liziniu/RLX

RLX is an RL codebase based on TensorFlow. It implements algorithms like SAC, ACER, GAIL and TRPO. It is easy to use.

RLX

Overview

This is a reinforcement learning (RL) codebase based on TensorFlow without tedious sess.run(). Its core components borrows lunzi module in slbo. With it, we can write TensorFlow in a Pythonic style. Overall, this codebase aims to provide a clean and easy tool to conduct research experiments on Atari and MuJoCo.

Algorithm

We don't aim to provide a universe interface for all algorithms on diverse environments. Thus, the listed algorithms are only implemented for specific tasks like Atari or MuJoCo.

• ACER[Atari]

• TRPO[MuJoCo]

• SAC[MuJoCo]

• GAIL[MuJoCo]

• TD3[MuJoCo]

• PPO[To be added]

These algorithms are tested on benchmark tasks, and the results can be found in the images folder.

Usage

Example scripts are provided in the scripts folder.

Reference

• baselines(https://github.com/openai/baselines)

• slbo(https://github.com/facebookresearch/slbo)

• dac(https://github.com/google-research/google-research/tree/master/dac)