repository·master·Indexed 25 days ago
https://github.com/google-research/footballAn RL environment based on the open-source Gameplay Football game, designed for reinforcement learning research. It provides a GYM-style API for creating environments, training agents using PPO, and managing game states via get_state and set_state. The documentation covers installation across Linux, macOS, and Windows, manual play keyboard mappings, and the use of the gfootball package.
To enable rendering of the game within a Docker container on Linux, you must grant the container access to your X server. Run the following command (this must be done after every reboot, or added to /etc/profile):
xhost +"local:docker@"gfootball/scenarios/ directory. It is recommended to examine existing scenario files in that directory as templates for your implementation.env/players directory. No other changes to the repository are required. You can refer to existing player implementations in that directory for reference.To build the environment on Linux, install the required system packages via apt-get, including cmake, build-essential, various sdl2 libraries, libboost-all-dev, and python3-pip.
Steps:
sudo apt-get install.pip, setuptools, and wheel, and install psutil.python3 -m pip install . to build the game environment.sudo apt-get install git cmake build-essential libgl1-mesa-dev libsdl2-dev \libsdl2-image-dev libsdl2-ttf-dev libsdl2-gfx-dev libboost-all-dev \libdirectfb-dev libst-dev mesa-utils xvfb x11vnc python3-pip
python3 -m pip install --upgrade pip setuptools wheel
python3 -m pip install psutil
git clone https://github.com/google-research/football.git
cd football
python3 -m venv football-env
source football-env/bin/activate
python3 -m pip install --upgrade pip setuptools wheel
python3 -m pip install psutil
python3 -m pip install .To build the environment on macOS (Intel or Apple Silicon), use brew to install system dependencies including git, python3, cmake, sdl2, sdl2_image, sdl2_ttf, sdl2_gfx, boost, and boost-python3.
Important Note on Python: It is highly recommended to use the Python distribution shipped with brew because boost-python3 is compiled against it. If your default python3 is not from Homebrew, create a virtual environment using $(brew --prefix python3)/bin/python3 -m venv football-env.
Steps:
brew.venv or conda).pip, setuptools, and wheel, and install psutil.python3 -m pip install . to build the game environment.brew install git python3 cmake sdl2 sdl2_image sdl2_ttf sdl2_gfx boost boost-python3
python3 -m pip install --upgrade pip setuptools wheel
python3 -m pip install psutil
git clone https://github.com/google-research/football.git
cd football
python3 -m venv football-env
source football-env/bin/activate
python3 -m pip install --upgrade pip setuptools wheel
python3 -m pip install psutil
python3 -m pip install .To perform Imitation Learning in Google Research Football, you can use game replays from the Kaggle competition. You can acquire this data in two ways:
To install the package in editable (development) mode, use the -e flag with pip. This allows you to edit Python source files in-place, and changes will be reflected the next time an interpreter process is started without requiring a reinstallation.
python3 -m pip install -e .Before building a Docker image, you must first clone the repository locally to access the Dockerfiles.
git clone https://github.com/google-research/football.git
cd footballTo train a policy that controls multiple players, follow these two steps:
number_of_players_agent_controls to the create_environment function to define how many players the agent should control..step() function, pass an array of actions (one action per player) instead of a single action.It is the responsibility of the caller to unpack or post-process these actions as needed. A reference implementation is available in examples/run_multiagent_rllib.py.
To run Reinforcement Learning training (specifically PPO), you must first install additional dependencies:
python3 -m pip install --upgrade pip setuptools wheelpython3 -m pip install tensorflow==1.15.* (or tensorflow-gpu==1.15.* for GPU support)python3 -m pip install dm-sonnet==1.* psutilpython3 -m pip install git+https://github.com/openai/baselines.git@masterRun PPO on academy_empty_goal scenario:
python3 -m gfootball.examples.run_ppo2 --level=academy_empty_goal_closeRun PPO on academy_pass_and_shoot_with_keeper scenario:
python3 -m gfootball.examples.run_ppo2 --level=academy_pass_and_shoot_with_keeperRun with rendering and episode saving:
python3 -m gfootball.examples.run_ppo2 --dump_full_episodes=True --render=TrueYou can install Google Research Football on Linux, macOS, or Windows. The recommended way for Linux users is to use Docker to avoid package version conflicts. For direct installation, follow the platform-specific steps below.
Install system dependencies:
sudo apt-get install git cmake build-essential libgl1-mesa-dev libsdl2-dev \
libsdl2-image-dev libsdl2-ttf-dev libsdl2-gfx-dev libboost-all-dev \
libdirectfb-dev libst-dev mesa-utils xvfb x11vnc python3-pip
python3 -m pip install --upgrade pip setuptools psutil wheelInstall Homebrew, then install dependencies:
brew install git python3 cmake sdl2 sdl2_image sdl2_ttf sdl2_gfx boost boost-python3
python3 -m pip install --upgrade pip setuptools psutil wheelInstall Git and Python 3. Update pip using python (not python3) in the Command Line:
python -m pip install --upgrade pip setuptools psutil wheelYou can build the Docker image using two different base configurations depending on whether you require GPU support for TensorFlow training.
### Tensorflow without GPU-training support version
```bash
docker build --build-arg DOCKER_BASE=ubuntu:20.04 . -t gfootballdocker build --build-arg DOCKER_BASE=tensorflow/tensorflow:1.15.2-gpu-py3 . -t gfootball