Note: These were designed for MineRL v0.4 and the code examples may not work on MineRL v1.0.

This repository contains examples and small tasks on getting started with MineRL environment.

To begin, install the requirements for MineRL, and then install Python requirements with with pip install -r requirements.txt. We also have Colab notebooks in case installing these libraries is not possible.

If you have any questions, you can reach us on Discord. If you spot typos/bugs in any of the tasks or this repo, do tell us via Github issues!

Stars indicate the difficulty of the task. Click the task to see more details.

⭐ Getting started with MineRL.

• Start by playing bit of Minecraft via MineRL with playing_with_minerl.py script.
• Check out getting_familiar_with_minerl_and_gym.py and follow the instructions to get familiar with the agent-environment (Gym) API.
• You can find the latter task on Colab here.

⭐⭐ Improve Intro baseline of the Diamond competition.

• Step-by-step instructions on how to improve a simple, fully-scripted agent for obtaining wood and stone in the MineRLObtainDiamond-v0 task.
• Start out by opening this document and following the instructions.
• If you are overwhelmingly stuck, you may look for reference answers from this documentation.

⭐⭐ Implementing behavioural cloning from (almost) scratch.

• Start by opening up behavioural_cloning.py and following the instructions at the beginning of the file.
• You can also find the task on Colab here.
• You can find a crude reference answers in this Colab notebook. This task is built on the BC + scripted baseline solution.

⭐⭐⭐ Learn how to use stable-baselines and imitation libraries with MineRL.

• This walk-through demonstrates how to combine well-established reinforcement learning (stable-baselines3) and imitation learning (imitation) libraries with MineRL to train more sophisticated agents.
• Start by opening this Colab link.

⭐⭐⭐ Improve Research baseline of the Diamond competition.

• Similar to the second task here, but in a more difficult setting where you may not manually encode actions.
• Get started by opening this documentation.
• Once done, you can check reference answers from here.

⭐⭐⭐ Useful utilities for the BASALT 2021 competition.

• A collection and a walkthrough of approaches and methods that are useful for the BASALT competition (learning without rewards).
• Get started by opening this Colab.