Reinforcement-Learning

This repository demonstrates the implementation of SARSA, Q-Learning and Expected SARSA to the GYM - FrozenLake8x8-v0, an OpenAI environmental simulator. The work in this repository is inspired by the article by Vaibhav Kumar.

Frozen Lake Environment

The frozen lake environment is an 8x8 grid world and has total 64 states. In each episode the agent starts at S and the episode terminates when it reaches either hole H or goal G. The reward for reaching the goal is 1 while going to any other state results in 0 reward.

All three algorithms are implemented as given in Reinforcement Learning: An Introduction by Richard S. Sutton and Andrew G. Barto.

TD Q-Learning

The figure above illustrates the results obtained with Q - Learning.

• The repository is still being updated for a better navigation through the code.