Reinforcement learning is a machine learning algorithm that aims to teach software agents taking actions in an environment by maximizing a reward function. In this project, I implemented several algorithms including Policy Gradient, Deep Q-Learning (DQN), and A2C for the atari games, such as LunarLander, Assault, and Mario. My results are shown above in Figure 1.

1. Policy Gradient for LunarLander

A policy-based policy gradient agent with REIFORCE algorithm. In addition, I implement some improvements including reward normalization and proximal policy optimization.

2. Deep Q-learning for Assault

A value-based deep Q-learning agent. In addition, I implement some improvements including Double DQN and Dueling DQN.

3. Advantage-Actor-Critic for Mario

An actor-critic agent to consider both policy and value. In addition, I implement some improvements including proximal policy optimization and generalized advantage estimation.

For more information, please refer to my technical report and my code is also publically available on GitHub.