I’m making a game where you can train your own neural network-based AI and let it fight in an arena. The core gameplay is similar to that of auto-battler games like Totally Accurate Battle Simulator, but with battlers controlled by code-based AI, human players, or neural network-based AI.
In this game, you can:
- Design game characters controlled by code, AI, or human players.
- Put characters in playgrounds (maps) to fight.
- Train an AI for your characters.
- The model is quite small, so you can train it locally, even without a GPU.
If you are interested in this idea, I have made a pre-release of the current game (you can also find the installation guide there):
And here’s the GitHub link with a few videos:
I’ve only finished the game’s scaffolding, so it severely lacks content right now, but the key functions are more or less there, and yes, you can train an AI.
(Godot itself cannot handle AI training, so it is basically a Godot game + a Python AI training environment running locally)
The game (especially the AI training part) isn’t very intuitive, so I attached a manual in the game folder. Yeah, reading a manual for a game is not so pleasant, but I haven’t been able to add an in-game tutorial. I’m sorry for that.
I appreciate any feedback/bug reports from you; have fun (hopefully) 
Finally, some AI declaration. I started using ChatGPT Codex for this project (mainly code and text) earlier this year. I apologise for any AI slop you may find in-game (especially text). The game has been tested and debugged by me and some friends, but I’m 100% sure bugs still exist. I apologise for any inconvenience it caused.