This ROS package aims at providing a simple ROS bridge for CARLA simulator.

This version requires CARLA 0.9.9

• Provide Sensor Data (Lidar, Cameras (depth, segmentation, rgb), GNSS, Radar, IMU)
• Provide Object Data (Transforms (via tf), Traffic light status, Visualization markers, Collision, Lane invasion)
• Control AD Agents (Steer/Throttle/Brake)
• Control CARLA (Support synchronous mode, Play/pause simulation, Set simulation parameters)

Beside the bridging functionality, there are many more features provided in separate packages.

For a quick overview, after following the Setup section, please run the CARLA AD Demo. It provides a ready-to-use demonstrator of many of the features.

First add the apt repository:

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 81061A1A042F527D &&
sudo add-apt-repository "deb [arch=amd64 trusted=yes] http://dist.carla.org/carla-ros-bridge-melodic/ bionic main"

sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 9BE2A0CDC0161D6C &&
sudo add-apt-repository "deb [arch=amd64 trusted=yes] http://dist.carla.org/carla-ros-bridge-kinetic xenial main"

Then simply install the ROS bridge:

sudo apt update &&
sudo apt install carla-ros-bridge-<melodic or kinetic>

This will install carla-ros-bridge- in /opt/carla-ros-bridge

Create a catkin workspace and install carla_ros_bridge package

#setup folder structure
mkdir -p ~/carla-ros-bridge/catkin_ws/src
cd ~/carla-ros-bridge
git clone https://github.com/carla-simulator/ros-bridge.git
cd ros-bridge
git submodule update --init
cd ../catkin_ws/src
ln -s ../../ros-bridge
source /opt/ros/kinetic/setup.bash
cd ..

#install required ros-dependencies
rosdep update
rosdep install --from-paths src --ignore-src -r

#build
catkin_make

For more information about configuring a ROS environment see http://wiki.ros.org/ROS/Tutorials/InstallingandConfiguringROSEnvironment

First run the simulator (see carla documentation: http://carla.readthedocs.io/en/latest/)

# run carla in background
SDL_VIDEODRIVER=offscreen ./CarlaUE4.sh -opengl

Wait a few seconds

export PYTHONPATH=$PYTHONPATH:<path-to-carla>/PythonAPI/carla/dist/carla-<carla_version_and_arch>.egg

source /opt/carla-ros-bridge/<melodic or kinetic>/setup.bash

source ~/carla-ros-bridge/catkin_ws/devel/setup.bash

Start the ros bridge (choose one option):

# Option 1: start the ros bridge
roslaunch carla_ros_bridge carla_ros_bridge.launch

# Option 2: start the ros bridge together with an example ego vehicle
roslaunch carla_ros_bridge carla_ros_bridge_with_example_ego_vehicle.launch

You can modify the ros bridge configuration by editing carla_ros_bridge/config/settings.yaml.

If the rolename is within the list specified by ROS parameter /carla/ego_vehicle/rolename, the client is interpreted as an controllable ego vehicle and all relevant ROS topics are created.

Certain parameters can be set within the launch file carla_ros_bridge.launch.

The bridge is able to load a CARLA map by setting the launch-file parameter town. Either specify an available CARLA Town (e.g. 'Town01') or a OpenDRIVE file (with ending '.xodr').

In default mode (synchronous_mode: false) data is published:

• on every world.on_tick() callback
• on every sensor.listen() callback

CAUTION: In synchronous mode, only the ros-bridge is allowed to tick. Other CARLA clients must passively wait.

In synchronous mode (synchronous_mode: true), the bridge waits for all sensor data that is expected within the current frame. This might slow down the overall simulation but ensures reproducible results.

Additionally you might set synchronous_mode_wait_for_vehicle_control_command to true to wait for a vehicle control command before executing the next tick.

It is possible to control the simulation execution:

• Pause/Play
• Execute single step

The following topic allows to control the stepping.

A CARLA Control rqt plugin is available to publish to the topic.

The ego vehicle sensors are provided via topics with prefix /carla/ego_vehicle/<sensor_topic>

Currently the following sensors are supported:

Radar data can be visualized on rviz using ainstein_radar_rviz_plugins.

There are two modes to control the vehicle.

1. Normal Mode (reading commands from /carla/<ROLE NAME>/vehicle_control_cmd)
2. Manual Mode (reading commands from /carla/<ROLE NAME>/vehicle_control_cmd_manual)

This allows to manually override a Vehicle Control Commands published by a software stack. You can toggle between the two modes by publishing to /carla/<ROLE NAME>/vehicle_control_manual_override.

carla_manual_control makes use of this feature.

For testing purposes, you can stear the ego vehicle from the commandline by publishing to the topic /carla/<ROLE NAME>/vehicle_control_cmd.

Examples for a ego vehicle with role_name 'ego_vehicle':

Max forward throttle:

 rostopic pub /carla/ego_vehicle/vehicle_control_cmd carla_msgs/CarlaEgoVehicleControl "{throttle: 1.0, steer: 0.0}" -r 10

Max forward throttle with max steering to the right:

 rostopic pub /carla/ego_vehicle/vehicle_control_cmd carla_msgs/CarlaEgoVehicleControl "{throttle: 1.0, steer: 1.0}" -r 10

The current status of the vehicle can be received via topic /carla/<ROLE NAME>/vehicle_status. Static information about the vehicle can be received via /carla/<ROLE NAME>/vehicle_info

CAUTION: This control method does not respect the vehicle constraints. It allows movements impossible in the real world, like flying or rotating.

You can also control the vehicle via publishing linear and angular velocity within a Twist datatype.

Currently this method applies the complete linear vector, but only the yaw from angular vector.

In certain cases, the Carla Control Command is not ideal to connect to an AD stack. Therefore a ROS-based node carla_ackermann_control is provided which reads AckermannDrive messages. You can find further documentation here.

The tf data is published for all traffic participants and sensors.

The child_frame_id corresponds with the CARLA actor id. If a role name is specified, an additional (static) transform with role name as child_frame_id is published.

Sensors publish the transform, when the measurement is done. The child_frame_id corresponds with the prefix of the sensor topics.

It is possible to draw markers in CARLA.

Caution: Markers might affect the data published by sensors.

The following markers are supported in 'map'-frame:

• Arrow (specified by two points)
• Points
• Cube
• Line Strip

You're missing Carla Python. Please execute:

export PYTHONPATH=$PYTHONPATH:<path/to/carla/>/PythonAPI/carla/dist/<your_egg_file>

Please note that you have to put in the complete path to the egg-file including the egg-file itself. Please use the one, that is supported by your Python version. Depending on the type of CARLA (pre-build, or build from source), the egg files are typically located either directly in the PythonAPI folder or in PythonAPI/dist.

Check the installation is successfull by trying to import carla from python:

python -c 'import carla;print("Success")'

You should see the Success message without any errors.