Integration of AutoWare AV software with the CARLA simulator
- ROS kinetic
- Autoware (tested with 1.11.0)
- object detection (especially traffic lights)
- no compliance with traffic rules (due to missing vector map)
Setup/build Autoware as described here: https://github.com/CPFL/Autoware
# Download pointcloud maps for Carla Towns
cd ~
git clone https://bitbucket.org/carla-simulator/autoware-contents
#download docker image (e.g. version 0.9.5)
docker pull carlasim/carla:<carla-version>
#extract the Carla Python API from the image
cd ~
mkdir carla-python
docker run --rm --entrypoint tar carlasim/carla:<carla-version> cC /home/carla/PythonAPI . | tar xvC ~/carla-python
The Carla Autoware Bridge is a ROS package. Therefore we create a catkin workspace (containing all relevant packages). The generic Carla ROS bridge (https://github.com/carla-simulator/ros-bridge.git) is included as GIT submodule and has to be initialized ("git submodule update --init") and updated ("git submodule update").
cd ~
git lfs clone https://github.com/carla-simulator/carla-autoware.git
cd carla-autoware
git submodule update --init
cd catkin_ws
catkin_init_workspace src/
cd src
ln -s <path-to-autoware>/ros/src/msgs/autoware_msgs
cd ..
catkin_make
To run Autoware within Carla please use the following execution order:
- Carla Server
- Autoware (including carla-ros-bridge and additional nodes)
You need two terminals:
#Terminal 1
#execute Carla
#For details, please refer to the CARLA documentation
nvidia-docker run -p 2000-2001:2000-2001 -it --rm carlasim/carla:<carla-version> ./CarlaUE4.sh /Game/Carla/Maps/Town01 -benchmark -carla-server -fps=20
#Terminal 2
export CARLA_AUTOWARE_ROOT=~/carla-autoware/autoware_launch
export CARLA_MAPS_PATH=~/autoware-contents/maps
source <path-to-autoware>/ros/install/setup.bash
source $CARLA_AUTOWARE_ROOT/../catkin_ws/devel/setup.bash
export PYTHONPATH=$PYTHONPATH:~/carla-python/carla/dist/carla-<carla-version>-py2.7-linux-x86_64.egg:~/carla-python/carla/
roslaunch $CARLA_AUTOWARE_ROOT/devel.launch
You can run Autoware and Carla on different machines. To let the carla autoware bridge connect to a remote Carla Server, execute roslaunch with the following parameters
roslaunch host:=<hostname> port:=<port number> $CARLA_AUTOWARE_ROOT/devel.launch
The setup of the sensors is defined within sensors.json.
carla_ego_vehicle reads the file and spawn the ego vehicle and the sensors.
When starting the carla_autoware_bridge a random spawn point and a fixed goal is used to calculate the route.
To override this, you can use RVIZ.
- selecting a Pose with '2D Pose Estimate' will delete the current ego_vehicle and respawn it at the specified position.
- selecting a Pose with '2D Nav Goal' will set a new goal within
carla_waypoint_publisher.
Press B to be able to steer the ego vehicle within ROS manual control.
Internally, this is done by stopping the conversion from the Autoware control message to AckermannDrive within the node vehiclecmd_to_ackermanndrive. The relevant ros-topic is /vehicle_control_manual_override.
The bridge contains three Carla Clients.
- ROS Bridge - Monitors existing actors in Carla, publishes changes on ROS Topics (e.g. new sensor data)
- Ego Vehicle - Instantiation of the ego vehicle with its sensor setup.
- Waypoint Calculation - Uses the Carla Python API to calculate a route.
