Table of contents
AbstractSim
AbstractSim is a low-fidelity, 2D soccer simulation developed by the BadgerRL lab. It is written in Python and uses StableBaselines3 and the PettingZoo Parallel API for RL training.
Installation
It is recommended to create a Python virtual environment using the provided requirements.txt file. You must use Python 3.9.
python -m venv env
source env/bin/activate
python -m pip install -r requirements.txt
Alternatively, a shell script is provided for setting up a conda environment.
Usage
Training Policies
To train a policy, run the following command:
python run.py --train --env=<env name>
| Flags (Optional) | Description |
|---|---|
--batch_size=64 | Set the batch size. |
--vec_envs=12 | Sets the number of SB3 vectorized envs. |
--wandb | Enable W&B. Must have W&B initialized to use. |
--total_timesteps=10000000 | Sets the number of training steps. |
Visualizing Policies
To render a policy, run the following command:
# policy path is relative to the top-level directory
python run.py --render --env=<env name> --policy_path=<path to policy>
If no policy is provided, the environment will be rendered with randomly sampled actions.
Exporting Policies
The export.py script can be used to convert a SB3 .zip policy into a format that is compatible with WisTex-United-system. From the top-level directory, run the following command:
# policy path is relative to the top-level directory
python export.py <path to policy>
The exported .onnx policy will be located in the exported folder. This file should then be moved into the Config/Policies/ folder in WisTex-United-system.
Development
Before creating your own environments, it is recommended to obtain a basic understanding of partially observable stochastic games (POSG) and how they are implemented by the PettingZoo Parallel API.