The main way to interact with a running simulation is TraCI which gives the complete flexibility of doing cross-platform, cross-language, and networked interaction with sumo acting as a server. One major drawback is the communication overhead due to the protocol and the socket communication. To have a more efficient coupling without the need for socket communication, the TraCI API is provided as a C++ library with the following properties:

  • C++ interface based on static functions and a few simple wrapper classes for results which can be linked directly to the client code
  • Function signatures similar to TraCI
  • Pre-built language bindings for Java and Python (using SWIG)
  • Support for other programming languages via SWIG


The following things currently do not work (or work differently than with the TraCI Python client):

  • running with sumo-gui does not work on Windows and is still highly experimental on other platforms
  • subscriptions that require additional arguments (except for vehicle.getLeader)
  • stricter type checking
    • the TraCI client sometimes accepts any iterable object where Libsumo wants a list
    • TraCI client may accept any object where Libsumo needs a boolean value
    • TraCI automatically converts every parameter into a string if a string is needed, Libsumo does not
  • using traci.init or traci.connect is not possible (you always need to use traci.start / libsumo.start)
  • with traci every TraCIException will generate a message on stderr, Libsumo does not generate this message
  • libsumo by itself cannot be used to connect multiple clients to the simulation (though connecting normal TraCI clients to a libsumo instance is possible)
  • running parallel instances of libsumo requires the multiprocessing module (in python)

To avoid the limitations with respect to GUI, multi-clients support, you can also use libraci. This is a C++ traci client library which is fully API-compatible with libsumo.

Building and Installing it#

The binary windows release already contains the readily compiled libsumo for C++ and Java. For Python you can install it via pip install libsumo. Only if your platform or language is not supported follow the steps below.

It currently requires cmake and swig being installed together with the developer packages for Python (and Java if needed), for Windows see Windows CMake. You need to (re-)compile sumo yourself under Windows following the remarks above, under Linux see Installing/Linux_Build (it is probably just a matter of calling cmake and make again if you previously did a build without swig). For the python bindings you will get a and a (or .pyd on Windows). If you place them somewhere on your python path you should be able to use them as described below. If you want to enable the experimental C# support, make sure that you have ENABLE_CS_BINDINGS set in your cmake configuration.


Make sure to add "/your/path/to/sumo/tools" to the PYTHONPATH environment variable.

Using libsumo#

If you want to use the (experimental) GUI then you need to have sumo-gui in your start command instead of sumo or define the environment variable LIBSUMO_GUI.


Make sure you have libsumo installed (pip install libsumo).

import libsumo
libsumo.start(["sumo", "-c", "test.sumocfg"])

Existing traci scripts can be reused (subject to the limitations mentioned above) by calling

import libsumo as traci

In case you have a lot of scripts you can also set the environment variable LIBSUMO_AS_TRACI to a non empty value which will trigger the import as above.


Please note the extra #define for enabling GUI code which is not needed if you do not or cannot use the GUI (Windows).

Example Code (test.cpp)#

#include <iostream>
#define HAVE_LIBSUMOGUI  // if you are on Windows or have libsumo compiled yourself without GUI you should remove this line
#include <libsumo/libsumo.h>

using namespace libsumo;

int main(int argc, char* argv[]) {
    Simulation::start({"sumo", "-c", "test.sumocfg"});
    for (int i = 0; i < 5; i++) {

compiling on Linux (make sure SUMO_HOME is set and sumo has been built)#

g++ -o test -std=c++11 -I$SUMO_HOME/src test.cpp -L$SUMO_HOME/bin -lsumocpp

running on Linux#



You might want to use the available Maven package.

Example Code (

import org.eclipse.sumo.libsumo.Simulation;
import org.eclipse.sumo.libsumo.StringVector;

public class Test {
    public static void main(String[] args) {
        Simulation.start(new StringVector(new String[] {"sumo", "-c", "test.sumocfg"}));
        for (int i = 0; i < 5; i++) {

Please note that starting with SUMO 1.16.0 it seems to be necessary to preload more libraries on Windows, see

compiling on Linux (make sure SUMO_HOME is set and sumo has been built)#

javac -cp $SUMO_HOME/bin/libsumo-1.8.0-SNAPSHOT.jar

running on Linux#

java -Djava.library.path=$SUMO_HOME/bin -cp $SUMO_HOME/bin/libsumo-1.8.0-SNAPSHOT.jar:. Test

casting subscription results#

Please be aware that casting subscription results is not straightforward with Java. You have to use the cast function as below. If the cast is not successful it will not throw an exception but return a null pointer.

TraCIResults ssRes = Simulation.getSubscriptionResults();
for (Map.Entry<Integer, TraCIResult> entry : ssRes.entrySet()) {
    TraCIResult sR = entry.getValue();
    TraCIStringList vehIDs = TraCIStringList.cast(sR);
    for (String vehID : vehIDs.getValue()) {
        System.out.println("Subscription Departed vehicles: " + vehID);


Please install the Python package. You can then use all commands inside your Matlab scripts just as in Python by adding the py. prefix.

py.libsumo.start(["sumo", "-c", "test.sumocfg"])