Skip to content

Latest commit

 

History

History
2 lines (2 loc) · 915 Bytes

README.md

File metadata and controls

2 lines (2 loc) · 915 Bytes

Super-Mario-Bros-RL-Player

This project aims to train a general Super Mario Bros. reinforcement learning agent, i.e. an agent that can solve any SMB level, even if it has not seen it before. This task is trickier than it seems at first. There are many tutorials out there that teach you how to train an RL agent to play SMB. The problem is that in these tutorials, they always train an agent to play the game, starting with World 1 Stage 1, then World 1 Stage 2, etc. Sometimes, they train a separate model for each stage! The problem with these approaches is that the agent does not learn to solve SMB generally. Instead, it learns to memorise each level and what to do when presented with a certain frame. Therefore, if the agent is asked to solve a level it has not seen before, e.g. a custom level, it will do poorly. With this project, I am trying to train a good enough agent that can play any SMB level.