The goal of this project is to build an intelligent, collaborative assistant bot in the game of Minecraft¹ that can perform a wide variety of tasks specified by human players. Its primary purpose is to be a tool for artifical intelligence researchers interested in grounded dialogue and interactive learning. This project is in active development.

A detailed outline and documentation is available in this paper

This release is motivated by a long-term research agenda described here.

The fastest way to install CraftAssist is by building a Docker image.

After installing Docker, go to your Docker resource settings and make sure that the memory allocated to Docker is at least 4GB (smaller amounts of memory can result in the build crashing). Then, clone the CraftAssist repo and run:

cd craftassist/docker
docker build -t craftassist -f Dockerfile.client .

Do this section before cloning the repo.

Make sure the following packages have already been installed before moving on:

• CMake
• Python3
• Glog
• Boost
• Eigen
• gcc version: 7.4.0 on ubuntu 18.04
• For Mac users:
  • LLVM version < 10 to successfully use clang. Working with multiple versions of Xcode.

# OSX
brew install git-lfs
git lfs install

# On Ubuntu
sudo add-apt-repository ppa:git-core/ppa
curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
sudo apt-get install git-lfs
git lfs install

Use this command, or your submodules will not be pulled, and your clone will take a very long time:

git lfs clone --recursive git@github.com:facebookresearch/craftassist.git

Now cd craftassist and copy the large data file and untar it to the correct directories:

curl http://craftassist.s3-us-west-2.amazonaws.com/pubr/models_folder.tar.gz -o models_folder.tar.gz
tar -xzvf models_folder.tar.gz -C python/craftassist/models/ --strip-components 1
curl http://craftassist.s3-us-west-2.amazonaws.com/pubr/ground_truth_data.txt -o python/craftassist/ground_truth_data.txt

To build a conda environment that supports this release:

# Create a new env preloaded with the conda install dependencies
conda create -n minecraft_env python==3.7.4 pip numpy scikit-learn==0.19.1 pytorch torchvision -c conda-forge -c pytorch
conda activate minecraft_env

# Install all of the rest of the dependencies with pip
pip install -r requirements.txt

Then activate this environment whenever you want to run the agent.

To build Cuberite and the C++ Minecraft client:

make

If you did a local install, run the following from the cloned repository directory:

python ./python/cuberite_process.py

to start an instance of Cuberite listening on localhost:25565

If you installed CraftAssist using Docker, start a shell in a new Docker container with:

docker run -p 25565:25565 -it --name craftassist-container craftassist

and run the above python command in the default directory of that shell.

Buy and download the official Minecraft client.

You can inspect the world and view the Minecraft agent's actions by logging into the running Cuberite instance from the game client.

To connect the client to the running Cuberite instance, click in the Minecraft client:

Multiplayer > Direct Connect > localhost:25565

Minecraft has recently release v1.15.2, and our Cuberite system supports at most v1.12

Please follow these instructions to add a 1.12.x profile and use it to connect.

Assuming you have set up the Cuberite server and the client:

• Docker users should open a second shell in their container with:

docker exec -it craftassist-container bash

• Local users should open a new terminal in the repository directory.

In the new shell, run:

python ./python/craftassist/craftassist_agent.py

You should see a new bot player join the game.

Chat with the bot by pressing t to open the dialogue box, and Enter to submit.

Use the w, a, s, and d keys to navigate, left and right mouse clicks to destroy and place blocks, and e to open your inventory and select blocks to place.

./python/craftassist/test.sh

Download links to the datasets described in section 6 of Technical Whitepaper are provided here:

• The house dataset: https://craftassist.s3-us-west-2.amazonaws.com/pubr/house_data.tar.gz
• The segmentation dataset: https://craftassist.s3-us-west-2.amazonaws.com/pubr/instance_segmentation_data.tar.gz
• The dialogue dataset: https://craftassist.s3-us-west-2.amazonaws.com/pubr/dialogue_data.tar.gz

In the root of each tarball is a README that details the file structure contained within.

If you would like to cite this repository in your research, please cite the CraftAssist paper.

@misc{gray2019craftassist,
    title={CraftAssist: A Framework for Dialogue-enabled Interactive Agents},
    author={Jonathan Gray and Kavya Srinet and Yacine Jernite and Haonan Yu and Zhuoyuan Chen and Demi Guo and Siddharth Goyal and C. Lawrence Zitnick and Arthur Szlam},
    year={2019},
    eprint={1907.08584},
    archivePrefix={arXiv},
    primaryClass={cs.AI}
}

CraftAssist is MIT licensed.

¹ Minecraft features: © Mojang Synergies AB included courtesy of Mojang AB