leoheidel / inf581

Project for INF581: Advanced Topics in Artificial Intelligence

Results

Diamonds	Lava

Abstract

The goal of this project is to use the reinforcement learning platform Malmo developed by Microsoft. Malmo allows to use Minecraft as a universe, for the purpose of researching and benchmarking reinforcement learning algorithms . In this project we evaluate various algorithms from stable-baselines, a fork from OpenAI Baselines. We also evaluate the importance of choosing the correct gain function.

Installing and running the environment

Since the installation of Minecraft and Malmo are quite complicated, everything has been automated inside a Docker container.

• First, head over to https://docs.docker.com/ and follow the instructions in order to install Docker on your machine.
• You can then execute the following command in the directory containing the code and the Dockerfile.

docker build . -t mymalmo

• Finally, you can launch the system with the command (replacing /path/to/this/directory with the actual absolute path)

docker run -it -p 5901:5901 -p 6901:6901 -p 8888:8888 -v /path/to/this/directory:/home/malmo/code -e VNC_PW=vncpassword mymalmo

You can then access a virtual desktop containing Minecraft on http://localhost:6901/?password=vncpassword.

A Jupyter Lab can be accessed on http://localhost:8888 (the link / token is printed in the console).

In the Jupyter Lab, our code can then be found in the code directory.