We introduce Voyager, the first LLM-powered embodied lifelong learning agent in Minecraft that continuously explores the world, acquires diverse skills, and makes novel discoveries without human intervention. Voyager consists of three key components: 1) an automatic curriculum that maximizes exploration, 2) an ever-growing skill library of executable code for storing and retrieving complex behaviors, and 3) a new iterative prompting mechanism that incorporates environment feedback, execution errors, and self-verification for program improvement. Voyager interacts with GPT-4 via blackbox queries, which bypasses the need for model parameter fine-tuning. The skills developed by Voyager are temporally extended, interpretable, and compositional, which compounds the agent’s abilities rapidly and alleviates catastrophic forgetting. Empirically, Voyager shows strong in-context lifelong learning capability and exhibits exceptional proficiency in playing Minecraft. It obtains 3.3× more unique items, travels 2.3× longer distances, and unlocks key tech tree milestones up to 15.3× faster than prior SOTA. Voyager is able to utilize the learned skill library in a new Minecraft world to solve novel tasks from scratch, while other techniques struggle to generalize.

In this repo, we provide Voyager code. This codebase is under MIT License.

Voyager requires Python ≥ 3.9 and Node.js ≥ 16.13.0. We have tested on Ubuntu 20.04, Windows 11, and macOS. You need to follow the instructions below to install Voyager.

Voyager requires Ollama to be installed as it magages and runs the open-source models, you can get it from here

Once installed run the following command:

ollama pull mxbai-embed-large

This will download an tested model that you can use as an base.

Now we can continue by downloading the model that we are going to use:

ollama pull llama2

After this is done you can continue to the next step.

git clone https://github.com/MineDojo/Voyager
cd Voyager
pip install --only-binary=:all: cchardet
pip install -e .

In addition to the Python dependencies, you need to install the following Node.js packages:

cd voyager/env/mineflayer
npm install -g npx
npm install
cd mineflayer-collectblock
npm install
npx tsc
cd ..
npm install

Voyager depends on Minecraft. You need to install Minecraft and set up a Minecraft (1.19 is tested and works. but newer and older versions should also work) instance.

Follow the instructions in Minecraft Login Tutorial to set up your Minecraft Instance.

You need to install fabric mods to support all the features in Voyager. Remember to use the correct Fabric version of all the mods.

Follow the instructions in Fabric Mods Install to install the mods.

Voyager uses OpenAI's GPT-4 as the language model. You need to have an OpenAI API key to use Voyager. You can get one from here.

After the installation process, you can run Voyager by:

from voyager import Voyager

# You can also use mc_port instead of azure_login, but azure_login is highly recommended
azure_login = {
    "client_id": "YOUR_CLIENT_ID",
    "redirect_url": "https://127.0.0.1/auth-response",
    "secret_value": "[OPTIONAL] YOUR_SECRET_VALUE",
    "version": "fabric-loader-0.14.18-1.19", # the version Voyager is tested on
}
openai_api_key = "YOUR_API_KEY"

voyager = Voyager(
    azure_login=azure_login,
    openai_api_key=openai_api_key,
)

# start lifelong learning
voyager.learn()

• If you are running with Azure Login for the first time, it will ask you to follow the command line instruction to generate a config file.
• For Azure Login, you also need to select the world and open the world to LAN by yourself. After you run voyager.learn() the game will pop up soon, you need to:
  1. Select Singleplayer and press Create New World.
  2. Set Game Mode to Creative and Difficulty to Peaceful.
  3. After the world is created, press Esc key and press Open to LAN.
  4. Select Allow cheats: ON and press Start LAN World. You will see the bot join the world soon.

If you dont have Java 21 installed already you can download it from here

Then install the Fabric minecraft server following the instructions at https://fabricmc.net/use/server/, or use the below:

curl -OJ https://meta.fabricmc.net/v2/versions/loader/1.19/0.14.18/0.11.2/server/jar
java -Xmx2G -jar fabric-server-mc.1.19-loader.0.14.18-launcher.0.11.2.jar nogui

This will start the server, creating the necessary folder structure. You should now kill the server by typing stop

You should now copy the mods into the mods directory as per https://github.com/MineDojo/Voyager/blob/main/installation/fabric_mods_install.md

Edit the server.properties file, changing the following values

difficulty=peaceful
gamemode=survival
online-mode=false
enable-rcon=true # Only needed if you wish to use RCON features
rcon.password=hunter2
rcon.port=25575
spawn-protection=0 # This is important, otherwise the bot can't do anything near the spawn

You can now start the server again. Now you can start Voyager, and have it connect with mc_port=25565. Once bot has joined the minecraft server, run op bot in the minecraft server console.

If you stop the learning process and want to resume from a checkpoint later, you can instantiate Voyager by:

from voyager import Voyager

voyager = Voyager(
    azure_login=azure_login,
    openai_api_key=openai_api_key,
    ckpt_dir="YOUR_CKPT_DIR",
    resume=True,
)

If you want to run Voyager for a specific task with a learned skill library, you should first pass the skill library directory to Voyager:

from voyager import Voyager

# First instantiate Voyager with skill_library_dir.
voyager = Voyager(
    azure_login=azure_login,
    openai_api_key=openai_api_key,
    skill_library_dir="./skill_library/trial1", # Load a learned skill library.
    ckpt_dir="YOUR_CKPT_DIR", # Feel free to use a new dir. Do not use the same dir as skill library because new events will still be recorded to ckpt_dir. 
    resume=False, # Do not resume from a skill library because this is not learning.
)

Then, you can run task decomposition. Notice: Occasionally, the task decomposition may not be logical. If you notice the printed sub-goals are flawed, you can rerun the decomposition.

# Run task decomposition
task = "YOUR TASK" # e.g. "Craft a diamond pickaxe"
sub_goals = voyager.decompose_task(task=task)

Finally, you can run the sub-goals with the learned skill library:

voyager.inference(sub_goals=sub_goals)

For all valid skill libraries, see Learned Skill Libraries.

If you have any questions, please check our FAQ first before opening an issue.

If you find our work useful, please consider citing us!

@article{wang2023voyager,
  title   = {Voyager: An Open-Ended Embodied Agent with Large Language Models},
  author  = {Guanzhi Wang and Yuqi Xie and Yunfan Jiang and Ajay Mandlekar and Chaowei Xiao and Yuke Zhu and Linxi Fan and Anima Anandkumar},
  year    = {2023},
  journal = {arXiv preprint arXiv: Arxiv-2305.16291}
}

Disclaimer: This project is strictly for research purposes, and not an official product from NVIDIA.