Distributed Deep Reinforcement Learning

Distributed Deep Reinforcement Learning by Ray and TensorFlow.

This framework inspired by general-purpose RL training system Rapid from OpenAI.

Rapid framework: This framework:

Tutorial (Chinese version)

• Parallelize your algorithm by Ray (1)
• Parallelize your algorithm by Ray (2)
• Parallelize your algorithm by Ray (3)

In short. This framework divides the reinforcement learning process into five parts:

• Replay buffer (option)
• Parameter server
• train (learn)
• rollout
• test

简单实验对比：

实验：LunarLanderContinuous-v2

算法：SAC

未调参，sac和dsac参数相同，dsac的worker数量：1。GPU：GTX1060

(dsac: distributed sac)