Commonly-used Multi-agent Environments Installation, Convenient Wrappers, and VectorEnv Implementation with PettingZoo (and Gymnasium) Compatibility.

Environment Supports

• Google Research Football
• Multi-agent MuJoCo
• StarCraft Multi-Agent Challenge (SMAC and SMACv2)

Installation

The scripts install_grf.sh, install_mamujoco.sh and install_smac.sh include detailed installation guides.

PyPi from sources

pip install git+https://github.com/xihuai18/Common-Cooperative-Multi-Agent-Environments.git

Install from GitHub sources

git clone https://github.com/xihuai18/Common-Cooperative-Multi-Agent-Environments.git
cd Common-Cooperative-Multi-Agent-Environments
pip install -r requirements.txt
pip install .

Parallel Env Wrappers

State, Observation, Action, Reward Wrappers

• AgentStateParallelEnvWrapper

Environment Pipelines

• OrderForcingParallelEnvWrapper
• AutoResetParallelEnvWrapper

Improved PettingZoo Wrappers

• AECToParallelWrapper
• ParallelToAECWrapper

Base Environment Abstractions

• ParallelEnv: ParallelEnv with randomness control as in Gymnasium.

VectorEnv Implementation

VectorEnv

• SyncVectorParallelEnv: VectorEnv with serial step and reset.
• AsyncVectorParallelEnv: VectorEnv with parallel step and reset.

VectorEnv Wrappers

• AgentStateVectorParallelEnvWrapper
• SyncAgentStateVectorParallelEnvWrapper
• AsyncAgentStateVectorParallelEnvWrapper

Examples

For examples using these environments, wrappers and vector environments, please refer to the tests/async_vector_env_test.py and tests/sync_vector_env_test.py.