Teaching Neural Nets to Fight with RL [P]

In this project I wanted to see if any interesting emergent behaviors would appear if we trained two agents to play a streetfighter-like game using RL. Maybe obvious in retrospect, but the agents are really good at reward hacking. I had to shape the rewards a bit to get them to even approach each other. I eventually used league play to improve the agent further. Without it, the agents don’t really learn general strategies, they just learn to exploit a particular opponent. You can read the article and try fighting the main bot yourself here: blog.lukesalamone.com/posts/fighting-game-rl