Naive-Reinforcement-Learning-With-Atari-Games

Game Environment

Atari game environments in the this project are all available on Open AI. The project focuses on centipede-ram-v0, BreakoutDeterministic-v4, and Taxi-v2. We have selected these games to observe naive reinforcement learning algorithms (without neutral networks) on how well the algorihtms can guide the agent to play markovian games (Taxi-v2), and non-markovian games (BreakoutDeterministic-v4, Centipede-ram-v0).

• check observation variables
• check environment variables
• more information about variables in openai

Algorithms

Naive Q-Learning

• check this website for more details
• check this page for implementation details

SARSA

• check thsi website
• check this page for implementation details

Monte Carlo Tree Search

• check this website
• check this page for implementation details

Results

• SARSA plays Taxi-v2

• Q-learning plays Taxi-v2

• For more information, please check this report.

Note

• pip install gym and pip install gym[atari] to run atari game environments