/f1tenth_gym_ros-agent

Containerized ROS communication bridge for F1TENTH gym environment.

Primary LanguagePythonMIT LicenseMIT

F1TENTH gym environment ROS2 communication bridge

This is a containerized ROS communication bridge for the F1TENTH gym environment that turns it into a simulation in ROS2.

Installation

Supported System:

  • Ubuntu (tested on 20.04) native with ROS 2
  • Ubuntu (tested on 20.04) with an NVIDIA gpu and nvidia-docker2 support
  • Windows 10, macOS, and Ubuntu without an NVIDIA gpu (using noVNC)

This installation guide will be split into instruction for installing the ROS 2 package natively, and for systems with or without an NVIDIA gpu in Docker containers.

Native on Ubuntu 20.04

Install the following dependencies:

  • ROS 2 Follow the instructions here to install ROS 2 Foxy.
  • F1TENTH Gym 
    git clone https://github.com/f1tenth/f1tenth_gym
cd f1tenth_gym && pip3 install -e .

Installing the simulation:

  • Create a workspace: cd $HOME && mkdir -p sim_ws/src
  • Clone the repo into the workspace: 
    cd $HOME/sim_ws/src
git clone https://github.com/f1tenth/f1tenth_gym_ros
  • Update correct parameter for path to map file: Go to sim.yaml https://github.com/f1tenth/f1tenth_gym_ros/blob/main/config/sim.yaml in your cloned repo, change the map_path parameter to point to the correct location. It should be '<your_home_dir>/sim_ws/src/f1tenth_gym_ros/maps/levine'
  • Install dependencies with rosdep: 
    source /opt/ros/foxy/setup.bash
cd ..
rosdep install -i --from-path src --rosdistro foxy -y
  • Build the workspace: colcon build

With an NVIDIA gpu:

Install the following dependencies:

  • Docker Follow the instructions here to install Docker. A short tutorial can be found here if you're not familiar with Docker. If you followed the post-installation steps you won't have to prepend your docker and docker-compose commands with sudo.
  • nvidia-docker2, follow the instructions here if you have a support GPU. It is also possible to use Intel integrated graphics to forward the display, see details instructions from the Rocker repo. If you are on windows with an NVIDIA GPU, you'll have to use WSL (Windows Subsystem for Linux). Please refer to the guide here, here, and here.
  • rocker https://github.com/osrf/rocker. This is a tool developed by OSRF to run Docker images with local support injected. We use it for GUI forwarding. If you're on Windows, WSL should also support this.

Installing the simulation:

  1. Clone this repo
  2. Build the docker image by:
$ cd f1tenth_gym_ros
$ docker build -t f1tenth_gym_ros -f Dockerfile .
  1. To run the containerized environment, start a docker container by running the following. (example showned here with nvidia-docker support). By running this, the current directory that you're in (should be f1tenth_gym_ros) is mounted in the container at /sim_ws/src/f1tenth_gym_ros. Which means that the changes you make in the repo on the host system will also reflect in the container.
$ rocker --nvidia --x11 --volume .:/sim_ws/src/f1tenth_gym_ros -- f1tenth_gym_ros

Without an NVIDIA gpu:

Install the following dependencies:

If your system does not support nvidia-docker2, noVNC will have to be used to forward the display.

  • Again you'll need Docker. Follow the instruction from above.
  • Additionally you'll need docker-compose. Follow the instruction here to install docker-compose.

Installing the simulation:

  1. Clone this repo
  2. Bringup the novnc container and the sim container with docker-compose:
docker-compose up
  1. In a separate terminal, run the following, and you'll have the a bash session in the simulation container. tmux is available for convenience.
docker exec -it f1tenth_gym_ros-sim-1 /bin/bash
  1. In your browser, navigate to http://localhost:8080/vnc.html, you should see the noVNC logo with the connect button. Click the connect button to connect to the session.

Launching the Simulation

  1. tmux is included in the contianer, so you can create multiple bash sessions in the same terminal.
  2. To launch the simulation, make sure you source both the ROS2 setup script and the local workspace setup script. Run the following in the bash session from the container:
$ source /opt/ros/foxy/setup.bash
$ source install/local_setup.bash
$ ros2 launch f1tenth_gym_ros gym_bridge_launch.py

A rviz window should pop up showing the simulation either on your host system or in the browser window depending on the display forwarding you chose.

You can then run another node by creating another bash session in tmux.

Configuring the simulation

  • The configuration file for the simulation is at f1tenth_gym_ros/config/sim.yaml.
  • Topic names and namespaces can be configured but is recommended to leave uncahnged.
  • The map can be changed via the map_path parameter. You'll have to use the full path to the map file in the container. The map follows the ROS convention. It is assumed that the image file and the yaml file for the map are in the same directory with the same name. See the note below about mounting a volume to see where to put your map file.
  • The num_agent parameter can be changed to either 1 or 2 for single or two agent racing.
  • The ego and opponent starting pose can also be changed via parameters, these are in the global map coordinate frame.

The entire directory of the repo is mounted to a workspace /sim_ws/src as a package. All changes made in the repo on the host system will also reflect in the container. After changing the configuration, run colcon build again in the container workspace to make sure the changes are reflected.

Topics published by the simulation

In single agent:

/scan: The ego agent's laser scan

/ego_racecar/odom: The ego agent's odometry

/map: The map of the environment

A tf tree is also maintained.

In two agents:

In addition to the topics available in the single agent scenario, these topics are also available:

/opp_scan: The opponent agent's laser scan

/ego_racecar/opp_odom: The opponent agent's odometry for the ego agent's planner

/opp_racecar/odom: The opponent agents' odometry

/opp_racecar/opp_odom: The ego agent's odometry for the opponent agent's planner

Topics subscribed by the simulation

In single agent:

/drive: The ego agent's drive command via AckermannDriveStamped messages

/initalpose: This is the topic for resetting the ego's pose via RViz's 2D Pose Estimate tool. Do NOT publish directly to this topic unless you know what you're doing.

TODO: kb teleop topics

In two agents:

In addition to all topics in the single agent scenario, these topics are also available:

/opp_drive: The opponent agent's drive command via AckermannDriveStamped messages. Note that you'll need to publish to both the ego's drive topic and the opponent's drive topic for the cars to move when using 2 agents.

/goal_pose: This is the topic for resetting the opponent agent's pose via RViz's 2D Goal Pose tool. Do NOT publish directly to this topic unless you know what you're doing.

Keyboard Teleop

The keyboard teleop node from teleop_twist_keyboard is also installed as part of the simulation's dependency. To enable keyboard teleop, set kb_teleop to True in sim.yaml. After launching the simulation, in another terminal, run:

ros2 run teleop_twist_keyboard teleop_twist_keyboard

Then, press i to move forward, u and o to move forward and turn, , to move backwards, m and . to move backwards and turn, and k to stop in the terminal window running the teleop node.

Developing and creating your own agent in ROS 2

There are multiple ways to launch your own agent to control the vehicles.

  • The first one is creating a new package for your agent in the /sim_ws workspace inside the sim container. After launch the simulation, launch the agent node in another bash session while the sim is running.
  • The second one is to create a new ROS 2 container for you agent node. Then create your own package and nodes inside. Launch the sim container and the agent container both. With default networking configurations for docker, the behavior is to put The two containers on the same network, and they should be able to discover and talk to each other on different topics. If you're using noVNC, create a new service in docker-compose.yml for your agent node. You'll also have to put your container on the same network as the sim and novnc containers.