Zombie-Maze-Land Q-Approximation in Malmo
Zombie-Maze-Land
Summary
The goal of this project is to have an agent survive for as long as possible in a 10x10 block environment with a maze and three zombies placed somewhere in the maze. The agent starts off in the middle of this environment. The zombies are placed at three pre-chosen spots. The agent must survive for as long as possible. To learn how to move, we train the agent with reinforcement learning, specifically, we use a Q-function approximation instead of a Q-table. See figure below for visualization:
Figure 1: Picture of 10x10 maze environment for Malmo-Minecraft agent. Black spots indicate walls. White spots indicate free path. Green spot indicates starting position of agent. Red spot indicates starting position of zombies.