andreped/super-ml-pets

Trains RL agents in a custom gym environment and deploys them via computer vision-based screen capture and keyboard/mouse control to play against real opponents in-game. Supports finetuning existing models and works cross-platform (Windows, macOS, Linux) with deployment tested on GitHub Codespaces and Google Colab. Uses Stable Baselines3 for reinforcement learning and requires 1920x1080 fullscreen gameplay with classic UI styling for the vision system to function.

No commits in the last 6 months.