Neural MMO 2.0: A Massively Multi-task Addition to Massively Multi-agent Learning

TL;DR

Neural MMO 2.0 introduces a flexible, multi-task environment for multi-agent reinforcement learning, enabling agents to generalize across diverse tasks, maps, and opponents, with improved performance and open-source accessibility.

Contribution

It presents a complete rewrite of Neural MMO with enhanced performance, a versatile task system, and broad compatibility, fostering research in generalization for multi-agent RL.

Findings

Three-fold performance improvement over previous version

Supports a wide range of customizable tasks and objectives

Facilitates research on generalization to unseen scenarios

Abstract

Neural MMO 2.0 is a massively multi-agent environment for reinforcement learning research. The key feature of this new version is a flexible task system that allows users to define a broad range of objectives and reward signals. We challenge researchers to train agents capable of generalizing to tasks, maps, and opponents never seen during training. Neural MMO features procedurally generated maps with 128 agents in the standard setting and support for up to. Version 2.0 is a complete rewrite of its predecessor with three-fold improved performance and compatibility with CleanRL. We release the platform as free and open-source software with comprehensive documentation available at neuralmmo.github.io and an active community Discord. To spark initial research on this new platform, we are concurrently running a competition at NeurIPS 2023.