The purpose of this reinforcement learning project is to create working agents in order to win several Atari 2600 games using a Deep Q Learning algorithm. This includes:
- AirRaid
- Air-Striker Genesis
- Assault
- Asterix
- Asteroids
- Centipede
- Defender
- DemonAttack
- Frostbite
- Hero
- IceHockey
- MsPacman
- Pong
- Riverraid
- Space-Invaders
In doing so it uses gym retro in order to import the necessary Atari games. If you're having trouble importing the necessary ROMs, refer to this.
Given that it's pixelated, this project uses stable_baseline's CnnPolicy with the Deep Q Learning algorithm in order to train the models.