This Python package provides Gym environments and data access for training agents in Minecraft, targeting researchers and developers in sample-efficient reinforcement learning. It offers a standardized interface to Minecraft for RL experimentation, enabling the development and benchmarking of advanced AI agents.

How It Works

MineRL integrates with Minecraft, providing a Gym-like API to interact with the game. It leverages a custom Minecraft client to render observations (pixels) and execute actions, abstracting away complex game mechanics. The package supports multiple versions, each tailored for specific competition datasets and Minecraft versions, offering flexibility for different research needs.

Quick Start & Requirements

• Install with pip install git+https://github.com/minerllabs/minerl.
• Requires Java JDK 8 (additional steps may be needed for macOS).
• Official documentation: https://minerl.io/docs/

Highlighted Details

• Supports multiple MineRL versions (v1.0, v0.4, v0.3) with distinct features and datasets.
• v1.0 features a new Minecraft version (11.2-16.5), higher resolution (640x360), and a near-human action space focusing on GUI and mouse control.
• Original MineRL-v0 datasets are available via Zenodo: https://zenodo.org/records/12659939.
• Example usage provided for basic environment interaction and agent control.

Maintenance & Community

• The project actively collects and showcases community projects.
• Edit the README to add your project: https://github.com/minerllabs/minerl/edit/master/README.md

Licensing & Compatibility

• The license is not explicitly stated in the README. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README mentions that original data mirrors for MineRL-v0 are down, with copies provided on Zenodo. Specific compatibility notes for macOS installation are also mentioned.