Teaching Neural Nets to Fight with RL [P]

A project explored emergent behaviors by training two agents to play a Street Fighter-like game using Reinforcement Learning (RL). The developer found that without "league play," agents would only learn to exploit specific opponents rather than developing general strategies. The project's findings and a playable bot are available for users to interact with.

This project uniquely demonstrates that "league play" is crucial for RL agents to develop general strategies, unlike training against a single opponent.