[P] I trained an AI to play Resident Evil 4 Remake using Behavioral Cloning + LSTM

I recorded gameplay trajectories in RE4’s village — running, shooting, reloading, dodging — and used Behavioral Cloning to train a model to imitate my decisions. Added LSTM so the AI could carry memory across time steps, not just react to the current frame.

The most interesting result: the AI handled single enemies reasonably well, but struggled with the fight-or-flee decision when multiple enemies were on screen simultaneously. That nuance was hard to imitate without more data.

Full video breakdown on YouTube. Source code and notebooks here: https://github.com/paulo101977/notebooks-rl/tree/main/re4

Happy to answer questions about the approach.