Gym Retro is a platform for reinforcement learning research on games. It turns out that old games are a perfect fit for benchmarking and improving RL agents in a simulated environment. After two years after the beta release of Gym, OpenAI released an extension of this amazing software by adding more game environments. The goal of the new platform is to study the ability of the agents to generalize between games with similar concepts but different appearances.
Click this button to open a Workspace on FloydHub that will setup Gym-Retro.
In this contest, participants try to create the best agent for playing custom levels of the Sonic games — without having access to those levels during development. You can find more detail in this page.
As mentioned in the contest's description this is a Transfer Learning task, which means that you are free to train your agent however you'd like, however, the OpenAI Team recommend using Sonic 1, 2, and 3 & Knuckles, which are available on Steam here:
In this notebook, we will show you how to use set up Gym-Retro on FloydHub.
We will:
- Install Gym-Retro
- Import and load the ROMs of the games
- Use random-policy for a couple of steps
- Visualize the agents