Quick links: Github | Baselines | NeurIPS 2021 Paper | Discord | Twitter

Introduction

Neural MMO is an open-source and computationally accessible research platform that simulates populations of agents in procedurally generated virtual worlds. We support basic foraging tasks involving a few agents for a couple of minutes, thousand-agent joint survival + exploration + combat over multiple hours, and everything between. Our goal is to support a broad base of multiagent research that would be impractical or impossible to conduct on other environments.

from nmmo import Env

env = Env(config=None) # Default environment. Keep reading for config options
obs = env.reset()

while True:
    actions = {} # Compute with your model
    obs, rewards, dones, infos = env.step(actions)

Environments provide a standard PettingZoo API. Join our community Discord and post in #support for help (do not raise Github issues for support). See the quick links above for source code, baselines, latest publications, social media, and news!

@inproceedings{NEURIPS DATASETS AND BENCHMARKS2021_44f683a8,
  author = {Suarez, Joseph and Du, Yilun and Zhu, Clare and Mordatch, Igor and Isola, Phillip},
  booktitle = {Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks},
  editor = {J. Vanschoren and S. Yeung},
  pages = {},
  title = {The Neural MMO Platform for Massively Multiagent Research},
  url = {https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/file/44f683a84163b3523afe57c2e008bc8c-Paper-round1.pdf},
  volume = {1},
  year = {2021}
}

In the presentation above, I coin the term Foundation Policies, analogous to Foundation Models, as a grounding motivation for investing in environment complexity and generality. A key contribution of Neural MMO is to do so efficiently and interpretably by adapting techniques from classic MMO development for deep learning.

The associated poster provides a high level overview of the workflow intended to support most research on the platform. For a more thorough walkthrough of features and usage, read on.

Installation

Official support: Ubuntu 20.04, WSL, and MacOS. Tested with Anaconda Python 3.9

Packaged installation

# Install NMMO with baseline dependencies (quotes for mac compatibility).
pip install "nmmo[cleanrl]"

# Clone baselines repository. Optional but recommended: setup WanDB integration.
git clone https://github.com/neuralmmo/baselines nmmo-baselines
echo YOUR_WANDB_API_KEY > nmmo-baselines/wandb_api_key

#Run a quick demo (download client below)
python -m demos.minimal

Download the latest client here (WSL users: do this on your Windows host). Start the demo and run the executable for your platform in client/UnityClient/. After a few seconds, the demo console will show a connection message and the client will load the map. The on-screen instructions demonstrate how to pan and zoom. You can also click on agents to examine their skill levels. The in-game console (which you can toggle with tab) gives you access to a number of overlay visualiztions.

Setup from source for developers (slow without –depth=1)

mkdir neural-mmo && cd neural-mmo

git clone https://github.com/neuralmmo/environment
git clone https://github.com/neuralmmo/baselines
git clone https://github.com/neuralmmo/client

echo YOUR_WANDB_API_KEY > baselines/wandb_api_key
cd environment && pip install -e .[all]

Gallery

Perspective and UI

Overlays

Multiscale Terrain Generation

Overhead Render

Call for Collaborations

Following the platform’s recent publication in NeurIPS 2021, we are excited to announce an open call for collaborations!

Eligibility (at least one of)

• You are affiliated with an academic lab (professor/PhD student/postdoc)

• You have previously published in a relevent area (RL, PCG, etc)

• You are a corporate researcher with substantial freedom to publish

• You have a substantial engineering background and want to help with core development

Excepting prospective developers, your objectives should include first-author publication at a top venue. We (Joseph Suarez and Phillip Isola, MIT) would be included as second authors. Compared to working independently, we can offer:

Benefits:

• Extended support and custom features

• Project-specific guidance, having developed the platform

• An invitation to our monthly group meetings

• Scoop insurance (we avoid duplicate projects)

• Introductions to other practitioners in our community

Contact me via Discord or email (in the publication) if you are interested.