This project is a 2D simulation of real world soccer.
It continues the work on gym-futbol(https://github.com/yc2454/gym-futbol).
Trained PPO2 agent on the left (red). Random action agent on the right (blue).
To install the Python module:
git clone https://github.com/zhengxingXue/gym_futbol_v1.git
cd gym_futbol_v1
pip install -e .
To try the environment with PPO2 agent (trained 1M timestep):
python main.py
The simulation is recorded and showed in the pop up window. The video file is saved in videos/ directory.
tutorial.ipynb contains example for training, to start:
jupyter notebook tutorial.ipynb
To create the futbol environment:
import gym, gym_futbol_v1
env = gym.make('futbol-v1') # 2v2 settings
env.render()To train the futbol environment with PPO2(MlpPolicy) algorithm for 100,000 time step:
python training/ppo2_mlp_policy.py --timestep 100000
The trained model and best model during the training are saved in training/logs/MlpPolicy-{time} directory.
Recording of initialized PPO2 agent vs. random opponents:
Recording of a trained PPO2 agent (1 Million time step) vs. random opponents:
