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)
• Group Relative Policy Optimization (GRPO)

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

• Proximal Policy Optimization (PPO)
  • continuous