See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224793504

SUMO (Simulation of Urban MObility); An open-source traffic simulation

Conference Paper · January 2002

Source: DLR

CITATIONS READS

535 11,702

4 authors, including:

Daniel Krajzewicz Christian Feld

German Aerospace Center (DLR) Forschungszentrum Jülich
140 PUBLICATIONS   4,640 CITATIONS    19 PUBLICATIONS   1,051 CITATIONS   

SEE PROFILE SEE PROFILE

Peter Wagner
German Aerospace Center (DLR)
255 PUBLICATIONS   5,445 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Scalasca View project

iTETRIS View project

All content following this page was uploaded by Peter Wagner on 05 June 2014.

The user has requested enhancement of the downloaded file.

 SUMO (Simulation of Urban MObility)
An open-source traffic simulation
Daniel Krajzewicz, Georg Hertkorn and Peter Wagner Christian Rössel
German Aerospace Centre, Centre for Applied Informatics Cologne
Institute for Transportation Research Weyertal 80, 50931 Köln, Germany
Rutherfordstr. 2, 12489 Berlin, Germany E-mail: roessel@zpr.uni-koeln.de
E-mail: Daniel.Krajzewicz@dlr.de,
Georg.Hertkorn@dlr.de, Peter.Wagner@dlr.de

KEYWORDS available, simulation is the only way to show weak

traffic simulation, microscopic, continuous, points of the street network or predict its traffic.
multimodal, open source, car-driver model, traffic For this purpose, many simulation software packages
research, road traffic were developed. Some of them were tested within the
SMARTEST-project (SMARTEST 1999).
Such traffic simulation software packages differ in
ABSTRACT their portfolios of modelled artifacts as well as in their
As no exact model of traffic flow exists due to its high usage paradigm: Some are conceived as Windows-
complexity and chaotic organisation, researchers using applications while other which are rather tools
mainly try to predict traffic using simulations. for traffic researchers, are simple command-line tools
Within this field, many simulation packages exist and or programs. SMARTEST also holds a list of features
differ in their software architecture paradigm as well as that such packages have.
in the models that describe traffic itself. SMARTEST was mostly dealing with microscopic
We will introduce yet another system which, in simulations. Simulations of this type regard single
contrast to most of the other simulation software vehicles as atomic parts, not the whole traffic flow
packages, is available as an open-source program and itself (macroscopic) or single parts of the vehicles or
may therefore be extended in order to fit a researcher’s drivers (sub-microscopic). Such vehicle models may
own needs and also be used as a reference testbed for be discrete in time and space using cellular automata or
new traffic models. only discrete in time (Nagel and Schreckenberg 1992,
Brockfeld et al 2001) or may even be fully continuous
models. Figure 1 shows the difference between space-
INTRODUCTION discrete and space-continuous simulations.
When trying to improve traffic, a valid model to work
with is needed.
Although some people may assume traffic can be
described by departure times and routes with certain
durations, traffic is highly conditioned by an
individual’s private wish for mobility – making up
around 60% of traffic – and due to this, neither
departure times nor fixed and earlier known routes are
available. This is a great problem for modelling traffic
itself. Especially the private transit leads to an
impossibility of describing traffic by the use of
Figure 1: Space-discrete vs. Space-continuous
mathematical formulas. Both, a modern human being’s
simulation
wishes to leave and arrive at certain places and at
certain times on the one hand, and the movement of the
From a researcher’s point of view, when using the
vehicle on the street on the other, influence traffic and
available simulation software packages, several
one another: the street network’s work load depends on
problems may arise due to their availability as a ready-
the drivers’ departure times and determines the speed
to-use software black-box, especially when
of movement. Vice versa the load affects the departure
commercial products are regarded.
times of the drivers’ as they wish to move fast and
At first, one can not examine the underlying model of a
arrive at a certain time.
simulation. Also, due to different software
Beside this, traffic is conditioned by values like the
architectures, a comparison of different models‘
weather, the infrastructure within the region or other
features (like simulation speed, its ability to describe
incidents affecting the system.
the reality, etc.) is difficult if not impossible.
This complexity yields in varied behaviour of the
Furthermore, such simulation tools can not be
whole system where system means the generation of
spontaneously extended by introducing ones own ideas
traffic and traffic itself, and as no valid mathematical
such as new types of sensors, measurements or models.
models that take into account all these influences are
To introduce a tool which accomplishes these not yet is moving on. The simulation of street vehicles is time-
supported tasks, our institute – together with the Centre discrete and space-continuous. As our car-driver model
for Applied Informatics (Cologne, Germany) – is is continuous - as the majority of car-driver models are
working on a traffic simulation software called SUMO - we decided to use this approach.
(„Simulation of Urban MObility“). In fact, this When simulating traffic, the street attributes, such as
software is a continuous, microscopic and multi-modal maximum velocity and right of way rules, are
traffic simulation and is – in spite of it’s name - also regarded.
capable of modelling traffic on networks larger than
single cities, e.g. highway networks, without any Features
changes.
In the current version – 0.7 – SUMO contains the
THE SIMULATION following features:
- collision free vehicle movement
Basic Paradigms - different vehicle types
- multi-lane streets with lane changing
SUMO is conceived to simulate a traffic road network - junction-based right-of-way rules (junctions with
of the size of a city. streets having equal / different priorities, e.g.
As the simulation is multi-modal, which means that not right-before-left)
only are car movements within the city modelled, but - lane-to-lane connections
also public transport systems on the street network, - a XML-raw-output containing information about
including alternative train networks, the atomic part of the state of the net for every time step
the simulation is a single human being. This human - detectors with independent GnuPlot or CSV
being is described by a departure time and the route (comma separated value) – output
he/she takes which again is made up of subroutes that - input from XML-files which may be spread over
describe a single traffic modality. several files for a better handling

Car-Driver Model
The model used currently within SUMO is the Gipps-
model extension (invented and described in: Krauß
1998, Janz 1998). It is capable of displaying main
Figure 2: Multimodality
features of traffic like free and congested flow.
In each time step the vehicle‘s speed is adapted to the
Thus, a simulated person may take his/her car to the
speed of the leading vehicle in a way that yields to a
nearest public transportation system station and
collision-free system behaviour within the following
continue his travel by other means of transport. Apart
simulation step(s).
from movements using motorised vehicles, a person
This velocity is called the safe velocity vsafe, and is
may also walk. Walking is not simulated at all but is
computed using the following equation:
modelled estimating the time the person needs to reach
the destination. Figure 2 displays a such a compound
route.

vl(t): speed of the leading vehicle in time t

g(t): gap to the leading vehicle in time t
tau: the driver’s reaction time (usually 1s)
b: the deceleration function

To bind the acceleration to the vehicle’s physical

abilities, the resulting “wished” or “desired” speed is
computed as the minimum of the vehicle‘s possible
maximum velocity, the vehicle’s speed plus the
Figure 3: Different simulation classes (from left to maximum acceleration with the safe velocity computed
right: macroscopic, microscopic, sub-microscopic) as shown above, therefore a vehicle will not “drive” or
“accelerate” faster than is possible for it:
The traffic flow is simulated microscopically. This
means, that every vehicle that moves within the
simulated network is modelled individually and has a
certain place and speed. In every time step which has a Further, the driver is simulated by assuming he is
duration of 1s, these values are updated in dependence making errors and so fails to perfectly adapt to the
to the vehicle ahead and the street network the vehicle
desired velocity. This is done by subtracting a random downloaded as free implementations exist (STLPort
“human error” from the desired speed: 2001). The code is well documented and formatted and
we will follow Ellemtel-guidelines (Ellemtel 1990-
1992) that assure portability compatibility with most
systems.
As the vehicle must not drive backwards, once again – Due to this, our software is compilable using most
after the previous computations – the maximum of the platforms and we validated this for the following
computed speed and zero must be taken and will be the environments:
vehicle’s final speed for the current time step. - Windows using MSVC++
This model is collision-free to allow simulations - Solaris using SUN-C++-compiler and STL-Port
without any artifacts that arise from the imperfection - Linux using gcc
of the underlying model (An impressive in-depth
description of the model and the underlying Simulation Benchmarks
assumptions and rules may be found in Krauß 1998
and Janz 1998). The simulation is capable of simulating large cities like
Berlin, Munich or Cologne on a normal desktop
computer. Most of our tests were done on Intel PC’s
Traffic Lights running at 933MHz and having 256MB of memory.
Traffic lights play an important role within the traffic The simulation is capable of simulating around
management as they improve traffic flow. Apart from 1Million vehicle movements per second depending on
simple right-of-way rules, each simulated junction may the network’s complexity. Further optimisations will
also be a junction with traffic lights. As some junctions follow.
in Germany allow to ignore the red stoplight when
turning right, an extension to the right-of-way rules Extensibility
regarding this is being implemented.
Hoping for the participation of other interested
persons, we try to supply potential developers with all
Simulation Output information needed to extend and modify the program.
By now, two different outputs are available. The first is Beside things like intelligent traffic lights, models for
a so-called „raw“ output which contains all edges cars equipped with ACC systems, etc., we hope to
(streets) and all lanes along with the vehicles driving supply interested researchers with a common testbed
on them for every time step, where vehicles are for their microscopic models.
described by their name, position and speed. This The documentation (SUMO 2002) shows how to
output is complete and may be used as input to post- extend the simulation – and other tools.
processing tools for evaluation. However, a large
simulation produces a nearly unmanageable amount of ONGOING PROJECTS
data, so other outputs have been invented.
The first processed output that may be generated by the Within our institute some projects use SUMO to
simulation, is a log-file made by simulated detectors validate their assumptions about new technologies. The
which may be positioned on a certain position of a following projects have already started using SUMO or
certain lane. These detectors are a simulation of induct will use it within the next months:
loops and are able to compute the flow, the average - A project from California, investigates whether
velocity on the lane, and other values. The results of the detector loops spread on different highway
this computation are written into a file using the CSV lanes may be used to predict jams.
or the GnuPlot-format. Every detector has it’s own file. - We try to validate our traffic flow predictions
based on floating car data retrieved from Taxis in
Berlin, Vienna and Würzburg .
Software Development - An internal project is trying to predict the
Being a tool for traffic researchers, SUMO is designed advantages of new sensor technologies, on route
to be fast and exact instead of trying to be a software prediction and the concluding traffic light phases.
that is pleasant to look at. So, although the - A project from the US tries to improve traffic
implementation of a GUI will be one of our next tasks, within a highway off-ramp area.
the main program is meant to be started from a
command line and produce an output which must be ADDITIONAL TOOLS
post-processed when one wants visual results. This
SUMO consists of more than a single application.
prevents data arising from the GUI from slowing down
Some other modules exist that allow to build assigned
the system, giving more memory and system time to
data needed for simulations and research.
the simulation itself.
The following modules are now being developed:
SUMO is implemented in C++. During development,
we try to use only standardised parts of this language.
One of them is the STL which – when not coming
directly with the compiler – may be additionally
Sumo-netconvert SUMO-NETCONVERT is also responsible for the
building of traffic light phases where either the
Due to it’s high complexity, the SUMO’s network priorities of the streets that build up a certain junction
description is not meant to be generated by a human or their traffic flows are used. As in the real world – at
user. Instead, we use this tool to convert common data least in Germany – phases of traffic are adapted
like lists of edges and optional nodes into a complete individually for each junction and the certain phase
SUMO-network. data may not be available for all junctions, we use
During this process, SUMO-NETCONVERT reads in heuristics to generate a realistic output.
the available data, computes the needed input for
SUMO and writes the results into a XML-file.
By now, four different input formats may be converted
into SUMO-networks:
- simple XML-data containing edge types, nodes
and edges
- CSV-data containing edge types, nodes and edges
- Cell-input files (Cell is a queue-simulation
developed by the ZAIK)
- VISUM-networks
As SUMO is used within the INVENT-project, some
further import functions will be implemented within
this year: ArcView, VISSIM and possibly GDF where
an indirect import of GDF is already possible by
converting it into XML using a script and then using
the generated XML-descriptions as input to SUMO-
NETCONVERT.
The next figure shows which data may be computed
from a simple list of nodes and edges. The flow is Figure 5: The city of Berlin build from high-quality
aligned from left to right and from top to bottom. The digital map data
first step is to determine the priorities on the junction,
the second to compute the relationships between the Sumo-router
lanes and edges that may be reached, and within the
third step, the destination edges are split among the Beside the static part – the network – the simulation
incoming lanes. The computation is flexible and consists of moving vehicles. With the increase of the
depends on the number of incoming and outgoing quality of simulations, the need to model a
edges and their sizes as well as the priorities they have populations’ mobility has increased as well. In such
within the network and on the resulting type of cases, vehicles are not spread statistically over the
junction. network, instead a single person’s daily plan consisting
of routes with certain departure times is used. While
data needed to describe the departure times and a
route’s origin and destination are given, the routes
themselves must be computed. To avoid online-
computation of these during the simulation, this
computation is done using a separate module, the
SUMO-ROUTER. This module reads the departure
times, origins and destinations for a set of virtual
humans that will be simulated, then computes the
routes through the network itself using the well-known
Dijkstra routing algorithm (Dijkstra 1959).

Figure 4: Conversion of simple, plain network data

into a complete description

The next picture shows the need of such a net-building

Figure 6: Routing in a small network
tool when a whole city is wished to be modelled.
Neither the number of the streets nor the complexity of
As the speed on the streets changes with the traffic
their relationships allows the processing of the network
amount and therefore the computation of routes using a
description by a human user.
network where the traffic is not yet known does not
 regard the real-world situation, the routing will be done programming languages. This, too, will be
using the Dynamic User Equilibrium approach investigated.
developed by Christian Gawron (Gawron 1998) where Also, as the amount of traffic to be simulated is
routing and simulation are repeated several times to growing and some research experiments need several
achieve a real-world behaviour of drivers. simulations to compute a single value (for instance in
Furthermore, the router supports dynamic network case of traffic light optimisation where several timing
load: the fact that the load on a depends on the time of schemes must be tested), the system will be extended
day is also regarded. to allow the usage of computer clusters.

Planned Participation
Other tools are planned due to the growing number of We also hope to gain help from other persons or
different projects investigating varoius topics and institutes that are interested in traffic simulation and
methods. want to participate on SUMO’s development by
Some planned modules are: extending, improving or simply using and criticizing it.
- traffic light optimisation We highly invite you to visit the SUMO-pages at
- post-processing tools for raw-output evaluation http://sumo.sourceforge.net.

CONCLUSIONS REFERENCES
We try to build and establish a common platform for Brockfeld, E. et al. 2001. „Optimizing Traffic Lights in a
traffic research by providing a simulation tool that may Cellular Automaton Model for City Traffic“. In: Physical
be applied by non-programming users in a simple way Review E 64, 056132
supporting them with methods and tools mostly needed E. W. Dijkstra. 1959. „A note on two problems in connexion
with graphs“. In: Numerische Mathematik, 1:269-271.
when working on traffic simulations. Due to it’s high Directorate General VII - Transport of the European
portability, the tool may be used on different operating Commission. „SMARTEST home page“.
systems. http://www.its.leeds.ac.uk/projects/smartest/
Besides this, the platform is also extendable by others Ellemtel. 1990-1992. Coding Standards Page.
in order to allow them to improve the software and http://membres.lycos.fr/pierret/cpp2.htm
include ideas we have not taken into account. Christian Gawron. 1998. „Simulation-Based Traffic
Additionally, models originally implemented may be Assignment“. Inaugural Dissertation.
replaced by own methods allowing their comparison to Stefan Janz. 1998. „Mikroskopische Minimalmodelle des
the existing models in respect to the simulation quality Straßenverkehrs“. Diploma Thesis.
Stefan Krauß. 1998. „Microscopic Modeling of Traffic Flow:
and speed. Investigation of Collision Free Vehicle Dynamics“.
Hauptabteilung Mobilität und Systemtechnik des DLR
FUTURE WORK Köln. ISSN 1434-8454
K. Nagel, M. Schreckenberg. 1992. Journal of Physics I 2,
Ongoing Work 2221
STLPort. 2001. Company Homepage.
Further work will be done as this simulation will be http://www.stlport.org/index.html
used by our institute for different purposes. Projects SUMO: G. Hertkorn, D. Krajzewicz, C. Rössel. 2002.
concerned about the verification of floating car data, SUMO Homepage. http://sumo.sourceforge.net
the prediction of improvements on traffic foreseeing TrollTech. 2002. Troll Tech Homepage.
and traffic light optimisation through new sensors that http://www.trolltech.com
observe traffic will increase the need to extend the
simulation software by artifacts reflecting these BIOGRAPHY
devices and so will increase the simulation’s
Born in Bydgoszcz, Poland, 1972, DANIEL
complexity.
KRAJZEWICZ has finished his study of computer
To trace and validate the simulation, a GUI will be
science at the Technical University in Berlin by the
implemented where the loaded map, together with
middle of the year 2000 with artificial intelligence and
vehicle movements and generalised information, will
computer graphics as main topics. After work on text
be displayed. To stay portable, the GUI will be
classification he changed to the Institute for
implemented using the Qt-windowing library
Transportation Research of the German Aerospace
(TrollTech 2002) which is free to use and available for
Centre where he now works on a cognitive driver
most systems including Windows, UNIX/Linux and
model and an open-source urban traffic simulation.
Macintosh. Two different types of GUI support will be
implemented. While the first application is the
simulation itself extended by a windowing system, the
second will be an interface that soley displays data
coming from simulation module(s).
Some running projects have shown the need to
integrate the software into other software packages or
to build interfaces between the software and other

View publication stats