CraKane / RLforAtari

RL for Atari

Introduction

• It is an attempt for model-free algorithms, including DQN/DDPG/PPO/AC

3 envs

• Point-v0
• CartPole-v0
• Gravitar-ram-v0

Methods

• As for Point & CartPole, we solved them by Policy Gradient algorithm to help us understand the basics of policy gradient algorithms.
• As for the most difficult env in Atari, Gravitar, we tried all kinds of tricks for model-free algorithms to see the performance.

Code Folder

• "Point_CartPole" is the solution and environments for Point-v0 & CartPole-v0.
• "Gravitar" is the solution for Gravitar in Atari (Gym).
• "Ass4_report.pdf" is the report for Point-v0 & CartPole-v0.
• "Gravitar.pdf" is the report for Gravitar-ram-v0.

NOTES

• other details are in "Ass4_report.pdf" and "Gravitar.pdf".