the joint action of the
agents in the group τ
i
, replaced by constant. Usually
A
c
τ
i
is taken as equivalent to removing the agents of
the group τ
i
from the system.
This investigation focuses on road networks and
grouping is done on a sequence of regulated intersec-
tions on one (linear) road. The vehicles intend to pass
several successive intersections without stops trying
to minimize their individual travel times, whereas an
optimal throughput in the network is desired by the
centralized traffic management and traffic lights can
be regulated accordingly. The idea is to create a
preference for vehicles to join groups and act coordi-
nated, in form of a “green wave” priority, which can
be found in public city transport nowadays. In this
way, an interaction between the centralized and the
decentralized approach is created which is beneficial
for both.
Another benefit of group formation is a simplified
control: Only the whole group or the group leader has
to make decisions, other group members only con-
tribute to the achievement of group goals, perform-
ing only local optimization. Furthermore, information
sharing between group members is possible.
4 VEHICULAR
COMMUNICATION
Wireless communication between vehicles and their
environment opens up a new dimension of innova-
tive applications which will increase safety and com-
fort for the driver in the future. Possible network
topologies include infrastructure networks (Vehicle-
to-Infrastructure, V2I) as well as Ad-Hoc networks
(Vehicle-to-Vehicle, V2V). However, in contrast to
conventional wireless LANs, these topologies are
highly dynamic and present a variety of challenges
for wireless communication.
Multiagent systems are usually modeled indepen-
dently from the underlying communication technol-
ogy and corresponding models assume perfect infor-
mation transfer between agents. However, the wire-
less channel causes limitations in terms of communi-
cation reliability and capacity which have to be con-
sidered as they are inherent to IEEE 802.11 wireless
LANs. Therefore, we use a dedicated communication
model for our investigations.
4.1 Communication Model
In order to simulate large scenarios with a manage-
able complexity, we make use of the traffic simula-
tor’s API module to build our own communication
model specifically for urban environments.
4.1.1 Application Model
We assume that vehicles are equipped with commu-
nication systems that periodically broadcast Cooper-
ative Awareness Messages (CAMs) using a repetition
interval of 500 ms. These CAMs contain status in-
formation (e.g. position, speed and driving direction)
which is used to detect the presence of neighboring
vehicles and to estimate the current traffic conditions
without need for any additional traffic detectors.
4.1.2 Radio Propagation Model
Typically, wireless network simulators assume a
generic propagation model, such as the Free Space or
Two-Ray Ground reflection model. While the former
is a completely idealized model, the latter considers
the effect of earth surface reflection and can be more
accurate. However, neither of them considers the ef-
fects of the surrounding topology on radio propaga-
tion, which is especially important in urban environ-
ments. Therefore, models which capture predictable
shadowing effects are appropriate for modeling urban
vehicular communication, where the effects of build-
ings should be taken into account.
SIMULTECH 2011 - 1st International Conference on Simulation and Modeling Methodologies, Technologies and
Applications
330

In (Sommer et al., 2010), the authors present a
computationally inexpensive simulation model for ra-
dio shadowing in urban environments based on real
world measurements, which comprises an estimation
of the effects that buildings have on the radio commu-
nication between vehicles. We combine their general
model with a Nagakami propagation model for deter-
mining the received signal power level.
Based on the calculation of received signal power
at each receiver, the arrived packets are determined
to be successfully received or lost. Our model deter-
mines the received power P
r
at a certain distance d:
P
r
[dBm] = 10log
10

X

m,
P
r,FS
m

− X
obs
, (3)
where X(m,
P
r,FS
m
) is a random variable following a
Gamma distribution with shape parameter m and scale
parameter
P
r,FS
m
describing the Nakagami multipath fa-
ding component. P
r,FS
is the received signal power
according to the deterministic free space path loss:
P
r,FS
[mW] = P
t
G
t
G
r
λ
2
(4π)
2
d
α
L
,
where P
t
represents the transmission power, G
t
and
G
r
the antenna gains and λ the carrier wavelength, d
is the linear distance between transmitter and receiver,
α the path loss coefficient and L a system loss factor.
The term X
obs
in equation 3 describes the addi-
tional attenuation of a transmission due to an obstacle
as introduced in (Sommer et al., 2010).
4.1.3 Medium Access Control Model
Vehicular communication relies on a wireless chan-
nel which is shared by all stations in the network, so
access to the shared channel needs to be coordinated
to avoid collisions. Medium access in vehicular net-
works is based on the carrier sense multiple access
with collision avoidance (CSMA/CA), but is subject
to some modifications, like communication without
prior association or authentication with a basic ser-
vice set (BSS) (ETSI, 2010).
As our communication model currently directly
interacts with the AIMSUN API (see Section 5), com-
munication modeling is restricted to AIMSUN’s gran-
ularity of simulation time, which provides a minimum
time step length of 100 ms, while frame durations and
MAC timings in IEEE 802.11 are in the range of µs.
Therefore the current version of the implementation
does not model medium access and collisions yet.
5 USE CASE AND
IMPLEMENTATION
In the following, we describe a use case comprising
efficient routing in urban road networks by central-
ized and decentralized decision support. Communi-
cation ensures distribution of information required for
centralized and decentralized decisions. We focus on
the implementation of functionality which integrates
traffic and realistic communication simulation.
The use case refers to the investigation of the im-
pact of centralized and decentralized decision support
during the morning rush hour within a part of the city
road network of Hannover, Germany. Here, at least
one junction is regularly overloaded, leading to traf-
fic jams and significant extensions of individual travel
times. Our aim is to automatically identify junc-
tions suffering from bad traffic quality and dynami-
cally adjust their traffic signal programs (see Section
2). Then, a centrally predefined rerouting strategy is
selected and communicated to the individual vehicles,
which may react to this new information and redefine
their route through the network by taking into account
their individual cognition of the traffic state (see Sec-
tion 3). Thus, congestion at a crowded junction may
be alleviated by spacious rerouting.
A model of the road network in the southern part
of Hannover is implemented in the traffic simulation
software AIMSUN. It allows detailed modeling of
city road traffic in terms of road infrastructure, be-
havior of the vehicles, traffic light control etc., being
embeddded in a microscopic traffic simulation. We
parameterize traffic flows and traffic signal programs
according to data from empirical traffic data collec-
tion as well as control programs being in operation.
While AIMSUN features precise modeling of single
vehicles, traffic lights and urban road infrastructure,
the simulation of vehicular communication and deci-
sion making is not supported readily.
In order to establish decentralized decision sup-
port, a modular software architecture has been devel-
oped. Each vehicle has a navigation system, contain-
ing a graph representation of the road network, a com-
munication module for reception and broadcasting of
V2V and V2I messages, a set of applications used for
decentralized decision making in terms of grouping
as well as individual routing based on updated infor-
mation by the centralized traffic management and a
learning application which features continuous obser-
vation of the vehicle’s anticipated and realized behav-
ior and thus updates its knowledge about the network.
First simulation results allow insights into effi-
ciency and applicability of the simulation framework.
Figure 1 shows the effective number of nodes in the
DECISION SUPPORT FOR DYNAMIC CITY TRAFFIC MANAGEMENT USING VEHICULAR COMMUNICATION
331

communication range of a specific node while it tra-
verses the scenario when using a radio propagation
model with and without consideration of obstacles.
Results obtained are based on the above mentioned
communication model (see Section 4). Compared
to a model without obstacle consideration, which
mainly depicts an ideal communication environment,
a more realistic model leads to a significant difference
of vehicles in communication range (at maximum:
145). This insight is crucial for agents’ decision mak-
ing, which usually assumes ideal communication and
availability of information. In sum, the consideration
of buildings significantly affects dynamic traffic man-
agement applications in city road networks.
0 20 40 60 80 100 120 140 160 180 200 220 240
0
50
100
150
200
time [s]
number of neighbors
without buildings
with buildings
145 neighbors
Figure 1: Mean number of neighbors with/without consid-
eration of buildings
6 CONCLUSIONS AND FUTURE
WORK
We have presented an integrated approach for simula-
tion of central and decentralized traffic management
in urban areas. Vehicular communication improves
central and enables decentral decision making.
A microscopic traffic simulation tool has been en-
hanced by dynamic traffic management functionality,
multiagent decision support and vehicular communi-
cation. First results show the importance of a realistic
simulation model for analyzing (de)centralized traffic
management in an urban context.
This paper presents work in progress. The archi-
tecture of the framework presented allows simple in-
tegration and investigation of future services for traf-
fic participants. Applications may be investigated
in terms of traffic performance, technology require-
ments and the reasonable level of centralized and de-
centralized control.
REFERENCES
Agogino, A. and Tumer, K. (2004). Team formation in par-
tially observable multi-agent systems. In Proceedings
of the International Joint Conference on Neural Net-
works, Budapest, Hungary.
Bergenthal, T., Frommer, A., and Paulerberg, D. (2004).
Wege auf Graphen. Mathe Prisma: Fachbereich C
/ Mathematik der Bergischen Universit¨a t Wuppertal.
Boltze, M. and Wolfermann, A. (2011). Der Einfluss von
Zwischenzeiten auf die Kapazit¨at von Knotenpunk-
ten mit Lichtsignalanlage. HEUREKA ’11, Stuttgart,
Germany.
ETSI (2010). Intelligent Transport Systems (ITS); Euro-
pean profile standard for the physical and medium ac-
cess control layer of Intelligent Transport Systems op-
erating in the 5 GHz frequency band. ETSI ES 202
663 V1.1.0 (2010-01).
FGSV (2001). Handbuch f¨ur die Bemessung von Straßen-
verkehrsanlagen HBS (Manual for the dimensioning
of road infrastructure). FGSV-Verlag, Cologne, Ger-
many.
FGSV (2010). Richtlinien f¨ur Lichtsignalanlagen RiLSA
(Guidelines for Traffic Signals). FGSV-Verlag,
Cologne, Germany.
Fiosins, M., Fiosina, J., M¨uller, J., and G¨ormer, J.
(2011). Agent-based integrated decision making for
autonomous vehicles in urban traffic. In Demazeau,
Y., Pechoucek, M., Corchado, J., and P´erez, J., editors,
Advances on Practical Applications of Agents and
Multiagent Systems, volume 88 of Advances in Intel-
ligent and Soft Computing, pages 173–178. Springer
Berlin / Heidelberg.
Shelby, S., Bullock, D., and Gettman, D. (2006). Transi-
tion methods in traffic signal control. Transportation
Research Record: Journal of the Transportation Re-
search Board, pages 130–140.
Sommer, C., Eckhoff, D., German, R., and Dressler, F.
(2010). A computationally inexpensive empirical
model of ieee 802.11p radio shadowing in urban en-
vironments. In Technical Report CS-2010-06, Univer-
sit¨at Erlangen-N¨urnberg, Erlangen.
Song, S. K., Han, S., and Youn, H. Y. (2007). A new
agent platform architecture supporting the agent group
paradigm for multi-agent systems. In IAT ’07: Pro-
ceedings of the 2007 IEEE/WIC/ACM International
Conference on Intelligent Agent Technology, pages
399–402, Washington, DC, USA. IEEE Computer So-
ciety.
Sutton, R. S. and Barto, A. G. (1998). Reinforcement Learn-
ing: An Introduction (Adaptive Computation and Ma-
chine Learning). The MIT Press.
SIMULTECH 2011 - 1st International Conference on Simulation and Modeling Methodologies, Technologies and
Applications
332