AI Squared Tournament: A Flexible Reinforcement Learning Framework for 1v1 Platform Fighting Agents

Abstract

Reinforcement Learning (RL) is an often overlooked area of Machine Learning, resulting in the number of opportunities for people to learn the subject oftentimes being limited. The goal of AI Squared is to create a way that allows for people of all backgrounds to learn RL in a fun, competitive, and exciting way. The AI Squared Project consisted of a tournament and a structured iPynb Notebook to allow people to design, train, and battle AI agents, teaching them RL concepts along the way. In the tournament, agents fight in a custom environment, a 1v1 knockout fighting game inspired by Brawlhalla.