Skip to content
Explainable & Easy-to-debug Deep Reinforcement Learning Framework
Python Jupyter Notebook
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Stable Baselines with TF 2.0

This repository is based on the original implementations of Stable Baselines. (

In this version, we pursuit following properties:

  1. Easy to debug using Eager-execution and Tensorflow 2.0.
  2. Easy to read, as simple as possible.

Quick start

We recommend to use Anaconda virtual environment. With using environment.yml (or environment_gpu.yml if you plan to use GPUs), you can easily install what you need to run! (To run experiments in Mujoco simulation tasks, make sure that you have the license for it.)

To start, enter the following commands in a terminal:

git clone
cd stable_baselines_tf2
conda env create -f environment.yml


  • Abstract classes :

    • base/ : Includes abstract classes for RL algorihtm
    • base/ : Includes abstract classes for RL policy
    • base/ : Includes replay buffer implementation (which is for off-policy RL algorithms)
    • TBU : Tensorboard Writer
  • Algorithms : (Not working yet / Working but not verified / Verified (reproduced) )

    • DQN : Working but not verified
    • SAC : Working but not verified
    • DDPG : Not working yet
    • Algorithms planning to add : TRPO/PPO, GAIL
You can’t perform that action at this time.