PCS 2502 (Projeto de Formatura)
This repository contains the code necessary to reproduce the results of my final project for the class PCS 2502 (Projeto de Formatura) at University of Sao Paulo in which we attempted to apply reinforcement learning techniques to the inverted pendulum control task.
The project is divided in the following folders:
- agents: Implementation of a few agents that were tested for solving the control task
- openai-gym: New environments created on OpenAI-gym used for testing
- tex: LaTeX files used to generate a report (in portuguese)
- videos: videos of the learning process for the DeepQLearning agent
Results
By applying a Deep learning agent to the control task we obtained encouraging results as we can see in the following chart:
Videos
To have an idea of how well the agent could perform in different stages of training you can take a look in the img folder:
