Reinforcement Learning

Self-study and implementations of deep reinforcement learning papers/algorithms with a friend.

The following algorithms can be found in the repo:

1. Tabular Q-Learning
2. Deep Q-Learning
   [Paper: Playing Atari with Deep Reinforcement Learning]
3. REINFORCE (Vanilla Policy Gradient with Monte Carlo returns)
4. Advantage Actor Critic (A2C)
   [Paper: Asynchronous Methods for Deep Reinforcement Learning]
5. Proximal Policy Optimization (PPO)
   [Paper: Proximal Policy Optimization Algorithms]
6. Deep Deterministic Policy Gradients (DDPG)
   [Paper: Continuous control with deep reinforcement learning]
7. Dynamics Randomization for RL Transfer Learning
   [Paper: Sim-to-Real Transfer of Robotic Control with Dynamics Randomization]

Usage

1. Install dependencies using pip3 install -r requirements.txt
2. Each script has train and test methods. To call them, do python3 <script_name> <method_name>. For example: python3 REINFORCE.py train
3. The test method will load a model from the models directory. Pre-trained models for some algorithms can be found in this repo.