Skybrain-AI 🧠🤖

AI which aims to solve the SkyBlock Minecraft survival challange. The network is created by using the Double DQN algorithm from StableBaselines3 customized with Prioritized Replay.

Prerequisits

• Java 1.8

Setup project

1. Installing Java JDK

It is required to have Java SDK 1.8.x.x installed, unfortunately a newer version such as 1.9 will not work.

MacOS

To install:

brew install --cask adoptopenjdk8

This will set Java 1.8 to your system-default Java version

Insert the following line to the file ~/.zshrc

export JAVA_HOME=$(/usr/libexec/java_home -v 1.8)

Linux (Ubuntu- or Debian-based)

apt-get install openjdk-8-jdk

Verify that the java-version is correct by running java -version. The output should be 1.8.x.x

2. Installing Dependencies:

pip install minedojo stable-baselines3 torchvision tensorboard

change variables guiScale and renderDistance in venv/lib/python../site-packages/minedojo/sim/Malmo/Minecraft/run/options.txt to 1

3. Run traning

To run the training, use the Jupyter Notebook CoolNotz.ipynb

View training progress

During training, the training-agent will log it's performance using tensorboard. A tensorboard-server will need to be booted up to view this data. This may be done by entering the following code in the terminal

python3 -m tensorboard.main --logdir ./DQN_steve_tensorboard/