b10902118/crtc

CRTC turns a mobile game the README only calls "CR" into an environment where computer programs can learn to play it through trial and error, a technique known as reinforcement learning. Instead of a person tapping the screen, an AI agent observes the game state and sends moves, over and over, to improve its play. Under the hood, the project runs the actual game engine on your own computer using a custom loader written in C++ that talks directly to the game's compiled code. A Python layer sits on top of that loader and exposes the game as a standard environment that reinforcement learning tools already know how to work with, following the same style of interface used by the popular Gymnasium and PettingZoo libraries, so existing RL agents and code can plug in with little modification. Getting started can be done two ways. A Google Colab notebook handles everything for you, so you can try it in a browser with a single click. For local use, the recommended path is Docker, which builds the loader for either x86 or arm processors and installs the necessary game files automatically, though the download step can be slow and a manually downloaded copy of the game speeds things up. Setting it up directly on your own machine without Docker is also possible but more involved, since the loader needs to be compiled for a 32 bit architecture. Once running, the environment can be controlled from a short Python script that resets the game, reads back observations, and steps through moves, or played interactively from the command line. The project is still early stage, with a listed to-do including more complete game state information, better logging, and eventually the ability to match bots against each other remotely. It is licensed under GPLv3, meaning any project built on top of it must also stay open source.