The Crawl Environment is an OpenAI Gym environment for testing learning machines on Dungeon Crawl Stone Soup (DCSS).

Note: I've only run this on Linux, so don't know if it works on other OSes

• DCSS ascii version (0.24 recommended)
• Python 3
• Pip 3 (Python package installer)

Several python modules.

• pip3 install setuptools
• pip3 install wheel
• pip3 install gym
• pip3 install pynput (needed for the test program only)

git clone https://github.com/powerbf/gym-crawl.git
cd gym-crawl
pip3 install -e .
export CRAWLDIR=<dir where crawl is installed>

Note: The program expects to find the DCSS executable at $CRAWLDIR/bin/crawl

The test program just sends random keystrokes to DCSS. It is not in any way intelligent.

python3 test-env.py -quick

(runs quickly, but only performs movement and eating)

python3 test-env.py

(runs slower, but performs a fuller range of actions - goes into menus for drop, wield, etc.)