Using Reinforcement Learning to Solve Games

Reinforcement Learning: The process of developing software agents that perform actions based on rewards in their environment.

This is a submission for OpenAI's Gym Retro contest. OpenAI, anon-profit artificial intelligence research company created a platform for reinforcement learning research on games, named Gym. Gym Retro is a toolkit to do research on reinforcement algorithms and study generalization on older games, which are different but retain many of the same goals. For this specific submission, we utilzed Python to script the motion of Sonic and utilized a basic algorithm known as "Just Enough Retained Knowledge" aka JERK.

OpenAI and reinforcement learning first captured my attention while watching a multi-million dollar video game tournement. OpenAI's bot was able to defeat a professional player without any outside input while using reinforcement learning. I was extremely curious about how this happened and how it developed to this point. I watched lectures on machine learning in my spare time to get my feet wet and decided that it was a field that would I would take heed of for the future.

Code Available On GitHub