Deep RL Algorithms in PyTorch

Models

• DQN
• Dueling Double DQN
• Categorical DQN (C51)
• Categotical Dueling Double DQN
• Proximal Policy Optimization (PPO)
  • discrete (episodic, n-step)
• Soft Actor-Critic (SAC)
  • debugging

Exploration

• Random Network Distillation (RND)

Experiments

The result of passing the environment-defined "solving" criteria.

• Dueling Double DQN
  • Only one hyperparameter "UP_COEF" was adjusted.

CartPole-v0

CartPole-v1

MountainCar-v0

LunarLander-v2

TODO

• Quantile Regression DQN (QR DQN)
• Implicit Quantile Networks (IQN)
• Intrinsic Curiosity Module (ICM)
• Rainbow
• Parametric DQN
• Proximal Policy Optimization (PPO)
  • continuous
• Deep Deterministic Policy Gradient (DDPG)
• MCTS Net
• Parallel Models
  • Ape-X
  • R2D2
  • PAAC