Using the "Advantage Actor Critic(A2C)" Reinforcement Learning method, the A.I 'Agent' is trained to play Atari's MsPacman.

The game used for this Model is Atari's MsPacman

The game was used by downloading ROM files.

AtariMania 'ROMS'

importingbreakoutGame

The A2C algortihm has been implemented for the Agent to play the game

refer the paper:

Asynchronous Methods for Deep Reinforcement Learning

The Python Libraries used:

StableBaselines-3

OpenAI-Gym

• Mark I (1 Million TimeSteps):

The first Model is a Preliminary Model which has been trained with 1 Million TimeSteps.

I had first set up for 20M timesteps but I am currently able to use my CPU instead of a GPU. This is the reason why the program is halted at 1M timesteps

Anyone could clone this repo and try the same on their own or for higher TimeSteps

Have Fun

Yours Truely,

Anuraag Rath :P