neural-mmo  by openai

Massively multiagent game environment for training and evaluating intelligent agents

created 6 years ago
1,624 stars

Top 26.4% on sourcepulse

GitHubView on GitHub
Project Summary

This project provides Neural MMO, a massively multiagent game environment designed for training and evaluating intelligent agents in complex, evolving worlds. It aims to serve as a proxy for real-world complexity, targeting AI researchers and engineers interested in artificial life and scalable agent training.

How It Works

Neural MMO employs a novel, framework-independent "native API" that offers significant efficiency gains over traditional Gym environments for multiagent settings. It decouples agent scalability from environment scalability, addressing each as a distinct research and engineering challenge. The architecture comprises four modules: Blade (core environment), Embyr (3D renderer), Ethyr (research tools), and Trinity (native API).

Quick Start & Requirements

  • Install: Clone openai/neural-mmo and jsuarez5341/neural-mmo-client, then run bash setup.sh in the client directory and bash scripts/setup/setup.sh followed by python setup.py in the environment directory.
  • Prerequisites: Anaconda with Python 3.6+, PyTorch (setup separately).
  • Run: python Forge.py --render and navigate to http://localhost:8080/forge/embyr/.
  • Docs: Client Repo

Highlighted Details

  • Native API designed for efficiency and simplicity in multiagent scenarios.
  • Persistent environment state, with dones always None.
  • Includes both a native API and a minimally modified Gym API.
  • Employs a Three.js-based web client for rendering.

Maintenance & Community

This repository is archived; active development continues at jsuarez5341/neural-mmo.

Licensing & Compatibility

The OpenAI repository is open-sourced under the MIT License. Some assets are from Runescape (Jagex) and used under fair use for research purposes; these may be removed upon request.

Limitations & Caveats

The project is archived and no longer actively maintained. Potential issues include computational infeasibility for very large-scale experiments and the inherent difficulty of the "agents that scale to their environment" problem (exploration). The complexity gap between the current environment and real-world MMOs remains.

Health Check
Last commit

2 years ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
18 stars in the last 90 days

Explore Similar Projects

Feedback? Help us improve.