International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.

org/systems_and_measurements/

128

Recent Development and Applications of SUMO – Simulation of Urban MObility

Daniel Krajzewicz, Jakob Erdmann, Michael Behrisch, and Laura Bieker

Institute of Transportation Systems
German Aerospace Center
Rutherfordstr. 2, 12489 Berlin, Germany
e-mail: Jakob.Erdmann@dlr.de, Daniel.Krajzewicz@dlr.de, Michael.Behrisch@dlr.de, Laura.Bieker@dlr.de

Abstract—SUMO is an open source traffic simulation package they help in preparing and performing a traffic simulation.
including the simulation application itself as well as supporting Then, major research topics, which can be addressed using
tools, mainly for network import and demand modeling. SUMO are presented. We then outline the usage of SUMO
SUMO helps to investigate a large variety of research topics, within some recent research projects. Finally, we present
mainly in the context of traffic management and vehicular recent extensions and discuss current development topics.
communications. We describe the current state of the package,
its major applications, both by research topic and by example, II. THE SUMO SUITE
as well as future developments and extensions.
SUMO is not only a traffic simulation, but rather a suite
Keywords-microscopic traffic simulation; traffic of applications, which help to prepare and to perform the
management; open source; software simulation of a traffic scenario. As the simulation application
“sumo”, which is included in the suite, uses own formats for
road networks and traffic demand, both have to be imported
I. INTRODUCTION or generated from existing sources of different kind. Having
SUMO (“Simulation of Urban MObility”) [1][2] is a the simulation of large-scale areas as the major application
microscopic, inter- and multi-modal, space-continuous and for sumo in mind, much effort has been put into the design
time-discrete traffic flow simulation platform. The and implementation of heuristics which determine missing,
implementation of SUMO started in 2001, with a first open but needed attributes.
source release in 2002. There were two reasons for making In the following, the applications included in the suite are
the work available as open source under the gnu public presented, dividing them by their purpose: network
license (GPL). The first was the wish to support the traffic generation, demand generation, and simulation.
simulation community with a free tool into which own
algorithms can be implemented. Many other open source A. Road Network Generation
traffic simulations were available, but being implemented SUMO road networks represent real-world networks as
within a student thesis, they got unsupported afterwards. A graphs, where nodes are intersections, and roads are
major drawback – besides reinvention of the wheel – is the represented by edges. Intersections consist of a position, a
almost non-existing comparability of the implemented shape, and right-of-way rules, which may be overwritten by
models or algorithms, and a common simulation platform is a traffic light. Edges are unidirectional connections between
assumed to be of benefit here. The second reason for making two nodes and contain a fixed number of lanes. A lane
the simulation open source was the wish to gain support contains geometry, the information about vehicle classes
from other institutions. allowed on it, and the maximum allowed speed. Therefore,
Within the past ten years, SUMO has evolved into a full changes in the number of lanes along a road are represented
featured suite of traffic modeling utilities including a road using multiple edges. Such a view on road networks is
network importer capable of reading different source common; though some other approaches, such as Vissim’s
formats, demand generation and routing utilities, which use a [3] network format or the OpenDRIVE [4] format, exist.
high variety of input sources (origin destination matrices, Besides this basic view on a road network, SUMO road
traffic counts, etc.), a high performance simulation usable for networks include traffic light plans, and connections between
single junctions as well as whole cities including a “remote lanes across an intersections describing which lanes can be
control” interface (TraCI, see Section II. D.) to adapt the used to reach a subsequent lane.
simulation online and a large number of additional tools and SUMO road networks can be either generated using an
scripts. The major part of the development is undertaken by application named “netgenerate” or by importing a digital
the Institute of Transportation Systems at the German road map using “netconvert”. netgenerate builds three
Aerospace Center (Deutsches Zentrum für Luft- und different kinds of abstract road networks: “manhattan”-like
Raumfahrt, DLR). External parties supported different grid networks, circular “spider-net” networks, and random
extensions to the simulation suite. networks. Each of the generation algorithms has a set of
In this paper, we will survey some of the recent options, which allow adjusting the network’s properties.
developments and future prospects of SUMO. We start with Figure 1 shows examples of the generated networks.
an overview of the applications in the suite, showing how

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

129

describing a road network for SUMO. This XML

representation is broken into five file types, each for
description of nodes, edges, optionally edge types,
connections, and (fixed) traffic light plans. Edge types name
sets of default edge attributes, which can be referenced by
the later loaded edges. Nodes describe the intersections,
edges the road segments. Connections describe which lanes
Figure 1. Examples of abstract road networks as built using incoming into an intersection are connected to which
“netgenerate”; from left to right: grid (“manhattan”, spider, and random
network)
outgoing lanes. The simulation network created by
netconvert contains heuristically computed values wherever
The road network importer netconvert converts networks the inputs are incomplete as well as derived values such as
from other traffic simulators such as VISUM [5], Vissim, or the exact geometry at junctions. It is also possible to convert
MATSim [6]. It also reads other common digital road a simulation network back into the “plain” format. Multiple
network formats, such as shapefiles or OpenStreetMap [7]. input formats can be loaded at the same time and are
Besides these formats, netconvert is also capable to read less automatically merged. Since the “plain” format allows
known formats, such as OpenDRIVE or the RoboCup specifying the removal of network elements and the adaption
network format. Figures 2 and 3 show the capabilities to of single edge and lane parameters, it can be used for a wide
import road networks from OpenStreetMap by example, range of network modifications. To support such
comparing the original rendering on OpenStreetMap’s web modifications SUMO additionally provides the python tool
pages against SUMO rendering of the imported network. netdiff.py, which computes the (human-readable) difference
D between two networks A and B. Loading A and D with
netconvert reproduces B.
Most of the available digital road networks are originally
meant to be used for routing (navigation) purposes. As such,
they often lack the grade of detail needed by microscopic
road traffic simulations: the number of lanes, especially in
front of intersections, information about which lanes
approach which consecutive ones, traffic light positions and
plans, etc., are missing. Sharing the same library for
preparing generated/imported road networks, see Figure 4,
both, netgenerate and netconvert, try to determine missing
values using heuristics. A coarse overview on this
preparation process can be found in [8]. However, most of
the algorithms described in [8] have been reworked since its
publication. Additional, optional heuristics guess locations of
highway on- and off-ramps, roundabouts, traffic lights, etc.
Figure 2. Original OpenStreetMap network of Gothenborg.

Figure 3. Gothenborg network imported into SUMO. Figure 4. Common network preparation procedure in netconvert and
netgenerate.
Additionally, netconvert reads a native, SUMO-specific,
XML-representation of a road network graph referred to as Even with the given functionality, it should be stated that
“plain” XML, which allows the highest degree of control for preparing a real-world network for a microscopic simulation

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

130

is still a time-consuming task, as the real-world topology of calculation under different cost functions. Details on the
more complicated intersections often has to be improved models used in SUMO can be found in Section III.B.
manually. A new tool named “netedit” allows editing road SUMO includes two further route computation
networks graphically. This is in many cases simpler and applications. The first, “jtrrouter”, uses definitions of turn
faster than preparing XML input files. It also combines the percentages at intersection for computing routes through the
otherwise separate steps of network generation and network. Such an approach can be used to set up the demand
inspection with netconvert and the simulation GUI. netedit is within a part of a city’s road network consisting of few
not yet available for public use. nodes. The second, “dfrouter”, computes routes by using
information from inductive loop or other cross-section
B. Vehicles and Routes detectors. This approach is quite successful when applied to
SUMO is a purely microscopic traffic simulation. Each highway scenarios where the road network does not contain
vehicle is given explicitly, defined at least by a unique rings and the highway entries and exits are completely
identifier, the departure time, and the vehicle’s route through covered by detectors. It fails on inner-city networks with
the network. By “route” we mean the complete list of rings or if the coverage with detectors is low.
connected edges between a vehicle's start and destination. If It should be noted, that, while digital representations of
needed, each vehicle can be described in a finer detail using real-world road networks became available in good quality
departure and arrival properties, such as the lane to use, the in recent years, almost no sources for traffic demand are
velocity, or the exact position on an edge. Each vehicle can freely available. Within most of our (DLR's) projects, a road
get a type assigned, which describes the vehicle’s physical administration authority was responsible for supporting the
properties and the variables of the used movement model. demand information, either in form of O/D-matrices or at
Each vehicle can also be assigned to one of the available least by supplying traffic counts, which were used to
pollutant or noise emission classes. Additional variables calibrate a model built on rough assumptions.
allow the definition of the vehicle’s appearance within the Two tools enclosed in the SUMO package try to solve
simulation’s graphical user interface. this problem by modeling the mobility wishes of a described
A simulation scenario of a large city easily covers one population. “SUMO Traffic Modeler” by Leontios G.
million vehicles and their routes. Even for small areas, it is Papaleontiou [9] offers a graphical user interface allowing
hardly possible to define the traffic demand manually. The the user to set up demand sources and sinks graphically.
SUMO suite includes some applications, which utilize “activitygen” written by Piotr Woznica and Walter
different sources of information for setting up a demand. Bamberger from TU Munich has almost the same
For large-scale scenarios usually so-called “origin/ capabilities, but has no user interface. Both tools are
destination matrices” (O/D matrices) are used. They describe included in the suite and both use own models for creating
the movement between so-called traffic analysis zones mobility wishes for an investigated area, requiring different
(TAZ) in vehicle numbers per time. For use in SUMO these data. They are both under evaluation, currently.
matrices must be disaggregated into individual vehicle trips Figure 5 summarizes the possibilities to set up a demand
with depart times spread across the described time span. for a traffic simulation using tools included in the SUMO
Unfortunately, often, a single matrix is given for a single package.
day, which is too imprecise for a microscopic traffic
simulation since flows between two TAZ strongly vary over inductive
turning
the duration of a day. For example, people are moving into loop
measures
flows
ratios
O/D-matrix

the inner-city centers to get to work in the morning, and

leave the inner-city area in the afternoon or evening. Such
direction changes cannot be retrieved from an aggregated
24h matrix. Much more useful but only sometimes available DFROUTER JTRROUTER OD2TRIPS

are matrices with a scale of 1h. The SUMO suite includes

“od2trips”, an application for converting O/D matrices to
single vehicle trips. An hourly load curve can be given as SUMO Traffic
additional input for splitting the daily flows into more Modeler
trips

realistic hourly slices. Besides disaggregating the matrix, the

application also optionally assigns an edge of the road
network as depart/arrival position, respectively. The mapping
from traffic assignment zones to edges must be supplied as ACTIVITYGEN DUAROUTER

another input.
The resulting trips obtained from od2trips consist of a
start and an end road together with a departure time.
However, the simulation requires the complete list of edges routes
to pass. Such routes are usually calculated by performing a
dynamic user assignment (DUA). This is an iterative process Figure 5. Supported methods for demand generation.
employing a routing procedure such as shortest path

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

131

C. Simulation
The application “sumo” performs a time-discrete
simulation. The default step length is 1s, but may be chosen
to be lower, down to 1ms. Internally, time is represented in
microseconds, stored as integer values. The maximum
duration of a scenario is so bound to 49 days. The simulation
model is space-continuous and internally, each vehicle’s
position is described by the lane the vehicle is on and the
distance from the beginning of this lane. When moving
through the network, each vehicle’s speed is computed using
a so-called car-following model. Car-following models
usually compute an investigated vehicle’s (ego) speed by
looking at this vehicle’s speed, its distance to the leading
vehicle (leader), and the leader’s speed. SUMO uses an
extension of the stochastic car-following model developed Figure 6. Screenshot of the graphical user interface coloring vehicles by
by Stefan Krauß [10] per default. Krauß’ model was chosen their CO2 emission.
due to its simplicity and its high execution speed.
The model by Krauß has proved to be valid within a set SUMO allows generating various outputs for each
of performed car-following model comparisons [11][12][13]. simulation run. These range from simulated inductive loops
Nonetheless, it has some shortcomings, among them its to single vehicle positions written in each time steps for all
conservative gap size, yielding in a too low gap acceptance vehicles and up to complex values such as information about
during lane changing, and the fact that the model does not each vehicle’s trip or aggregated measures for all streets
scale well when the time step length is changed. To deal with and/or lanes. Besides conventional traffic measures, SUMO
these issues, an application programmer interface (API) for was extended by a noise emission and a pollutant emission /
implementing other car-following models was added to fuel consumption model, see also Section V.A. All output
sumo. Currently, among others, the following models are files generated by SUMO are in XML-format.
included: the intelligent driver model (IDM) [14], Kerner’s
three-phase model [15], and the Wiedemann model [16]. It D. On-Line Interaction
must be stated, though, that different problems were In 2006, the simulation was extended by the possibility to
encountered when using these models in complex road interact with an external application via a socket connection.
networks, probably due to undefined side-constraints and/or This API, called “TraCI” for “Traffic Control Interface” was
assumptions posed by the simulation framework. For this implemented by Axel Wegener and his colleagues at the
reason, the usage of different car-following models should be University of Lübeck [18], and was made available as a part
stated to be experimental only, at the current time. Being a of SUMO’s official release. Within the iTETRIS project, see
traffic flow simulation, there are only limited possibilities to Section IV.B, this API was reworked, integrating it closer
reflect individual driver behavior; it is however possible to into SUMO’s architecture.
give each vehicle its own set of parameters (ranging from To enable on-line interaction, SUMO has to be started
vehicle length to model parameters like preferred headway with an additional option, which obtains the port number to
time) and even to let different models run together. The listen to. After the simulation has been loaded, SUMO starts
computation of lane changing is done using a model to listen on this port for an incoming connection. After being
developed during the implementation of SUMO [17]. connected, the client is responsible for triggering simulation
Two versions of the traffic simulation exist. The steps in SUMO as well as for closing down the connection
application “sumo” is a pure command line application for what also forces the simulation to quit. The client can access
efficient batch simulation. The application “sumo-gui” offers values from almost all simulation artifacts, such as
a graphical user interface (GUI) rendering the simulation intersections, edges, lanes, traffic lights, inductive loops, and
network and vehicles using openGL. The visualization can of course vehicles. The client may also change values, for
be customized in many ways, i.e., to visualize speeds, example instantiate a new traffic light program, change a
waiting times and to track individual vehicles. Additional vehicle’s velocity or force it to change a lane. This allows
graphical elements – points-of-interest (POIs), polygons, and complex interaction such as online synchronization of traffic
image decals – allow to improve a scenario’s visual lights or modeling special behavior of individual vehicles.
appearance. The GUI also offers some possibilities to While DLR uses mainly a client-library written in Python
interact with the scenario, e.g. by switching between when interacting with the simulation, the client can be
prepared traffic signal programs, changing reroute following written in any programming language as long as TCP sockets
grades, etc. Figure 6 shows a single intersection simulated in are supported. A Python API as well as a freely available
sumo-gui. sumo-gui offers all features the command line Java API [19] are included with SUMO and support for other
version sumo supports. programming languages may follow.

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

132

III. RESEARCH TOPICS microscopic traffic flow models are not capable of modelling
In the following, the major research topics addressed real collisions and thus derive safety-related measures
using SUMO are presented. The list is mainly based on indirectly, for instance by detecting full braking. SUMO’s
observations of published papers which cite SUMO. strength lies in simulation of V2X applications that aim at
improving traffic efficiency. Additionally, evaluating
A. Vehicular Communication concepts for forwarding messages to their defined
The probably most popular application for the SUMO destination (“message routing”) can be done using SUMO,
suite is modeling traffic within research on V2X – vehicle- see, for example, [26] or [27].
to-vehicle and vehicle-to-infrastructure – communication. In B. Route Choice and Dynamic Navigation
this context, SUMO is often used for generating so-called
“trace files”, which describe the movement of The assignment of proper routes to a complete demand or
communication nodes by converting the output of a SUMO a subset of vehicles is investigated both, on a theoretical base
simulation into a format the used communication simulator as well as within the development of new real-world
can read. Such a post-processing procedure allows feeding a applications. On the theoretical level, the interest lies in a
communication simulator with realistic vehicle behavior, but proper modeling of how traffic participants choose a route –
fails on simulating the effects of in-vehicle applications that a path through the given road network – to their desired
change the vehicles’ behavior. To investigate these effects, a destination. As the duration to pass an edge of the road graph
combined simulation of both, traffic and communication is highly depends on the numbers of participants using this
necessary [20]. For such research, SUMO is usually coupled edge, the computation of routes through the network under
to an external communication simulation, such as ns2 or ns3 load is a crucial step in preparing large-scale traffic
[21] using TraCI. For obtaining a functioning environment simulations. Due to its fast execution speed, SUMO allows
for the simulation of vehicular communications, a further to investigate algorithms for this “user assignment” or
module that contains the model of the V2X application to “traffic assignment” process on a microscopic scale. Usually,
simulate is needed. Additionally, synchronization and such algorithms are investigated using macroscopic traffic
message exchange mechanisms have to be involved. flow models, or even using coarser road capacity models,
TraNS [22] was a very popular middleware for V2X which ignore effects such as dissolving road congestions.
simulation realizing these needs. It was build upon SUMO The SUMO suite supports such investigations using the
and ns2. Here, TraNS’ extensions to ns2 were responsible for duarouter application. Two algorithms for computing a user
synchronizing the simulators and the application had also to assignment are implemented, c-logit [28] and Gawron’s [29]
be modeled within ns2. TraNS was the major reason for dynamic user assignment algorithm. Both are iterative and
making TraCI open source. After the end of the projects the therefore time consuming. Possibilities to reduce the duration
original TraNS authors were working on, TraNS was no to compute an assignment were evaluated and are reported in
longer maintained. Since the TraCI API was changed after [30]. A further possibility to reduce the computational effort
the last TraNS release, TraNS only works with an outdated is given in [31]. Here, vehicles are routed only once, directly
version of SUMO. by the simulation and the route choice is done based on a
A modern replacement for TraNS was implemented continuous adaptation of the edge weights during the
within the iTETRIS project [23]. The iTETRIS system simulation.
couples SUMO and ns2’s successor ns3. ns3 was chosen Practical applications for route choice mechanisms arise
because ns2 was found to be unstable when working with a with the increasing intelligence of navigation systems.
large number of vehicles. Within the iTETRIS system, the Modern navigation systems as Tom Tom’s IQ routes ([32])
“iTETRIS Control System”, an application written in c++ is use on-line traffic information to support the user with a
responsible for starting and synchronizing the involved fastest route through the network regarding the current
simulators. The V2X applications are modeled as separate, situation on the streets. One research topic here is to develop
language-agnostic programs. This clear distribution of new traffic surveillance methods, where vehicular
responsibilities allows to implement own applications communication is one possibility. With the increased
conveniently in the user’s favorite programming language. penetration rate of vehicles equipped with a navigation
The Veins framework [20] couples SUMO and device, further questions arise: what happens if all vehicles
OMNET++ [24], a further communication simulator. A get the same information? Will they all use the same route
further, very flexible approach for coupling SUMO with and generate new congestions? These questions are not only
other applications is the VSimRTI middleware developed by relevant for drivers, but also for local authorities as
Fraunhofer Fokus [25]. Its HLA-inspired architecture not navigation devices may invalidate concepts for keeping
only allows the interaction between SUMO and other certain areas calm by routing vehicles through these areas.
communication simulators. It is also able to connect SUMO SUMO allows addressing these topics, see, e.g., [33].
and Vissim, a commercial traffic simulation package. In C. Traffic Light Algorithms
[25], a system is described where SUMO was used to model The evaluation of developed traffic light programs or
large-scale areas coarsely, while Vissim was used for a fine- algorithms for making traffic lights adaptable to current
grained simulation of traffic intersections. traffic situation is one of the main applications for
Many vehicular communication applications target at microscopic traffic flow simulations. As SUMO’s network
increasing traffic safety. It should be stated, that up to now,

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

133

model is relatively coarse compared to commercial investigation is the evaluation of hyperspectral sensors
applications such as Vissim, SUMO is usually not used by reported in [38].
traffic engineers for evaluating real-life intersections. Still, Besides evaluating developed surveillance systems,
SUMO’s fast execution time and its open TraCI API for possibilities to incorporate traffic measurements of various
interaction with external applications make it a good kinds into a simulation are evaluated, see for example
candidate for evaluating new traffic control algorithms in Section IV.C. on “VABENE”.
abstract scenarios.
The first investigation in traffic lights was performed IV. RECENT AND CURRENT PROJECTS
within the project “OIS” [34] where a traffic light control SUMO was used in past research projects performed by
algorithm, which used queue lengths determined by image the DLR and other parties. In the following, some of the
processing should have been evaluated. As a real-world recently performed projects are described.
deployment of the OIS system was not possible due to legal
constraints, the evaluation had to be done using a simulation. A. TrafficOnline
The simulation was prepared by implementing a real-world Within the TrafficOnline project, a system for
scenario, including real-world traffic light programs. The determining travel times using GSM telephony data was
simulation application itself was extended by a simulated designed, implemented, and evaluated. SUMO was used to
sensor, which allowed retrieving queue lengths in front of the validate this system’s functionality and robustness. In the
intersection similar to the real image processing system. The following, we focus on the simulation’s part only, neither
traffic light control was also implemented directly into the describing the TrafficOnline system itself, nor the evaluation
simulation. At the end, the obtained simulation of OIS-based results.
traffic control was compared against the real-world traffic The outline for using the simulation was as follows.
lights, [34] shows the results. Real-world scenarios were set up in the simulation. When
In ORINOKO, a German project on traffic management, being executed, the simulation was responsible for writing
the focus was put on improving the weekly switch plans per-edge travel time information as well as simulated
within the fair trade center area of the city of Nürnberg. telephony behavior values. The TrafficOnline system itself
Here, the initial and the new algorithm for performing the obtained the latter, only, and computed travel times in the
switch procedure between two programs were implemented underlying road network. These were then compared to the
and evaluated. Additionally, the best switching times were travel times computed by the simulation. The overall
computed by a brute-force iteration over the complete procedure is shown in Figure 7.
simulated day and the available switch plans.
By distinguishing different vehicle types, SUMO also
allowed to simulate a V2X-based emergency vehicle
prioritization at intersections [35]. Other approaches for
traffic light control were also investigated and reported by
other parties, see, e.g., [36], or [37].
As mentioned before, the first investigations were
performed by implementing the traffic light algorithms to
evaluate directly into the simulation’s core. Over the years, Figure 7. Overall process of TrafficOnline validation.
this approach was found to be hard to maintain. Using TraCI
seems to be a more sustainable procedure currently. The evaluation was performed using scenarios located in
and around Berlin, Germany, which covered urban and
D. Evaluation of Traffic Surveillance Systems highway situations. The road networks were imported from a
Simulation-based evaluation of surveillance systems NavTeq database. Manual corrections were necessary due to
mainly targets on predicting whether and to what degree the the limits of digital road networks described earlier in
developed surveillance technology is capable to fulfill the Section II.A.
posed needs at an assumed rate of recognized and/or
equipped vehicles. Such investigations usually compare the
output of the surveillance system, fed with values from the
simulation to the according output of the simulation. An
example will be given later, in Section IV.A. on the project
“TrafficOnline”.
A direct evaluation of traffic surveillance systems’
hardware, for example image processing of screenshots of
the simulated area, is uncommon, as the simulation models
of vehicles and the environment are too coarse for being a Figure 8. Validation of the traffic flows in TrafficOnline.
meaningful input to such systems. Nonetheless, the
simulation can be used to compute vehicle trajectories, Measurements from inductive loops were used for traffic
which can be enhanced to match the inputs needed by the modeling. Figure 8 shows two examples of validating the
evaluated system afterwards. An example of such an traffic simulation by comparing simulated (black) and real-

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

134

world inductive loop measures (blue, where dark blue inputs for the simulations Vissim and VISUM, both
indicates the average value). For validating the robustness of commercial products of PTV AG. These scenarios were
the TrafficOnline system, scenario variations have been converted into the SUMO-format using the tools from the
implemented, by adding fast rail train lines running parallel SUMO package. Besides the road networks and the demand
to a highway, or by implementing additional bus lanes, for for the peak hour between 8:00am and 9:00am, they included
example. Additionally, scenario variations have been built by partial definitions of the traffic lights, public transport, and
scaling the simulated demand by +/- 20%. other infrastructure information.
For validating the TrafficOnline system, a model of One of the project’s outputs is a set of in-depth
telephony behavior was implemented, first. The telephony descriptions of V2X-based traffic management applications,
model covered the probability to start a call and a started including different attempts for traffic surveillance,
call’s duration, both retrieved from real-world data. For an navigation, and traffic light control. In the following, one of
adequate simulation of GSM functionality, the real-world these applications, the bus lane management, is described,
GSM cell topology was put onto the modeled road networks. showing the complete application design process, starting at
It should be noted, that dynamic properties of the GSM problem recognition, moving over the design of a
network, such as cell size variations, or delays on passing a management application that tries to solve it, and ending at
cell border, have not been considered. Figure 9 shows the its evaluation using the simulation system. A more detailed
results of validating the simulated telephone call number report on this application is [39].
(black) against the numbers found in real-world data (green, Public transport plays an important role within the city of
dark green showing the average call number) over a day for Bologna, and the authorities are trying to keep it attractive by
two selected GSM cells. giving lanes, and even streets free to public busses only. On
the other hand, the city is confronted with event traffic – e.g.,
visitors of football matches, or the fair trade centre – coming
in the form of additional private passenger cars. One idea
developed in iTETRIS was to open bus lanes for private
traffic in the case of additional demand due to such an event.
The application was meant to include two sub-systems. The
first one was responsible for determining the state on the
roads. The second one used this information to decide
whether bus lanes shall be opened for passenger cars and
Figure 9. Validation of the telephony behavior in TrafficOnline. should inform equipped vehicles about giving bus lanes for
usage.
B. iTETRIS
The interest in V2X communication is increasing but the
deployment of this technology is still expensive, and ad-hoc
implementations of new traffic control systems in the real
world may even be dangerous. For research studies where
the benefits of a system are measured before it is deployed, a
simulation framework, which simulates the interaction
between vehicles and infrastructure is needed, as described
in III.A. The aim of the iTETRIS project was to develop
such a framework, coupling the communication simulator
ns3 and SUMO using an open source system called “iCS” –
iTETRIS Control System – which had to be developed
within the project. In contrary to other, outdated solutions
such as TraNS, iTETRIS was meant to deliver a sustainable Figure 10. Speed information collection by RSUs. Each dot represents one
product, supported and continued to be developed after the data point, the color represents the speed (green means fast, red slow).
project’s end.
Besides implementing the V2X simulation system itself, In order to use standardised techniques, traffic
which was already presented in Section III.A., the work surveillance was implemented by collecting and averaging
within iTETRIS included a large variety of preparation tasks the speed information contained in the CAMs (cooperative
and – after completing the iCS implementation – the awareness messages) at road side units (RSUs) placed at
evaluation of traffic management applications as well as of major intersections (see Figure 10). As soon as the average
message routing protocols. speed falls below a threshold, the application, assuming a
The preparations mainly included the investigation of high traffic amount, gives bus lanes free for passenger cars.
real world traffic problems and their modelling in a The RSU sends then the information about free bus lanes to
simulation environment. The city of Bologna, who was a vehicles in range.
project partner in iTETRIS, supported traffic simulation The evaluations show that the average speed was usable
scenarios covering different parts of the town, mainly as as an indicator for an increase of traffic demand. Though, as

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

135

the usage of this measure is rather uncommon, further adapted manually where needed. The basic traffic demand
investigations and validations should be performed. When was computed from O/D-matrices supplied by traffic
coming to measure the benefits of using bus-lanes for private authorities.
vehicles, the application did not prove its benefits at all. At The simulation is restarted every 10 minutes, loads a
higher penetration rates, the average travel time of all previously saved state of the road network and computes the
distinguished transport modes – busses, vehicles not state for half an hour ahead. While running, the simulation
equipped with V2X devices, equipped vehicles, as well as state is calibrated using traffic measurements from various
rerouted vehicles – climbs above the respective average sources such as inductive loops, floating car data and (if
travel times without the application. The main reason is that available) an airborne traffic surveillance system. This
vehicles, which use bus lanes tend to either decelerate the calibration is performed by comparing simulated vehicle
busses or are blocked by busses. counts with measured vehicle counts at all network edges for
which measurements are currently available. Depending on
this comparison, vehicles are removed prematurely from the
simulation or new vehicles are inserted. Also, the maximum
speed for each edge is set to the average measured speed.
A crucial part of this calibration procedure is the
selection of a route for inserted vehicles. This is
accomplished by building a probability distribution of
possible routes for each network edge out of the basic traffic
demand and then sampling from this distribution.
The accuracy of the traffic prediction depends crucially
on the accuracy of the basic traffic demand. To lessen this
problem we are currently investigating the use of historical
traffic measurements to calibrate the simulation wherever
current measurements are not yet available. However, this
approach carries the danger of masking unusual traffic
developments, which might already be foreseeable from the
latest measurements.
Figure 11. Average travel times changes per vehicle class over equipment Both, the current traffic state as well as the prediction of
rates.
the future state is presented to the authorities in a browser-
The results show, that a naive implementation of the based management interface. The management interface
application does not take into account traffic behaviour and allows to investigate the sources of collected information,
degrades with increasing penetration rate. This effect was including inductive loops, airborne and conventional images,
observed in studies on other V2X-based traffic management as well as to monitor routes or evaluate the network’s current
applications as well. It also shows that proper design and a accessibility, see Figure 12.
fine-grained evaluation of developed applications are
needed.
C. VABENE
Big events or catastrophes may cause traffic jams and
problems to the transport systems, causing additional danger
for the people who live in the area. Public authorities are
responsible for taking preparatory actions to prevent the
worst case. The objective of VABENE is to implement a
system that supports public authorities to decide which
action should be taken. This system is the successor of
demonstrators used during the pope’s visit in Germany in
2005 and during the FIFA World Cup in 2006.
One focus of VABENE lies on simulating the traffic of
large cities. The system shows the current traffic state of the
whole traffic network, helping the traffic manager to realize Figure 12. Screenshot of the “EmerT” portal used in VABENE showing
travel time isochrones.
when a critical traffic state will be reached. To simulate the
traffic of a large region such as Munich and the area around
Munich at multiple real-time speed, a mesoscopic traffic D. CityMobil
model was implemented into SUMO. This model has not yet Microscopic traffic simulations also allow the evaluation
been released to the public and is available for internal of large scale effects of changes in vehicle or driver behavior
proposes only. such as the introduction of automated vehicles or
Similar to the TrafficOnline Project (Section A), the road electromobility. The former was examined with the help of
networks were imported from a NavTeq database and

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

136

SUMO in the EU project CityMobil where different passenger. It is important to note that earlier delays influence
scenarios of (partly) automated cars or personal rapid transit later trips of a simulated person. The above concept is
were set up on different scales, from a parking area up to reflected in an extension of the SUMO route input. One can
whole cities. now specify a person as a list of rides, stops and walks. A
On a small scale, the benefits of an autonomous bus ride can stand for any vehicular transportation, both private
system were evaluated. In this scenario, busses are informed and public. It is specified by giving a starting edge, ending
about waiting passengers and adapt their routes to this edge and a specification of the allowed vehicles. Stops
demand. On a large scale, the influence of platooning correspond to non-traffic related activities such as working
vehicles was investigated, using the model of a middle-sized or shopping. A walk models a trip taken by foot but it can
city of 100.000 inhabitants. Both simulations showed also stand for other modes of transport that do not interfere
positive effects of transport automation. with road traffic. Another extension concerns the vehicles. In
addition to their route, a list of stops and a line attribute can
V. RECENT EXTENSIONS be assigned. Each stop has a position, and a trigger which
may be either a fixed time, a waiting time or the id of a
A. Emission and Noise Modeling person for which the vehicle must wait. The line attribute can
Within the iTETRIS project, SUMO was extended by a be used to group multiple vehicles as a public transport
model for noise emission and a model for pollutant emission route.
and fuel consumption. This was required within the project These few extensions are sufficient to express the above
for evaluating the ecological influences of the developed mentioned person trips. They are being used within the
V2X applications. TAPAS [43][44] project to simulate intermodal traffic for
Both models are based on existing descriptions. 7 models the city of Berlin. Preliminary benchmarks have shown that
for noise emission and 15 pollutant emission / fuel the simulation performance is hardly affected by the
consumption models were evaluated, first. The parameter overhead of managing persons. In the future the following
they need and their output were put against values available issues will be addressed:
within the simulation and against the wanted output, • Online rerouting of persons. At the moment routing
respectively. Finally, HARMONOISE [40] was chosen as
across trips must be undertaken before the start of the
noise emission model. Pollutant emission and fuel
consumption is implemented using a continuous model simulation. It is therefore not possible to compensate a
derived from values stored in the HBEFA database [41]. missed bus by walking instead of waiting for the next
The pollutant emission model’s implementation within bus.
SUMO allows to collect the emissions and fuel consumption • Smart integration of bicycles. Depending on road
of a vehicle over the vehicle’s complete ride and to write infrastructure bicycle traffic may or may not interact
these values into a file. It is also possible to write collected with road traffic.
emissions for lanes or edges for defined, variable • Import modules for importing public time tables.
aggregation time intervals. The only available noise output
collects the noise emitted on lanes or edges within pre- VI. CURRENT DEVELOPMENT
defined time intervals, a per-vehicle noise collecting output As shown, the suite covers a large variety of
is not available. Additionally, it is possible to retrieve the functionalities, and most of them are still under research. In
noise, emitted pollutants, and fuel consumption of a vehicle the past, applications from the SUMO suite were adapted to
in each time step via TraCI, as well as to retrieve collected currently investigated projects’ needs, while trying to keep
emissions, consumption, and noise level for a lane or a road. the already given functionality work. This development
Besides measuring the level of emissions or noise for context will be kept for the next future, and major changes
certain scenarios, the emission computation was also used
in functionality are assumed to be grounded on the
for investigating new concepts of vehicle routing and
dependencies between the traffic light signal plans and investigated research questions. Nonetheless, a set of
emissions [42]. “strategic” work topics exist and will be presented in the
following sub-sections. They mainly target on increasing the
B. Person-based Intermodal Traffic Simulation simulation’s validity as well as the number of situations the
A rising relevance of intermodal traffic can be expected simulation is able to replicate, and on establishing the
due to ongoing urbanization and increasing environmental simulation as a major tool for evaluation of academic
concerns. To accommodate this trend SUMO was extended models and algorithms for both traffic simulation as well as
by capabilities for simulating intermodal traffic. We give a for evaluation of traffic management applications.
brief account of the newly added concepts.
The conceptual center of intermodal traffic is the A. Car-Following and Lane-Change API
individual person. This person needs to undertake a series of One of the initial tasks SUMO was developed for was the
trips where each may be taken with a different mode of comparison of traffic flow models, mainly microscopic car-
transport such as personal car, public bus or walking. Trips following and lane-changing models. This wish requires a
may include traffic related delays, such as waiting in a jam, clean implementation of the models to evaluate. Within the
waiting for a bus or waiting to pick up an additional iTETRIS project, first steps towards using other models than

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

137

the used Krauß extension for computing the vehicles’ [4] OpenDRIVE consortium, OpenDRIVE homepage [Online],
longitudinal movement were taken by implementing an API http://www.opendrive.org/, accessed July 03, 2012.
for embedding other car-following models. Some initial [5] PTV AG, VISUM homepage [Online], http://www.ptv-
vision.com/de/produkte/vision-traffic-suite/ptv-visum/,
implementations of other models exist, though not all of accessed July 03, 2012.
them are able to deal correctly with multi-lane urban traffic.
[6] MATSim homepage [Online], http://www.matsim.org/,
What is already possible to do with car-following models is accessed July 03, 2012.
also meant to be implemented for lane-change models. [7] OpenStreetMap homepage [Online],
http://www.openstreetmap.org/, accessed July 03, 2012.
B. Model Improvements
[8] D. Krajzewicz, G. Hertkorn, J. Ringel, and P. Wagner,
While evaluation of academic driver behavior models is “Preparation of Digital Maps for Traffic Simulation; Part 1:
one of the aimed research topics, most models are Approach and Algorithms,” in Proceedings of the 3rd
concentrating to describe a certain behavior, e.g., Industrial Simulation Conference 2005, pp. 285–290.
spontaneous jams, making them inappropriate to be used EUROSIS-ETI. 3rd Industrial Simulation Conference 2005,
Berlin (Germany). ISBN 90-77381-18-X.
within complex scenarios which contain a large variety of
[9] L. G. Papaleondiou and M. D. Dikaiakos, “TrafficModeler: A
situations. In conclusion, next steps of SUMO development Graphical Tool for Programming Microscopic Traffic
will go beyond established car-following models. Instead, an Simulators through High-Level Abstractions,” in Proceedings
own model will be developed, aiming on its variability of the 69th IEEE Vehicular Technology Conference, VTC
mainly. Spring 2009, Spain, 2009.
[10] S. Krauß, “Microscopic Modeling of Traffic Flow:
C. Interoperability Investigation of Collision Free Vehicle Dynamics,” PhD
SUMO is not the only available open source traffic thesis, 1998.
simulation platform. Some other simulators, such as [11] E. Brockfeld, R. Kühne, and P. Wagner, “Calibration and
Validation of Microscopic Traffic Flow Models,” in
MATsim, offer their own set of tools for demand generation, Transportation Research Board [ed.]: TRB 2004 Annual
traffic assignment etc. It is planned to make these tools being Meeting, pp. 62–70, TRB Annual Meeting, Washington, DC
usable in combination with SUMO by increasing SUMO’s (USA) 2004.
capabilities to exchange data. Besides connecting with other [12] E. Brockfeld and P. Wagner, “Testing and Benchmarking of
traffic simulation packages, SUMO is extended for being Microscopic Traffic Flow Models,” in Proceedings of the
capable to interact with driving or world simulators. Within 10th World Conference on Transport Research, pp. 775-776,
WCTR04 - 10th World Conference on Transport Research,
the DLR project “SimWorld Urban”, SUMO is connected to Istanbul (Turkey) 2004.
the DLR driver simulator, allowing to perform simulator test [13] E. Brockfeld, R. Kühne, and P. Wagner, “Calibration and
drives through a full-sized and populated city area. Validation of Microscopic Traffic Flow Models,”
Transportation Research Records, 1934, pp. 179–187, 2005.
VII. SUMMARY [14] M. Treiber and D. Helbing, “Realistische Mikrosimulation
We have presented a coarse overview of the microscopic von Strassenverkehr mit einem einfachen Modell,”
Symposium Simulationstechnik (ASIM), 2002.
traffic simulation package SUMO, presenting the included
applications along with some common use cases, and the [15] B. Kerner, S. Klenov, and A. Brakemeier, “Testbed for
wireless vehicle communication: A simulation approach
next development steps. The number of projects and the based on three-phase traffic theory,” in Proceedings of the
different scales (from single junction traffic light control to IEEE Intelligent Vehicles Symposium (IV’08), pp. 180–185,
whole city simulation) present the capabilities of the 2008.
simulation suite. Together with its import tools for networks [16] R. Wiedemann, „Simulation des Straßenverkehrsflußes,“ in
and demand and recently added features such as emission Heft 8 der Schriftenreihe des IfV, Institut für Verkehrswesen,
modeling and the powerful TraCI interface, SUMO aims to Universität Karlsruhe, 1974.
stay one of the most popular simulation platforms not only in [17] D. Krajzewicz, “Traffic Simulation with SUMO - Simulation
of Urban Mobility,” in J. Barceló, “Fundamentals of Traffic
the field of vehicular communication. We kindly invite the Simulation,” International Series in Operations Research and
reader to participate in the ongoing development and Management Science. Springer, pp. 269–294, ISBN 978-1-
implement his or her own algorithms and models. Further 4419-6141-9. ISSN 0884-8289, 2010.
information can be obtained via the project’s web site [2]. [18] A. Wegener, M. Piórkowski, M. Raya, H. Hellbrück, S.
Fischer, and J.-P. Hubaux, “TraCI: An Interface for Coupling
Road Traffic and Network Simulators,” in Proceedings of the
REFERENCES 11th communications and networking simulation symposium,
2008.
[19] Politecnico di Torino, TraCI4J Homepage [Online],
[1] M. Behrisch, L. Bieker, J. Erdmann, and D. Krajzewicz, http://sourceforge.net/apps/mediawiki/traci4j/index.php?title=
“SUMO - Simulation of Urban MObility: An Overview,” in Main_Page, accessed July 09, 2012.
SIMUL 2011, The Third International Conference on [20] C. Sommer, Z. Yao, R. German, and F. Dressler, “On the
Advances in System Simulation, 2011. need for bidirectional coupling of road traffic microsimulation
[2] DLR and contributors, SUMO Homepage [Online], and network simulation,” in Proceedings of the 1st ACM
http://sumo.sourceforge.net/, accessed July 03, 2012. SIGMOBILE workshop on Mobility models, pp. 41–48,
[3] PTV AG, Vissim homepage [Online], http://www.ptv- 2008.
vision.com/en-uk/products/vision-traffic-suite/ptv- [21] ns3 Homepage [Online], http://www.nsnam.org/, accessed
vissim/overview/, accessed July 03, 2012. January 26, 2011.

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org

International Journal on Advances in Systems and Measurements, vol 5 no 3 & 4, year 2012, http://www.iariajournals.org/systems_and_measurements/

138

[22] M. Piórkowski, M. Raya, A. Lugo, P. Papadimitratos, M. [33] D. Krajzewicz, D. Teta Boyom, and P. Wagner, “Evaluation
Grossglauser, and J.-P. Hubaux, “TraNS: Realistic Joint of the Performance of city-wide, autonomous Route Choice
Traffic and Network Simulator for VANETs,” ACM based on Vehicle-to-vehicle-Communictaion,” TRB 2008 (87.
SIGMOBILE Mobile Computing and Communications Annual Meeting), 2008, Washington DC, USA.
Review, pp. 31-33, 2008. [34] D. Krajzewicz, E. Brockfeld, J. Mikat, J. Ringel, C. Rössel,
[23] iTETRIS Homepage [Online], http://www.ict-itetris.eu/10-10- W. Tuchscheerer, P. Wagner, and R. Woesler, “Simulation of
10-community/, accessed July 09, 2012. modern Traffic Lights Control Systems using the open source
[24] OMNET++ Homepage [Online], http://www.omnetpp.org/, Traffic Simulation SUMO,” in Proceedings of the 3rd
accessed July 09, 2012. Industrial Simulation Conference 2005, pp. 299–302,
EUROSIS-ETI, 3rd Industrial Simulation Conference 2005,
[25] D. Rieck, B. Schuenemann, I. Radusch, and C. Meinel, Berlin (Germany). ISBN 90-77381-18-X.2005.
“Efficient Traffic Simulator Coupling in a Distributed V2X
Simulation Environment,” in SIMUTools '10: Proceedings of [35] L. Bieker, “Emergency Vehicle prioritization using Vehicle-
the 3rd International ICST Conference on Simulation Tools to-Infrastructure Communication,” Young Researchers
and Techniques, Torremolinos, Malaga, Spain, 2010. ICST Seminar 2011 (YRS2011), 2011, Copenhagen, Denmark.
(Institute for Computer Sciences, Social-Informatics and [36] D. Greenwood, B. Burdiliak, I. Trencansky, H. Armbruster,
Telecommunications Engineering), ICST, Brussels, Belgium, and C. Dannegger, “GreenWave distributed traffic
pp. 1-9, ISBN: 978-963-9799-87-5. intersection control,” in Proceedings of The 8th International
[26] D. Borsetti and J. Gozalvez, “Infrastructure-Assisted Geo- Conference on Autonomous Agents and Multiagent Systems -
Routing for Cooperative Vehicular Networks,” in Proceedings Volume 2, pp. 1413–1414, 2009.
of the 2nd IEEE (*) Vehicular Networking Conference (VNC [37] O. H. Minoarivelo, “Application of Markov Decision
2010), 2010, New Jersey (USA). Processes to the Control of a Traffic Intersection,”
[27] M. A. Leal, M. Röckl, B. Kloiber, F. de Ponte-Müller, and T. postgraduate diploma, University of Barcelona, 2009
Strang, “Information-Centric Opportunistic Data [38] J. Kerekes, M. Presnar, K. Fourspring, Z. Ninkov, D.
Dissemination in Vehicular Ad Hoc Networks,” in Pogorzala, A. Raisanen, A. Rice, J. Vasquez, J. Patel, R.
International IEEE Conference on Intelligent Transportation MacIntyre, and S. Brown, “Sensor Modeling and
Systems (ITSC), 2010, Madeira Island (Portugal). Demonstration of a Multi-object Spectrometer for
[28] E. Cascetta, A. Nuzzolo, F. Russo, and A. Vitetta, “A Performance-driven Sensing,” in Proceedings of Algorithms
modified logit route choice model overcoming path and Technologies for Multispectral, Hyperspectral, and
overlapping problems,” in Transportation and traffic theory: Ultraspectral Imagery XV, SPIE Vol. 7334, Defense and
Proceedings of the 13th International Symposium on Security Symposium, Orlando, Florida, 2009, DOI:
Transportation and Traffic Theory. Pergamon Press, Lyon, 10.1117/12.819265.
France, 1996. [39] L. Bieker and D. Krajzewicz, “Evaluation of opening Bus
[29] C. Gawron, “Simulation-based traffic assignment – Lanes for private Traffic triggered via V2X Communication,”
computing user equilibria in large street networks,” Ph.D. (FISTS 2011), 2011, Vienna, Austria.
Dissertation, University of Köln, Germany, 1998 [40] R. Nota, R. Barelds, and D. van Maercke, “Harmonoise WP 3
[30] M. Behrisch, D. Krajzewicz, and Y.-P. Wang, “Comparing Engineering method for road traffic and railway noise after
performance and quality of traffic assignment techniques for validation and fine-tuning,” Technical Report Deliverable 18,
microscopic road traffic simulations,” in Proceedings of HARMONOISE, 2005.
DTA2008. DTA2008 International Symposium on Dynamic [41] INFRAS. HBEFA web site [Online], http://www.hbefa.net/,
Traffic Assignment, Leuven (Belgien), 2008. accessed July 09, 2012.
[31] M. Behrisch, D. Krajzewicz, P. Wagner, and Y.-P. Wang, [42] D. Krajzewicz, L. Bieker, E. Brockfeld, R. Nippold, and J.
“Comparison of Methods for Increasing the Performance of a Ringel, “Ökologische Einflüsse ausgewählter
DUA Computation,” in Proceedings of DTA2008. DTA2008 Verkehrsmanagementansätze,“ In Heureka '11, 2011,
International Symposium on Dynamic Traffic Assignment, Stuttgart, Germany.
Leuven (Belgien), 2008. [43] R. Cyganski and A. Justen, “Maßnahmensensitive Nach-
[32] R.-P. Schäfer, “IQ routes and HD traffic: technology insights fragemodellierung in mikroskopischen Personenverkehrs-
about tomtom's time-dynamic navigation concept,” in modellen,“ Deutsche Verkehrswissenschaftliche Gesellschaft,
Proceedings of the the 7th joint meeting of the European Schriftenreihe B, 2007.
software engineering conference and the ACM SIGSOFT [44] G. Hertkorn and P. Wagner, “Travel demand modelling based
symposium on The foundations of software engineering on time use data,” in 10th International conference on Travel
(ESEC/FSE '09). ACM, New York, NY, USA, 171-172. Behaviour Research, August 2004.
DOI=10.1145/1595696.1595698, 2009.

2012, © Copyright by authors, Published under agreement with IARIA - www.iaria.org