Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HELLBRAIN

Reinforcement-learning agents that learn to play classic games from scratch.

HELLBRAIN drops a PPO agent into a game with no instructions and lets it figure the thing out — fumbling, then flailing, then (eventually) flexing. This repo is a family of standalone trainers, one per game, each on its own branch, so you clone only the one you want.

Status: experimental R&D. These are research toys, not products. Training an agent to play well takes a long time and good results aren't guaranteed. (Street Fighter is coming soon…)

DOOM            an agent learning to survive VizDoom
POKÉMON RED     an agent learning to explore Kanto, earn badges, level up
STREET FIGHTER  coming soon...???

Pick your game (each branch is a complete, standalone project)

git clone -b doom    https://github.com/revjmoney/HELLBRAIN-AI.git hellbrain-doom
git clone -b pokemon https://github.com/revjmoney/HELLBRAIN-AI.git hellbrain-pokemon
# Street Fighter — coming soon...???

Each branch has its own README with exact setup + run steps. The short version is always the same three beats:

pip install -r requirements.txt
# (commercial games: drop in your own legally-owned ROM — see the branch README)
python train.py            # add --allow-commercial for Pokémon

Training prints progress to your terminal and checkpoints to checkpoints/. Add --render to watch the game's own window while it learns.

What's shared across every branch

  • A lean, game-agnostic train.py (PPO CnnPolicy, auto-resume from the newest checkpoint, plain stdout progress — no GUI required).
  • A tiny env_factory.py that hands the trainer the right environment.
  • A config.py you can actually read in 30 seconds.

No telemetry, no accounts, no cloud. It runs on your box. Bring your own ROMs for the commercial titles (DOOM uses the free/libre Freedoom assets, so it just runs).

📺 Want the broadcast kit?

These trainers run great bare. But the part that turns "an AI is training" into an actual show — the HELLBRAIN Stream Kit — lives separately:

live OBS-ready dashboard · real-time stat panels · reward/score graphs · scrolling event ticker · swappable themes · automatic highlight clipping · instant replay · one-click highlight-reel builder.

It's available for a small, negotiable fee — more if you want a custom skin, custom graphics, or a brand-new game brought online and wired up end-to-end.

📧 therealrevjmoney@gmail.com

Built by Rev. J. Money. I teach the machines to fight so you don't have to. 😉

License

MIT (see LICENSE). Use it, fork it, ship it. The Stream Kit is a separate, proprietary add-on — that's the part that keeps the lights on.

About

PPO reinforcement-learning agents that learn to play classic games from scratch - DOOM, Pokemon Red, and Street Fighter II - one standalone trainer per branch. Experimental R&D; bring your own ROMs.

Topics

Resources

Contributing

Security policy

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors