This repository contains the maze scenario where an agent must learn to navigate within randomly generated maze environments to obtain target objects and avoid non-target objects scattered within the map.

1. To run the random agent, MazeExplorer must be installed. Please follow this link for installation.
2. Other requirements can be installed as below:

pip3 install numpy torch
pip3 install setproctitle

3. gen_mazes.py generates maze config files and WAD files in mazes_dir/ directory.

python gen_mazes.py

4. main.py runs the random agent with multiprocessing. Prior to running main.py, the log_file of class Params in params.py must be changed to a nonexisting file name. The raw data will be saved to the CSV file of the specified log_file path.

python main.py

Significant portion of environment code and scripts under mazeenv directory are modified versions of microsoft/MazeExplorer repository (link). The copyright notice is included in NOTICE.

main.py and eval.py are modified versions of the corresponding scripts in the kibeomKim/House3D_baseline repository (link).