Skip to content

Latest commit

 

History

History
34 lines (28 loc) · 1.03 KB

readme.md

File metadata and controls

34 lines (28 loc) · 1.03 KB

Simrl

Simrl stands for Simple Implementations of Reinforcement Learning. This repository is still under development. If you find any problem, feel free to open an issue.

Installation

git clone https://github.com/IcarusWizard/simrl.git
cd simrl 
pip install -e .

Usage

You can start training by calling the algorithm, i.e.

python -m simrl.algo.<type>.<name> --env <env_name>

Supported Algorithems

  • Model-based
    • PETS
    • MBPO
  • Model-free
    • TRPO
    • PPO
    • DQN
    • SAC

Supported Environments

The default installation only supports classical gym environments with flatten states. You can enable more supports with pip install -e .[env], which will install box2d and pybullet.

We all support dm_control, please see their repo for instruction of installation. Please not that for test convenience, the default backend is set as egl.

We are planning to add support for more diverse environments and POMDP environments with visual inputs.