This repository reimplements the most popular reinforcement learning algorithms & components.
- Deep Q-Learning (parallelized) (DQN)
- Double Deep Q-Learning (DDQN)
- Prioritized Experience Replay (PER)
- Multi-step learning with n-step TD-targets (Sutton and Barto, Chapter 7.1)
- Dueling network architectures (Dueling DDQN)
- Noisy networks for exploration (Noisy Nets)
- Distributional Perspective on Reinforcement Learning (C51)
- Rainbow: Combining Improvements in Deep Reinforcement Learning (parallelized) (Rainbow)
- Proximal Policy Optimization Algorithms (parallelized) (PPO)
- Soft Actor-Critic (parallelized) (SAC)
- Contrastive Unsupervised Representations for Reinforcement Learning (CURL)
- Atari tricks: frame stacking, action repetitions, no-ops actions