cosynus-lix/f1tenth_quickstart_ros2

A quickstart for using f1tenth_gym_ros to launch a virtual F1tenth compeition.

This repository is for providing a quick start on using the f1tenth_gym_ros to launch a F1tenth virtual competition using ROS2. For a ROS1 version quick start, please refer to f1tenth_quickstart version ROS1.

Quick demonstration by using example controller

Preparation:

The simulator can be built in 3 different ways:

• natively installed as a ROS package
• built as a Docker image (without NVIDIA GPU)
• built as a Docker image (with NVIDIA GPU)

For purpose of simplicity, we will assume the 3rd option is used. For the other 2 options, please refer to the README file of f1tenth_gym_ros. Similarly, controllers can be built either natively or within a Docker container. In this tutorial, we set up controllers in a Docker container.

Build the Docker image for the simulator:

$ git clone [email protected]:f1tenth/f1tenth_gym_ros.git
$ cd f1tenth_gym_ros
$ docker build -t f1tenth_gym_ros -f Dockerfile .

Single vehicle mode:

Step 1 - Launch the simulator

Launch the Docker image for the simulator (f1tenth_gym_ros). You might need to take a few seconds to wait for RVIZ to fully function.

$ rocker --nvidia --x11 --volume .:/sim_ws/src/f1tenth_gym_ros -- f1tenth_gym_ros
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 launch f1tenth_gym_ros gym_bridge_launch.py

Step 2 - Launch the controller

In another terminal, clone this repository and build a Docker image (f1tenth_quickstart_ros2) for controllers, then launch it:

$ git clone [email protected]:cosynus-lix/f1tenth_quickstart_ros2
$ cd f1tenth_quickstart_ros2
$ docker build -t f1tenth_quickstart_ros2 -f Dockerfile .
$ docker run -it --rm --volume $(pwd)/src/:/ctrl_ws/src/ f1tenth_quickstart_ros2

Note that the folder src is mounted in the container using --volume $(pwd)/src/:/ctrl_ws/src/. Any changes to the /ctrl_ws/src/ folder within the Docker environment will be automatically saved.

In the same Docker image, run example controller wall_following:

$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 run wall_following sing_vehicle_mode

You can also use keyboard as the controller to move the ego vehicle:

$ sudo apt-get install ros-foxy-teleop-twist-keyboard
$ ros2 run teleop_twist_keyboard teleop_twist_keyboard

Step 3 - Pose reset

To reset vehicle's pose, use 2D Pose Estimate in RVIZ.

Head-to-head mode, ego + opp:

Step 1 - Modify the configuration file, recompile simulator package and launch it:

$ rocker --nvidia --x11 --volume .:/sim_ws/src/f1tenth_gym_ros -- f1tenth_gym_ros
$ vim src/f1tenth_gym_ros/config/sim.yaml (change 'num_agent:' to 2)
$ colcon build
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 launch f1tenth_gym_ros gym_bridge_launch.py

Step 2 - Launch the controller

In another terminal, launch the docker image and run the example controller:

$ docker run -it --rm --volume $(pwd)/src/:/ctrl_ws/src/ f1tenth_quickstart_ros2
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 run wall_following head_to_head_mode

If the opponent vehicle does not appear correctly in RVIZ, try to add it manually: click Add at the right bottom of RVIZ window -> select RobotModel -> change description topic to /opp_robot_description.

Step 3 - Add/remove obstacles and pose reset

To reset vehicles' pose, use 2D Pose Estimate in RVIZ for the ego vehicle and 2D Goal Pose for the opponent vehicle.

Change the track map of simulator

Step 1

Copy map files (e.g. berlin.png and berlin.yaml in the map library) to the folder maps/ in f1tenth_gym_ros.

Step 2

Modify the map configuration, there are 2 options:

option 1 (temporary modification) - Modify the configuration file inside Docker image:

$ rocker --nvidia --x11 --volume .:/sim_ws/src/f1tenth_gym_ros -- f1tenth_gym_ros
$ vim src/f1tenth_gym_ros/config/sim.yaml (change 'map_path:' to the new map name)
$ colcon build
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 launch f1tenth_gym_ros gym_bridge_launch.py

option 2 (permanent modification) - Modify the configuration before compiling Docker image:

$ (in host PC terminal) cd f1tenth_gym_ros
$ vim config/sim.yaml (change 'map_path:' to the new map name)
$ docker build -t f1tenth_gym_ros -f Dockerfile .

$ rocker --nvidia --x11 --volume .:/sim_ws/src/f1tenth_gym_ros -- f1tenth_gym_ros
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 launch f1tenth_gym_ros gym_bridge_launch.py

You can also DIY a map (design a new one or add obstacles on an old one) by drawing pixels on map image (png/pgm/... files) under any picture editor! Just remember: white for free space and black for obstacles. Moreover, in the .yaml file, choose a small value for resolution and set the third coordinate of origin to a null value (e.g. 0).

It should just work perfectly! For more details on changing maps, we refer to the description here.

Write the controller code by yourself!

What to change?

Using Python

1. in src/wall_following/: modify the file single_vehicle.py / head_to_head.py, or write an independent file with a similar structure.
2. the main point is to get environment information from the topic /scan(Message type - sensor_msgs/LaserScan, more details) and(or) /odom(Message type - nav_msgs/Odometry, more details), calculate the controll command and send to the topic /drive (Message type - ackermann_msgs/AckermannDrive, more details).
3. file setup.py: in case that you write a node in a new python file e.g. new_node.py, you need to indicate the entry for launching this node:

    entry_points={
        'console_scripts': [
            'sing_vehicle_mode = wall_following.single_vehicle:main',
            'head_to_head_mode = wall_following.head_to_head:main',
            'new_node_mode' = wall_following.new_node:main
        ],
    }

4. file package.xml: in case that you use new dependancies (such as ackermann_msgs etc), add them in this file.

Using C++

you can also write the controller in C++, refer to ROS2 Tutorial.

What is missing in the example code?

1. in the example code, we only use the information of Lidar via the topic /scan but not /odom. You can use /odom to obtain the ego agent's odometry for your own algorithm!
2. for head-to-head mode, we suppose that the ego and opp vehicle use the same controller and put their controller inside a same file. Of course you can use different controllers for each! In practice, it is better to write their controllers in different files.