Super-Auto-Pets

Developing AIs for Super Auto Pets

Current status: Using Deep Q-Learning to train AI. Simulated training works well, real world training quite bad...

V1.0: Trains AI directly with the game (plays against real people)

V6.0: Smarter training scheme, training against itself.

To Do:

1. (V6.0): Train an accurate model in the simulation-based enviroment (training against itself)
2. (V1.0): Need to be able to get information from screen, to use that with what is used in V6.0
3. Git gud
4. ...
5. $$

Acknowledgements

This implementation is based on the work by HJK-Z, with aim to improve the design with smarter reinforcement learning schemes. The game logic is built around sapai.