weibinkou / carla_ros_slam

RGBD-SLAM , GMAPPING, HECTORMAPPING, OCTOGRAPH , KARTOSLAM , GOOGLE CARTOGRAPHER SLAM implementation in CARLA SIMULATOR

CarlaRosBridge

System information on which the above project is tested

• Ubuntu 18.04 64bit
• Nvidia 1050TI (435- Vulkan enabled)
• i7 7th Gen
• RAM 24GB
• ROS melodic

How to load the project

• Clone the git repo by issueing the below command

   `git clone --recurse-submodules https://gitlab.kpit.com/kankans/carlarosbridge.git`.  
  

• Browse to the cloned folder cd carlarosbridge.

• Compile the project by catkin_make.

• Source the project source devel/setup.sh , Everytime when a new terminal is opened the carlarosbridge environment is sourced.

• Run ROS roscore.

• Run carla ./CarlaEU4.sh -windowed -ResX=400 -ResY=320 wait for the carla window to load.

• If you want to launch the carla client with RVIZ then roslauch carlabg CarlaVehicleWithRviz.

• If you want to launch the carla client without RVIZ then roslauch carlabg CarlaVehicleWithoutRviz.
  

  • You can edit the lauch file to play with the carla environment.
    • You can change the spawn point of the vehicle by changing the X,Y,Z location in the above line <arg name="spawn_point" default="73.193413,-136.704315,9.837398,-0.647311,-178.772690,0.000000"/> .
    • If you want to change the vehicle model edit the default model in the <arg name="vehicle_filter" default='vehicle.tesla.model3'/>.
    • If you want to change the weather of the simulation then change(Cloudiness, Precipitation , SunAltitudeAngle) <arg name="weather" default="90.0,0.0,1.0"/>
    • If you want to change the other actor models then change the below parameters <arg name="NumberofVehicles" default="1"/> <arg name="NumberofPedestrians" default="1"/> <arg name="SafeTravel" default="False"/>.
  • You can change the sensor config in sensors.json (carlarosbridge/carlabg/config/sensors.json).

• The Navigation is broken and does not work properly, however the navigation can be launched by the following command

  • roslaunch carlaconfig amcl_demo.launch <-- this also launches RVIZ with the map

Standard Slam Implementation

• Gmapping
  • Open Terminal and run roscore.
  • Open the carla folder in terminal and run the CarlaSimulator CARLAUE4.sh -ResX=320 -ResY=480.
  • Start the carla-ros bridge node , Open another terminal and run roslaunch carlabg CarlaVehicleWithoutRviz.launch command.
  • Start the Gmapping node roslaunch carlabg Gmapping_Carla.launch.
  • If you want to view the live mapping , lauch the rviz node rosrun rviz rviz. Either you can load the Gmapping-rviz config from /carla/carlabg/config folder or enable a map view under frame map.
  • Save the map by running map_server rosrun map_server map_saver.
• Hectormapping
  • Launch the carla-ros bridge same as Gmapping , and then launch the hectormapping node by running roslaunch carlabg Hectormapping_Carla.launch.
  • Save the map by running map_server rosrun map_server map_saver.
• Octomap
  • Launch octomap by roslaunch carlabg Octomapping_Carla.launch.
  • The octomap can be viewed using the rviz octomap plugin , load the Octomap-rviz config from /carla/carlabg/config.
• RTAB mapping
  • Launch RTAB map by roslaunch carlabg RtabMapping_Carla.launch this node will automatically launch the RTABmap viewer for map visualization.

DNN based Visual SLAM Implementation

• Launch the Carla Simulator 0.9.7.
• Run the Visual SLAM detector by running carlarosbridge/Slam_Tes/Visual_SLAM.py file. It runs the detection offline edit the config here Edit Source Config, the weights and config can be edited here , the detector speed can be increased by resizing the image refer here, if you dont want to save the output comment out this line.
• Run the Sensor Fusion Algorithm by opening carlarosbridge/Slam_Test/SensorFusionCarla.py file, if you want to edit the sensor noise change this line , the sensor deviation is in g , by default the script will automatically record the data in KalmanTest_XXX.csv.

Note

• The slam node did not work for me unless the messages are transmitted slowly.Hence while performing manual maneuver bag the messages by the following command
  rosbag record /tf /camera/scan /carla/ego_vehicle/lidar/lidar1/point_cloud
• Launch Gmapping node by running following commad
  roslaunch carlabg Gmapping_Carla.launch
• After the messages are logged we can playback the messages by following command
  rosbag play -r 0.01 --clock "Your recorded file name"
• This will take significant amount of time to build the map (better than the below error message)
  (Error Message to be included ;-P)
• Once you are done with the playback , save the message by following command
  rosrun map_server map_saver -f "Your File Name"
• All recorded data is stored in /src/data/
• Hector and GMAPPING is tested against the two scenarios, Results are stored in the data folder ExperintData.tar.xz