2018 21st International Conference on Intelligent Transportation Systems (ITSC)

Maui, Hawaii, USA, November 4-7, 2018

Microscopic Traffic Simulation using SUMO

Pablo Alvarez Lopez, Michael Behrisch, Laura Bieker-Walz, Jakob Erdmann, Yun-Pang Flötteröd,
Robert Hilbrich, Leonhard Lücken, Johannes Rummel, Peter Wagner and Evamarie Wießner

Abstract— Microscopic traffic simulation is an in- lot of model extensions, simulation enhancements and
valuable tool for traffic research. In recent years, both improvements have been made. This publication presents
the scope of research and the capabilities of the tools an overview of the functionality and the used simula-
have been extended considerably. This article presents
the latest developments concerning intermodal traffic tion models in SUMO. First, a general introduction of
solutions, simulator coupling and model development the work flow with traffic simulations is given. Second,
and validation on the example of the open source example scenarios are presented which are also freely
traffic simulator SUMO. available to give researchers the chance to start with
their research without the need to setup a complex
I. Introduction
scenario first. Next, different models, concepts and tools
For the implementation of traffic management solu- are explained to support the work of traffic researchers.
tions accurate knowledge of the traffic conditions and Finally, an outlook of the works with SUMO is stated.
dynamics is necessary. Traffic simulation frameworks pro-
vide a helpful tool to answer complex research questions, II. Workflow
to evaluate or to test traffic management strategies and In order to simulate traffic, a number of elements are
their impacts. The traffic simulation tools can mainly be needed. The most important ones are the following
divided into four different groups [17]:
• Network data (e.g. roads and footpaths)
1) Macroscopic: average vehicle dynamics like traffic • Additional traffic infrastructure (e.g. traffic lights)
density are simulated • Traffic demand
2) Microscopic: each vehicle and its dynamics are
Together these elements form a simulation scenario.
modeled individually
Since traffic simulation models are typically used for
3) Mesoscopic: a mixture of macroscopic and micro-
stochastic behavior, it makes sense to simulate such a
scopic model
scenario a number of times and draw statistical conclu-
4) Submicroscopic: each vehicle and also functions
sions.
inside the vehicle are explicitly simulated e.g. gear
shift It is often a time-consuming process to prepare a
simulation scenario based on real-world data. SUMO
The advantage of macroscopic models are normally its provides a large package of applications to help with this
fast execution speed. However the detailed simulation of task. The process for preparing the road network data
microscopic or submicroscopic models are more precise and traffic lights is described in section IV. Preparing
especially when emissions or individual routes should be traffic demand is explained in Sections V. Section VI
simulated. Therefore, this paper focuses on microscopic deals with multi- and intermodal traffic while Section
traffic simulation. VII focuses on simulation validation. Moreover, sections
There are several microscopic simulation tools avail- VIII and IX describe pedestrian simulation and model
able to support the evaluation of research questions. For enhancements respectively. After that, model coupling
example, the Simulation PTV Vissim is a commercial work will be explained in Section IX.
software package, which is well known and used [7]. PTV After defining a scenario it is immensely useful to
Vissim is well supported and provides a user friendly observe the simulation objects (vehicles, pedestrians,
interface with also 3D visualization. In contrast to the traffic lights) in a visual representation for qualitative
commercial PTV Vissim software, the activity-based validation. To this end, SUMO provides the SUMO-
traffic simulation MATSim is open source and freely GUI application, which allows observing the simulation
available over the internet [14] 1 . at different speeds and with various coloring options to
Furthermore, SUMO is also freely available and pub- highlight various aspects such as speeds, traffic densities,
lished under the Eclipse Public License V2 2 . SUMO road elevation or right-of-way rules.
is used worldwide and is downloaded over 35.000 times To evaluate a simulation scenario quantitatively, the
every year. A reference publication of SUMO was already simulation provides a wide array of output files that can
published and highly referenced in [15]. Since then a be enabled selectively, such as:
1 https://www.matsim.org/ • Vehicle trajectories (positions and speeds)
2 http://sumo.dlr.de/ • Traffic data collected from modeled detectors

978-1-7281-0323-5/18/$31.00 ©2018 IEEE 2575

Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 Fig. 1. Street Network of the Real-World Bologna Scenario in
SUMO’s WebWizard

Fig. 2. Example Traffic Flow of an Induction Loop

•Traffic data aggregated over network elements
(edges or lanes)
• Traffic data aggregated over the whole trip of a
observed days. The data is measured every 5 minutes and
vehicle or person
are smoothed with the Savitzky-Golay filter. Similar to
• Protocols of traffic light switching
other cities, only few vehicles were driving during the
• Traffic data aggregated for the whole simulation
night, while the traffic demands increased during the
• Emissions, energy consumption, and so on.
rush hours in the morning and in the afternoon. Based
These output files can be visualized using SUMO tools on these raw data, a one-hour time span, representing a
or imported into other applications. To help with this, typical 8am-9am rush hour, was created and used as the
SUMO also provides tools for converting output files into traffic demand in the “Real-World Bologna” scenario.
other formats and for importing them with python or Different types of personal vehicles are modeled in
matlab. order, to simulate a more realistic distribution of differ-
III. Example: Bologna Scenario ent vehicles considering produced emissions, acceleration
and deceleration behaviors. In addition to the vehicular
A scenario is used throughout this entire paper to
traffic, the scenario includes public transport, bus stops
illustrate the concepts and work flow of SUMO. The
and special lanes. To show the functionality of inter-
scenario is called “Real-World Bologna” as it is based
modal traffic, the Acosta scenario was extended with
on a part of the inner city of Bologna (Italy). It focuses
some fictional person trips.
on the area between the two main streets Andrea Costa
Furthermore, the vehicles in the simulation have all
and Pasubio and it includes the football stadium as well
different types of emission classes which are defined
as the hospital, see Figure 1.
in the vehicle types. For the emission simulation, the
The “Real-World Bologna” scenario is publicly avail-
database of HBEFA (Handbook Emission Factors for
able 3 as part of SUMO to provide a quick start. However,
Road Transport) has been used. It is possible to simulate
for more complex or larger evaluations the scenario may
the produced emissions of CO, CO2, NOx, PMx, HC,
be too small or the traffic demand could be too less.
as well as the fuel consumption of every single vehicle
Figure 1 shows the street network that is modeled in
[16]. In addition to the HBEFA model the produced
the scenario. It is large enough to develop and to eval-
emissions can be also simulated by the PHEM model
uate complex traffic management strategies, while still
(Passenger car and Heavy duty emission Model) of the
executing quickly on regular desktop computers.
Technical University Graz. PHEM and HEBEFA include
The positions of the traffic lights were given as teleme-
both emission data from real world vehicle. From this
try data files by the municipal of Bologna and have been
data different emission classes for vehicles types have
imported in the SUMO scenario. In addition, the signal
been extracted which could be used in SUMO for the
time plans of the traffic lights were also included in the
emission simulation.
simulation scenario.
The traffic demand for the scenario is based on induc- IV. Network Setup
tion loop data from real-world detectors in the city of SUMO networks consist of nodes and unidirectional
Bologna. The vehicle counts of the induction loops during edges representing street, waterways, tracks, bike lanes
11 and 13 November 2008 were provided by the city and walkways. Each edge has a geometry described by
administration. As an example, Figure 2 shows the traffic a series of line segments and consists of one or more
flows detected by a single induction loop for the three lanes running in parallel. Attributes such as width,
3 https://sourceforge.net/projects/sumo/files/traffic_ speed limit and access permissions (e.g. bus only) are
data/scenarios/Bologna_small/ modeled as constant along a lane. As a consequence a

2576
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 stretch of road must be modeled as a sequence of edges An example of the tool and the Bologna scenario can
when either of these attributes changes along its length. be seen in Figure 1.
SUMO networks include detailed information regarding
possible movements at intersections and corresponding B. Development of the Real-World Bologna scenario
right of way rules which are used to determine the The municipal of Bologna provided the traffic demand
dynamic simulation behavior. To ensure consistent net- and the information about the traffic network for the
work representation, SUMO networks are created using Real-World Bologna scenario as a Vissim scenario [7]
the NETCONVERT and NETEDIT applications. NET- which could be converted to a SUMO scenario and was
CONVERT is a command-line tool which can be used extended and corrected automatically and manually. One
to import road networks from different data sources e.g. major extension of the traffic network was to include
OpenStreetMap (OSM), OpenDRIVE, Shapefile or from lanes which are explicitly restricted for buses only. The
other simulators such as MATSim and Vissim. A key positions of the traffic lights and their signal time plans
feature of NETCONVERT is the heuristic refinement of could also be provided from the municipal of Bologna as
missing network data to achieve the necessary level of telemetry data.
detail for microscopic simulation (i.e. synthesizing traffic
light plans, right-of-way rules and intersection geometries V. Demand Modeling and Adaption
for OSM networks). One of the premises to be able to simulate traveling
NETEDIT is a graphical network editor which can be activities in a given network is to provide traffic demand.
used to create, analyze and edit network files. This serves In SUMO, traffic demand can be defined as individual
to complement the network generation heuristics with trips, flows or as routes. The basic information should
manual refinements and also supports defining additional include departure time, origin, destination and transport
traffic infrastructure which could not be imported by mode where transport mode can be different kind of
NETCONVERT. Among the support features are defin- vehicles or pedestrian. When traffic demand is defined as
ing: routes, information about origin and destination is not
• Basic network elements (junctions, edges and lanes) mandatory necessary. Instead of that additional route
• Advanced network elements (e.g. traffic lights) information, defined as a sequence of edges, is required.
• Additional infrastructure (e.g. bus stops) Once traffic demand is generated, traffic assignment in
• Polygons and points of interest (POI) SUMO, i.e., DUAROUTER, MAROUTER or Oneshot,
can be executed for understanding the traffic state of
Due to the frequent mismatch between available input
the investigated network. The aforementioned assign-
data and the necessary level of detail for microscopic sim-
ment methods are mainly different in their assignment
ulation, network and infrastructure preparation is often a
principles (User Equilibrium (UE), Stochastic User Equi-
challenging task. For this reason, NETCONVERT is un-
librium (SUE) or the fastest route at a given departure
der constant development to improve it’s heuristics and
time). In the following, different ways to generate or to
reduce the required amount of manual editing. Recent
import traffic demand in SUMO will be firstly intro-
years have seen a number of improvements in regard to:
duced. The possibilities to modify traffic demand will
• Easy of use (OSM wizard) then be explained.
• Importing new data formats (i.e. OpenDRIVE)
• Modeling networks with left-hand traffic A. Demand importing and generation
• Importing public transport information Origin and destination (O-D) matrix, as one of the out-
• Modeling 3D Topology puts in the traditional 4-steps transportation forecasting
• Generating multi-modal networks for vehicles, bikes model, is the typical way to describe traffic demand
and pedestrians between given traffic analysis zones since decades. Each
O-D matrix cell describes the respective traffic demand
A. Generation of scenario with OSM as the amount of transport mode, such as passenger cars
A useful tool for initial scenario preparation is the and trucks. Such data cannot be directly used in SUMO.
osmWebwizard application which is part of the SUMO The tool OD2TRIPS provides a way to convert such O-D
python tools. This tool allows generating a simulation matrices into individual trips. In order to generate the
scenario using a few clicks in in a browser-based interface. departure time for each vehicle/person either uniform
The user selects an area from a map display along with or random distribution with in a given time period can
a set of parametrized traffic modes. The tool uses this be chosen. Currently, O-D matrices either in Vissim or
information to download and import the network data Visum format can be processed.
from OpenStreetMap with parameters corresponding to It is also possible to import and generate traffic
the selected traffic modes. Random traffic is then gener- demand from an agent-based demand model. SUMO
ated to populate the network. This moves through the can already couple with the agent-based demand model
network according to a dynamic route assignment which TAPAS [25], also developed by DLR. Through iterative
includes public transport routing for individual persons. computation the data about simulated people’s routes,

2577
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 departure times and chosen transport mode can be de- rently, there are known deficiencies when applying
rived. Such route data can be then directly used as traffic this tool with highly meshed networks, such as city
demand in SUMO. networks. Flowrouter is suggested to use as alterna-
Moreover, SUMO has four tools to generate demand tive.
with use of different data sources. • JTRROUTER
The purpose of JTRROUTER is to generate routes
• ACTIVITYGEN
with the given traffic volumes and turning ratios at
This tool was contributed by the SUMO users from
intersections, on ramps or at interchanges. When
the University of Munich and was a part of the
using JTRROUTER for intersections in a city area,
project Fidens [5]. This tool aims at generating
it is possible that some vehicles may run in circle
demand for a synthetic population in a large net-
since traffic volumes will be split only according to
work. It requires a road network generated by NET-
turning ratios. The edges, already used by vehicles,
CONVERT or NETGENERATE, and a population
are not taken into consideration.
definition. The latter one is done by using a sim-
• randomTrips.py
ple activity-based traffic model where the activities
This is the simplest way to generate trips as syn-
related to work, school, leisure, and work and the
thetic traffic demand for a given network. Edges
four transport modes, i.e. walk, bicycle, car and
selected as sources or/destinations are decided either
bus, are considered. The required input statistics
uniformly at random or with a modified distribu-
data should include general information, such as the
tion. The probabilities to select an edge as source
number of inhabitants, the number of households,
or/destination can be basically weighted by edge
the age to retire, probability to be unemployed, the
length, edge speed and/or the number of lanes. Trips
special context or state of the city, population’s age
are generated according to the defined repetition
brackets, work hours, population and work position
rate and will be distributed evenly in an interval
distribution, city boundaries, schools and bus lines.
between the defined begin and end time. Different
More detail information can be found in [22]. AC-
transport modes can be considered as well. If a given
TIVITYGEN’s output will be a route file, which
network is not fully connected, it may happen that
contains start edge, end edge and optionally some
no routes between certain sources and destinations
stopovers only. Complete routes need to be gener-
can be found. Such trips will be discarded during
ated by traffic assignment models, such generated
simulation. It may be necessary to increase the
by SUMO’s DUAROUTER.
number of trips to ensure the number of usable trips.
• Flowrouter
When using the option “route-file”, a respective
The main concept of Flowrouter is to solve a max-
route file will be automatically generated with use
imum flow problem in a given network where avail-
of DUAROUTER.
able detector data are used as road capacities. When
no type information in the input detector file or B. Demand adaptation
the option revalidate-detectors is set, all edges are
Due to limited demand data availability, changes in
classified into the following three types:
daily activities or unexpected events either imported or
– Source: any edge without incoming edges generated traffic demands may not correspond to actual
– Sink: any edge without outgoing edges traffic situations. To address this issue, three methods are
– In-between: any edge which is neither source nor available and explained below, where the first two meth-
sink ods are mainly for daily traffic demand while the last
Both routes and flows with respective routes can be method mainly deals with unexpected situations, which
generated as outputs. To increase route plausibility, result in road or area closure, and traffic guidance, such
two options: limit and edge restriction can be set. as guidance to alternative parking lots or to alternative
The former one helps to distribute flows more evenly routes.
on possible routes while the latter one specifies the The first method is to apply the open source program
maximum flow for a route which consists of the given Cadyts (Calibration of Dynamic Traffic Simulations),
edges. which is Bayesian-based route choice calibration model
• DFROUTER [9]. This program aims at the correspondence between
The main concept of DFROUTER is to generate estimated and measured flows by adjusting a prior dis-
routes on each detector with the detected flows tribution of route choices given traffic flow measure-
on highways and the respective ramps and inter- ments. A coupling between SUMO and Cadyts has been
changes, since such data has become more and more established and applied. In addition to flow measure-
available due to the extensive usage of detectors. It ments, an initial route set needs to be given, e.g. from
is possible to generate routes only between source DUAROUTER. Cadyts and SUMO will then interplay
detectors and destination detector or between all with each other until the convergence or the maximum
detectors, by setting the option “routes-for-all”. Cur- number of the iterations is reached. In each iteration,

2578
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 SUMO provides the updated traffic state to Cadyts for SUMO does not separate the different individual vehicle
adjusting route distribution. According to the updated types by default, it just gives every person a list of
route distribution, SUMO delivers the corresponding available vehicles and it is up to the user to define the
traffic state. Traffic demand and departure times can be types of those vehicles. So every person can have multiple
also scaled and adjusted respectively. cars and/or bikes to their availability. Every vehicle has a
The second method is to use the calibrators imple- defined capacity and can as such transport multiple per-
mented within SUMO package. These calibrators allow sons. The feature which separates public from individual
the dynamic adaption of traffic flows and speeds. Accord- transport in SUMO is that public transport consists of
ing to the given induction loop measurements, i.e. speeds a set of vehicles (line) serving a given sequence of stops
and flows, the calibrator will remove or insert vehicles (not necessarily on a fixed schedule) where the person
so that the simulated flows comply with the measured is free to choose a different vehicle of the same line to
flows. Moreover, road speeds will be also adjusted accord- fulfill the mobility request. Defining the public transport
ingly. The respective adjustment is similar to the way in SUMO is not different from defining individual traffic.
a variable speed sign works. With the consideration of All vehicles can have stops with predefined leave times
measured flows and speeds invalid jams will be avoided which can be used to model the schedule of a public
by removing jammed vehicles so that jams will not spill transport vehicle.
back to a calibrator. When inserting new vehicles, the The list of stops for a public transport line can be ex-
applied routes area based on the route distributions on tracted automatically from sources like OpenStreetMap,
the respective road sections. Such route distributions a feature which is available with OSMWebWizard de-
can be either defined by users or based on the route scribed in Section IV. Using this as an input one can
distribution for all vehicles that passed the corresponding define a schedule for vehicles using this route and so
road sections in a given interval. More details about the provide a scheduled public transport. With the web
applied algorithm can be found in [6]. wizard a schedule of one vehicle every ten minutes is
The last method is to use the REROUTER, imple- used. To allow the transfer between the road and the
mented within SUMO package. With use of REROUTER rail network it is possible to introduce access facilities
vehicles’ routes and even destinations can be modified at stopping positions which model the transfer with a
when they travel over the predefined roads, on which fixed time penalty for the individual person. These access
they can make a favorable choice, i.e. route, parking lot points can also be generated automatically within a fixed
or destination, according to the given choice set and the range around a rail station.
current traffic state, i.e. travel time. When choosing new Having individual vehicles and public transport at
routes, it is also possible to terminate routes. The respec- their disposal, persons can find their fastest way through
tive vehicles will then immediately leave the simulation the network using all available modes. To facilitate
and counted as arrived at their current positions on the this, SUMO can calculate intermodal routes with the
rerouting roads. While choosing new destinations, the included DUAROUTER for the process of calculating
fastest routes from vehicle’s current positions to the new routes before the simulation as well as directly within
destinations will be computed accordingly. When choos- the simulation. This way it suffices for the user to define
ing parking lots, the available parking capacities in the depart and end position as well as the available modes
available parking plots are considered. If vehicle’s final and the depart time and SUMO will find the fastest
destinations are not the respective parking lots, vehicles, routes for walking driving and using public transport
which reroute to a new parking area, will continue to (and combinations thereof). Together with a randomized
their original destinations after finishing the stops. In demand as in the fictional Bologna scenario this can give
addition, REROUTER can also be used to define closed a very fast pseudo-realistic scenario with walking and
roads with the consideration of vehicle types. Researches riding persons.
with the REROUTER application can be found in [12]. Independently, it is also possible to give a fine grained
individual route to persons and also model the transfer
VI. Multi- and Intermodal Traffic
of goods (containers) in a similar way. Containers can
The central element in modeling mobility with dif- currently not be used with the automatic routing system
ferent modes of transport is the person. Every person and need a hand crafted route to use different modes of
in SUMO can have an individual plan when and with transport.
which means she wants to travel. Therefore she has an
availability of transport modes which can change from VII. Validation
trip to trip. The available modes and activities in SUMO Microsimulation tools seek to describe reality on many
are: different levels, which raises questions of the validity of
• Walking the results obtained. To name but a few of these levels,
• Riding an individual vehicle (car, bike, etc.) this is the input itself (which is often coming from a
• Public transport digital network), the demand for transport which is also
• Stopping or other non-mobility related activities not intern to SUMO but stems from some outside tools,

2579
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 the modeling of the dynamics and behavior of the traffic
objects, up to the computation of emissions and other
side-effects of traffic.
Unfortunately, there is no one-stop-shop when it comes
to these issues, instead, permanent vigilance and scrutiny
is needed. Whenever there is an opportunity to check
for the correctness of your model, then this should be
done. Any piece of real data is of help here, the data are
often not too far away. It sometimes helps to position
yourself near the intersection you want to model and
simply compare real-life with simulation-life on a visual
basis.
On a more quantitative level, which is always recom-
mendable, this is called calibration and validation [4],
[8]. The output of the simulation should be as close Fig. 3. Simulation data in relation to measured traffic data from
as possible to real data (validation), and in order to induction loops
reach this goal it might be necessary to adapt some of
the parameters of SUMO (calibration). Which raises the
question how to quantify ”closeness”, and here many parison on the network level has been performed, which
different measures have been proposed [4]. In the experi- is presented in Fig. 3. This presentation is typical for
ence of these authors, there is not much to be gained by a network validation; it displays the simulated counts
using something more complex as the RMSE (root-mean on the y-axis with the measured ones on the x-axis.
squared error). Perfect agreement would need the results to lay on the
To give some ideas about the size: when truly micro- diagonal, and from the parameters of a linear fit to these
scopic behavior is analyzed (i.e. the gaps, the speeds, and data additional metrics can be computed to compare
the accelerations), typical validation errors for different simulation with reality.
microscopic models are between 10 and 20 % for trying VIII. Pedestrians
to predict the gap between two vehicles [3]. For networks,
SUMO allows the simulation of pedestrians using
a typical metric used tells that more than 85 % of the
configurable pedestrian models. This can be used to
counts measured in the simulation should deviate less
model the itineraries of persons which alternate between
than 15 % from the measured ones [8].
walking and riding in vehicles. It can also be used to
Obviously, the data {xi }i do not need to come from
model the interactions between vehicles and pedestrians
real data, it might be asked how closely such a simulation
whenever their paths intersect in the road network.
came to reproduce e.g. the data that have been written
SUMO defines an abstract interface to describe the
into handbooks such as the HCM (Highway Capacity
interaction between pedestrians and other simulation
Manual) [18] or it German counter-part, the HBS [2]. The
objects. This interface is currently implemented by three
results [19] indicate, that a SUMO simulation can come
pedestrian models.
fairly close to the curves in such handbooks representing
the fundamental diagram. While this is quite reasonable, A. The Non-Interacting Model
the question of the break-down of traffic flow is a much The simplest pedestrian model in SUMO does not
more complicated one, and here one can see differences consider the interactions among pedestrians nor those
between different models. between pedestrians and vehicles. It is only useful to
A very nice example is also present in the examples track itineraries across the network (i.e. to model the
that come with any SUMO installation: the San Pablo effect of public transport delays) or to generate rough
Dam Road data that have been provided by Carlos trajectories.
Daganzo will be used as the standard example for this.
Given these data, and given a model from SUMO which B. The Striping Model
has a bunch of free parameters, the job of a calibration This model is the default model in SUMO and it
is to find parameters so that the difference between the does model all the interactions listed above. Pedestrians
measured data and the simulated parameters becomes occupy space along foot paths, sidewalks, and pedestrian
minimal. Often, people do this minimization by hand, crossings and may impede the movements of vehicles or
which is not wrong. However, we strongly advise against other pedestrians. To make use of this, pedestrian cross-
doing this, since those by-hand results are almost impos- ings must be modeled as part of the road network either
sible to reproduce. It is much better to use numerical by generating them heuristically with NETCONVERT
tools like a non-linear minimization routine to perform or editing them manually in NETEDIT.
this stunt. The model achieves high execution speed (linear in the
For the above introduced Bologna scenario, a com- number of pedestrians) by only considering close-range

2580
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 interactions along piecewise linear paths. This is done by a constant acceleration, while the calculations for the
separating the available space into narrow stripes along execution of an action step include the whole logic for
the walking direction and only interacting with at most evaluation and decision logic corresponding to the actual
one other traffic object per stripe. traffic situation in the vehicle’s surroundings.
IX. Modeling enhancements C. Sublane Model
A. Numerical Integration Scheme By default, SUMO models traffic where each vehicle
This subsection takes a short glance at some in- drives at the center of its lane and overtaking only takes
ner workings of the simulation process in SUMO. The place using an additional lane. Where this behavior is
simulation updates the vehicles’ positions in temporal insufficient, the Sublane Model within SUMO may be
steps of a user specified duration ∆t moving them by used to model heterogeneous traffic and reduced lane
a positional increment ∆x(t). By default, SUMO uses a discipline. When using this model, lane widths and ve-
first order Euler scheme ∆x(t) = ∆t · v(t + ∆t), where hicle widths are taken into account with a configurable
v(t + ∆t) = v(t) + ∆t · a(t) and a(t) is the acceleration lateral resolution. This allows overtaking within a single
chosen by the underlying car-following model. Physically, lane and also models lateral dynamics such as overtaking
this may be interpreted as the vehicle traveling at a speed, lateral distance keeping and virtual lane forma-
fixed speed within a time step and changing this speed tion.
instantaneously in between two steps. D. Overtaking through the oncoming lane
Since SUMO 0.28 the user can choose to use the ballis-
By default SUMO models lane-changing only with
tic scheme ∆x(t) = ∆t · (v(t) + ∆ta(t)/2), corresponding
lanes that go in the same direction. This implies that
to a movement with constant acceleration for each time
overtaking cannot take place where there is only one
step and therefore yielding smoother vehicle trajectories
lane per direction. To overcome this limitation, SUMO
[24]. As the Euler scheme allows vehicles to change their
includes an optional model for overtaking through the
speed more abruptly than the ballistic scheme, the model
opposite direction lane. To enable this model it is suf-
calibration depends on the chosen scheme. This may
ficient to include additional information regarding the
become especially apparent for relatively large values ∆t
oppositionality of lanes when preparing the network.
of the integration time step, which are frequently chosen
This model is not yet compatible with the Sublane Model
for speeding up computations of large scenarios.
described above.
B. Action Points
X. Simulator Coupling
Per default the effective reaction time for a simulated
With SUMO it is possible to simulate many features
vehicle in SUMO equals the simulation step length ∆t.
around traffic, but there are also limitations. For example
This means that a vehicle’s behavior is affected by the
in the field of vehicular communication normally the ex-
simulation state at time t not before t + ∆t (some
pertise of two fields are needed. First, the communication
exceptions apply to the lane changing process, but this
from the vehicles needs to be modeled. Second, the ve-
out of the scope of this work).
hicles and its trajectories have to be generated. Further-
Since SUMO version 0.32 the simulation step length
more, the applications of vehicular communications often
can be decoupled from the effective reaction time. This
have to influence the traffic situation and therefore the
was realized by allowing to schedule action points at
impact of the application has to be evaluated. Therefore,
intervals, which are multiples of the simulation step
it is necessary to couple SUMO with other simulators to
length [23]. When the action step length (the temporal
produce reliable results.
interval between two successive action points) is set to
For the coupling of two or more simulators an inter-
a different value than the simulation step length, the
face for in-line communication must be provided. This
effective reaction time for a vehicle at a specific point
interface is given in SUMO via TraCI (Traffic Control
in time equals the time until the next scheduled action
Interface) [26]. The user can write a program (e.g. using
point. In particular, the effective reaction time is not
Python) or use a software to retrieve data, start, stop or
constant and bounded from below by the simulation step
modify the simulation via a socket connection to TraCI.
length and from above by the specified action step length.
TraCI has been used for several projects to couple
The action point mechanism lends itself to two impor-
different simulators and evaluate their simulation results.
tant applications. Firstly, it gives a mechanism to control
For example the iTETRIS 4 and COLOMBO 5 projects
driving performance by an individual configuration of
used the iCS (iTETRIS Control System) to couple
the drivers’ maximal effective reaction times. Secondly,
SUMO with ns3 and an application for vehicle-to-vehicle
it can be employed to speed up simulations, which use
(V2V) and vehicle-to-infrastructure (V2I) communica-
a short simulation step length. This is due to the fact
tion. Another program is Veins which uses OMNeT++
that a vehicle’s positional update in a simulation step,
which does not correspond to an action point, is merely 4 http://www.ict-itetris.eu/

a uniform increment of speed and position according to 5 http://www.colombo-fp7.eu/

2581
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.
 and SUMO to simulate vehicular communication [21]. [12] Flötteröd, Y.P., Erdmann, J.: Experiment study on the evacu-
Furthermore, there is also an extension of Veins available ation of bomb alert with sumo. In: Proceedings of the SUMO
User Conference - Traffic, Mobility, and Logistics, pp. 39–50.
which is called Plexe [20]. Plexe allows for example to Berlin, Germany (2016)
simulate platooning for autonomous vehicles and cruise [13] Flötteröd, Y.P., Flötteröd, G.: Route choice calibration from
control models. multi-point vehicle stream measurements. In: 2nd Interna-
tional Conference on Models and Technologies for Intelligent
XI. Outlook Transportation Systems. Leuven, Belgium (2011)
[14] Horni, A., Nagel, K., Axhausen, K. (eds.): Multi-Agent Trans-
This paper showed that SUMO already provides a port Simulation MATSim. Ubiquity Press, London (2016).
large framework with helpful tools for the generation, DOI 10.5334/baw
[15] Krajzewicz, D., Erdmann, J., Behrisch, M., Bieker, L.: Recent
validation and evaluation of large traffic scenarios. But development and applications of SUMO - Simulation of Urban
there are still more features and extensions, which are MObility. International Journal On Advances in Systems and
planned for enhancing and extending SUMO. For ex- Measurements 5(3&4), 128–138 (2012)
[16] Krajzewicz, D., Hausberger, S., Wagner, P., Behrisch, M.,
ample a fictional Bologna scenario to show the SUMO Krumnow, M.: Second generation of pollutant emission models
features, which are not used in the real world scenario, for sumo. In: SUMO2014 - Second SUMO User Confer-
was prepared, but not presented here due to the page ence, Reports of the DLR-Institute of Transportation Systems
(2014). URL http://elib.dlr.de/89398/
limitation. [17] Krauss, S.: Microscopic modeling of traffic flow: Investigation
SUMO already supports a number of features related of collision free vehicle dynamics. Ph.D. thesis, Universität zu
to intermodal railway simulation such as time tables, Köln (1998)
[18] Manual, H.C.: Hcm2010. Transportation Research Board,
railway crossings, railway signals and train dynamic National Research Council, Washington, DC (2010)
models. To expand the capabilities for railway simula- [19] Rummel, J.: Replication of the hbs autobahn with sumo. In:
tion a number of extensions are planned for the future. SUMO 2017 – Towards Simulation for Autonomous Mobility,
pp. 171–178 (2017). URL http://elib.dlr.de/114523/
Among these are: [20] Segata, M., Joerer, S., Bloessl, B., Sommer, C., Dressler, F.,
• Network and time table import from public transit Lo Cigno, R.: PLEXE: A Platooning Extension for Veins.
• Improved support for bidirectional track usage In: 6th IEEE Vehicular Networking Conference (VNC 2014),
pp. 53–60. IEEE, Paderborn, Germany (2014). DOI 10.1109/
• Modeling individual rail cars and their couplings
VNC.2014.7013309
• Further dynamic models [21] Sommer, C., German, R., Dressler, F.: Bidirectionally Cou-
• Adding graphical tools for defining traffic demand pled Network and Road Traffic Simulation for Improved IVC
Analysis. IEEE Transactions on Mobile Computing 10(1), 3–
References 15 (2011). DOI 10.1109/TMC.2010.133
[22] SUMO: Demand/Activity-based Demand Generation web
[1] Bieker, L., Krajzewicz, D., Morra, A.P., Michelacci, C., Car- site. Accessed 2018 (2018)
tolano, F.: Traffic simulation for all: a real world traffic sce- [23] Todosiev, E.P.: The action point model of the driver-vehicle
nario from the city of bologna. In: SUMO 2014 (2014). URL system. Ph.D. thesis, The Ohio State University (1963)
http://elib.dlr.de/89354/ [24] Treiber, M., Kanagaraj, V.: Comparing numerical integration
[2] Brilon, W., REICHARDT, G., WEISER, F., WESTPHAL, S., schemes for time-continuous car-following models. Physica
WU, N.: Handbuch für die bemessung von straßenverkehrsan- A: Statistical Mechanics and its Applications 419, 183–195
lagen. Schlussbericht zum Forschungsauftrag FE 2, R96A (2015)
(2001) [25] Varschen, C., Wagner, P.: Mikroskopische modellierung
[3] Brockfeld, E., Kühne, R.D., Wagner, P.: Calibration and val- der personenverkehrsnachfrage auf basis von zeitverwen-
idation of microscopic traffic flow models. Transportation dungstagebüchern. In: Stadt Region Land - Heft 81 -
Research Records 1876, 62–70 (2004) Tagungsband AMUS 2006, vol. 81, pp. 63–69. Aachen, Ger-
[4] Daamen, W., Buisson, C., Hoogendoorn, S.P. (eds.): Traffic many (2006). URL http://elib.dlr.de/45058/2/SRL_81_-
Simulation and Data – Validation Methods and Applications. _Beitrag_Varschen.pdf
Taylor and Francis, CRC Press (2014). URL http://www. [26] Wegener, A., Piórkowski, M., Raya, M., Hellbrück, H., Fischer,
taylorandfrancis.com/books/details/9781482228700/ S., Hubaux, J.P.: Traci: An interface for coupling road traffic
[5] Department of Electrical and Computer Engineering, Techni- and network simulators. In: Proceedings of the 11th Commu-
cal University of Munich: Fidens - Trust between Cooperative nications and Networking Simulation Symposium, CNS ’08,
Systems in Application to VANETs web site. Accessed 2018 pp. 155–163. ACM, New York, NY, USA (2008). DOI 10.
(2018). URL https://www.ldv.ei.tum.de/forschung/fidens 1145/1400713.1400740. URL http://doi.acm.org/10.1145/
[6] Erdmann, J.: Online-kalibrierung einer mikroskopischen 1400713.1400740
verkehrssimulation. In: Proceedings of VIMOS 2012. Dresden,
Germany (2012)
[7] Fellendorf, M.: Vissim: A microscopic simulation tool to eval-
uate actuated signal control including bus priority. In: 64th
Institute of Transportation Engineers Annual Meeting, pp. 1–
9. Springer (1994)
[8] FHWA: Traffic Analysis Tools. FHWA (2018).
Https://ops.fhwa.dot.gov/trafficanalysistools/index.htm
[9] Flötteröd, G.: Traffic state estimation with multi-agent simu-
lations. Ph.D. thesis, Technical University of Berlin, Berlin,
Germany (2008)
[10] Flötteröd, G., Bierlaire, M., Nagel, K.: Bayesian demand
calibration for dynamic traffic simulations. Transportation
Science (45(4)), 541–561 (2011)
[11] Flötteröd, Y.P.: Sumo-cadyts calibration with limited data
quality. In: Proceedings of the SUMO User Conference -
Towards Simulation for Autonomous Mobility, pp. 155–162.
Berlin, Germany (2006)

2582
Authorized licensed use limited to: AMRITA VISHWA VIDYAPEETHAM AMRITA SCHOOL OF ENGINEERING. Downloaded on July 25,2024 at 20:12:57 UTC from IEEE Xplore. Restrictions apply.