In this project, we analyze some popular ways that are used to train agents in the domain of Reinforcement Learning. Initially, we tried having our own environment and using it, but since we wanted to move to a more complex environment, we ended up using OpenAI's Gym.
After working on our environment, we tried using Q-Learning on Berkley AI Course's Pacman Environment.
-
Q-learning on Berkley Pacman
-
Q-learning on Cartpool
-
Deep Q Network directly using observation space vector
-
Deep Q Network using convolutions on the render of the environment
-
Double Q Learning
A detailed report of our work can be found in final_report.prf