Stanford CS 221 (Principles of Artificial Intelligence) Final Project: Finding Water in Minecraft

Lydia Chan, Russell Tran
13 December 2019

We use Deep Q-Learning (DQN) to train an agent to search for bodies of water in the video game Minecraft. The agent reads in raw pixel inputs and has the controls of a normal player.

Directory

• code: Run the training algorithms and simulate Minecraft. To run, call python3 {find_water_baseline.py, find_water_dqn0.py, find_water_q_learning.py}
• environments: These are the xml files which represent different Minecraft worlds/environments in which the agent can roam. These xml files are parsed by Minecraft Malmo--refer to their documentation for the formatting. The MineRL platform is capable of taking these Minecraft Malmo environments (which Malmo calls "missions") and using them as OpenAI gym environmetns.
• out: Data output on our runs
• poster data: Assets for our poster

Below are the necessary dependencies for macOS and Windows:

tensorflow==1.14

minerl==0.2.9

pandas==0.24.8

gym==0.15.3

mujoco-py>=2.0.2.8

mpi4py==3.0.3

baselines==0.1.5

lxml==4.4.1

psutil>=5.6.2