Train RL agents to play Pokemon Red

New 10-19-24! Updated & Simplified V2 Training Script - See V2 below

New 1-29-24! - Multiplayer Live Training Broadcast 🎦 🔴 View Here

Stream your training session to a shared global game map using the Broadcast Wrapper

See how in Training Broadcast section

Watch the Video on Youtube!

Join the discord server

Running the Pretrained Model Interactively 🎮

🐍 Python 3.10+ is recommended. Other versions may work but have not been tested.
You also need to install ffmpeg and have it available in the command line.

Windows Setup

Refer to this Windows Setup Guide

For AMD GPUs

Follow this guide to install pytorch with ROCm support

Linux / MacOS

V2 is now recommended over the original version. You may follow all steps below but replace baselines with v2.

1. Copy your legally obtained Pokemon Red ROM into the base directory. You can find this using google, it should be 1MB. Rename it to PokemonRed.gb if it is not already. The sha1 sum should be ea9bcae617fdf159b045185467ae58b2e4a48b9a, which you can verify by running shasum PokemonRed.gb.
2. Move into the baselines/ directory:
   cd baselines
3. Install dependencies:
   pip install -r requirements.txt
   It may be necessary in some cases to separately install the SDL libraries. For V2 MacOS users should use macos_requirements.txt instead of requirements.txt
4. Run:
   python run_pretrained_interactive.py

Interact with the emulator using the arrow keys and the a and s keys (A and B buttons).
You can pause the AI's input during the game by editing agent_enabled.txt

Note: the Pokemon.gb file MUST be in the main directory and your current directory MUST be the baselines/ directory in order for this to work.

Training the Model 🏋️

V2

• Trains faster and with less memory
• Reaches Cerulean
• Streams to map by default
• Other improvements

Replaces the frame KNN with a coordinate based exploration reward, as well as some other tweaks.

1. Previous steps but in the v2 directory instead of baselines
2. Run: python baseline_fast_v2.py

Tracking Training Progress 📈

Training Broadcast

Stream your training session to a shared global game map using the Broadcast Wrapper on your environment like this:

env = StreamWrapper(
            env, 
            stream_metadata = { # All of this is part is optional
                "user": "super-cool-user", # choose your own username
                "env_id": id, # environment identifier
                "color": "#0033ff", # choose your color :)
                "extra": "", # any extra text you put here will be displayed
            }
        )

Hack on the broadcast viewing client or set up your own local stream with this repo:

https://github.com/pwhiddy/pokerl-map-viz/

Local Metrics

The current state of each game is rendered to images in the session directory.
You can track the progress in tensorboard by moving into the session directory and running:
tensorboard --logdir .
You can then navigate to localhost:6006 in your browser to view metrics.
To enable wandb integration, change use_wandb_logging in the training script to True.

Static Visualization 🐜

Map visualization code can be found in visualization/ directory.

Follow up work

Check out our follow up projects & papers!

Pokemon Red via Reinforcement Learning 🔗

  @misc{pleines2025pokemon,
    title={Pokemon Red via Reinforcement Learning},
    author={Marco Pleines and Daniel Addis and David Rubinstein and Frank Zimmer and Mike Preuss and Peter Whidden},
    year={2025},
    eprint={2502.19920},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
  }

Pokemon RL Edition 🔗

PokeGym 🔗

Supporting Libraries

Check out these awesome projects!

PyBoy

Stable Baselines 3