This repository provides the code for RoboSumo, a suite of competitive multi-agent environments designed for research in continuous adaptation via meta-learning. It is targeted at researchers and practitioners in reinforcement learning and multi-agent systems who need a platform for studying adaptation in nonstationary and competitive settings. The primary benefit is a standardized environment for evaluating meta-learning algorithms in dynamic, adversarial scenarios.

How It Works

RoboSumo utilizes the MuJoCo physics simulator and OpenAI Gym for environment creation. The core innovation lies in its competitive, multi-agent setup where agents continuously adapt to each other's evolving strategies. This nonstationary nature challenges standard RL algorithms, making it suitable for testing meta-learning approaches that aim to learn adaptation rules.

Quick Start & Requirements

• Install via pip: pip install -r requirements.txt followed by pip install -e . after cloning the repository.
• Prerequisites: numpy, gym, mujoco_py>=1.5. Running demos requires tensorflow>=1.1.0 and click.
• Demo execution: python demos/play.py with options to select environments, policy architectures, and parameter versions.
• Official documentation: Installation guide for MuJoCo is referenced.

Highlighted Details

• Implements competitive multi-agent environments for reinforcement learning research.
• Focuses on nonstationary and adversarial settings to test adaptation capabilities.
• Supports different policy architectures (MLP, LSTM) and parameter versions for agent configurations.

Maintenance & Community

The project is marked as "Archive" and no updates are expected. It originates from OpenAI.

Licensing & Compatibility

The repository does not explicitly state a license in the provided README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is archived, indicating no further development or support. The dependency on tensorflow>=1.1.0 may pose compatibility issues with modern TensorFlow versions.