Exploration and Evaluation of Deep Reinforcement Learning in the CartPole environment

In this work we present an overview of a Deep Reinforcement Learning method using Q-learning that recently brought the state-of-the-art results in the Atari Emulator Environment. This method is called DQN algorithm that combines successfully a convolutional neural network with Q-learning. Successively it was improved with Double Q-learning and other valid techniques. Attracted by this hot research area we replicated the DQN algorithm and run it in the CartPole environment, a famous toy control challenge that allowed us to deal with the main issues related to the implementation of DQN. In this paper we reported the results that we obtained, our consideration on them and some personal thoughts about the Deep Reinforcement Learning field.

Report (pdf)

https://github.com/Codernauti/Exploration-of-DQN-in-CartPole-Environment/releases/download/v1.0/ExplorationAndEvaluationDRL_CartPole.environment.pdf

Presentation

https://docs.google.com/presentation/d/1DylaW-f55jYYyvB7s36DhKgBVZDhve7be8MmX9abGA0/edit?usp=sharing

Source code

See the following repository: https://github.com/EduBic/DQN-in-CartPole

This works is based on the following papers:

• Playing Atari with Deep Reinforcement Learning [1]: https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf
• Human-level control through deep reinforcement learning [2]: https://www.nature.com/articles/nature14236
• Double Q Learning [3]: https://arxiv.org/abs/1509.06461
• Deep Reinforcement Learning with Double Q-learning [4]: https://arxiv.org/abs/1509.06461