This repository provides environment generation code for multi-agent reinforcement learning research, specifically for the "Emergent Tool Use From Multi-Agent Autocurricula" paper. It enables the creation of complex, dynamic environments for training agents, targeting researchers in multi-agent RL and AI.

How It Works

The system constructs environments by starting with a Base environment and layering modular components (EnvModule classes) for objects and dynamics, and gym.Wrapper classes for game mechanics, rewards, and observations. This modular approach minimizes code duplication and facilitates the creation of novel environments by extending existing functionality rather than deep subclassing.

Quick Start & Requirements

• Install by cloning mujoco-worldgen, installing its requirements (pip install -r mujoco-worldgen/requirements.txt), then installing mujoco-worldgen and this repo (pip install -e mujoco-worldgen/, pip install -e multi-agent-emergence-environments/).
• Requires Python 3.6. Tested on Mac OS X and Ubuntu 16.04.
• Additional packages for policy playback: pip install -r multi-agent-emergence-environments/requirements_ma_policy.txt.
• Official documentation and examples are available within the repository structure.

Highlighted Details

• Implements environments for Hide and Seek, Box Locking, Blueprint Construction, and Shelter Construction.
• Supports modular environment construction via EnvModule and gym.Wrapper classes.
• Includes a bin/examine script for testing environments and playing saved policies.

Maintenance & Community

• Status: Archive (code is provided as-is, no updates expected).

Licensing & Compatibility

• License: Not explicitly stated in the README. Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

The project is archived and will not receive updates. It has only been tested on older OS versions (Ubuntu 16.04) and Python 3.6, suggesting potential compatibility issues with modern systems.