Generating-Interesting-Maps-in-Doom using Evolutionary Algorithms

Libraries used:

ViZDoom, MazeExplorer, Omgifol, Numpy

To Run:

a. cd into ~/MazeExplorer

b. python3 gen_maze_sandbox.py

Changes to make per individual:

a. Replace ~/MazeExplorer/mazeexplorer/maze.py with attached file (change line 163 to match your own path based on you local filesystem)

b. Replace ~/MazeExplorer/mazeexplorer/mazeexploreer.py with attached file (change line 82 to match your own path based on you local filesystem)

Evolutionary Algorithm: CMA-ES

We create a Vector of random points for the healthpacks (5) and random points for the wall verticies (15 points) and send that vector through CMA-ES to return multiple new map vectors. Each map was evaluated using a pre-trained agent that traveled through the map and determined how well it played.

DEMO