rg

rl-games

RL implementations

Showing:

Popularity

Downloads/wk

0

GitHub Stars

101

Maintenance

Last Commit

7d ago

Contributors

6

Package

Dependencies

11

License

MIT

Categories

Readme

RL Games: High performance RL library

Some results on interesting environments

Ant_running Humanoid_running

Allegro_Hand_400 Shadow_Hand_OpenAI

Config file

Implemented in Pytorch:

  • PPO with the support of asymmetric actor-critic variant

  • Support of end-to-end GPU accelerated training pipeline with Isaac Gym and Brax

  • Masked actions support

  • Multi-agent training, decentralized and centralized critic variants

  • Self-play

    Implemented in Tensorflow 1.x (not updates now):

  • Rainbow DQN

  • A2C

  • PPO

Installation

For maximum training performance a preliminary installation of Pytorch 1.9+ with CUDA 11.1 is highly recommended:

conda install pytorch torchvision cudatoolkit=11.1 -c pytorch -c nvidia or: pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 -f https://download.pytorch.org/whl/torch_stable.htm

Then:

pip install rl-games

Training

NVIDIA Isaac Gym

Download and follow the installation instructions from https://developer.nvidia.com/isaac-gym
Run from python/rlgpu directory:

Ant
python rlg_train.py --task Ant --headless
python rlg_train.py --task Ant --play --checkpoint nn/Ant.pth --num_envs 100

Humanoid
python rlg_train.py --task Humanoid --headless
python rlg_train.py --task Humanoid --play --checkpoint nn/Humanoid.pth --num_envs 100

Shadow Hand block orientation task
python rlg_train.py --task ShadowHand --headless
python rlg_train.py --task ShadowHand --play --checkpoint nn/ShadowHand.pth --num_envs 100

Atari Pong
python runner.py --train --file rl_games/configs/atari/ppo_pong.yaml
python runner.py --play --file rl_games/configs/atari/ppo_pong.yaml --checkpoint nn/PongNoFrameskip.pth

Brax Ant
python runner.py --train --file rl_games/configs/brax/ppo_ant.yaml
python runner.py --play --file rl_games/configs/atari/ppo_ant.yaml --checkpoint nn/Ant_brax.pth

Release Notes

1.1.0

  • Added to pypi: pip install rl-games
  • Added reporting env (sim) step fps, without policy inference. Improved naming.
  • Renames in yaml config for better readability: steps_num to horizon_length amd lr_threshold to kl_threshold

Troubleshouting

  • Some of the supported envs are not installed with setup.py, you need to manually install them
  • Starting from rl-games 1.1.0 old yaml configs won't be compatible with the new version:
    • steps_num should be changed to horizon_length amd lr_threshold to kl_threshold

Rate & Review

Great Documentation0
Easy to Use0
Performant0
Highly Customizable0
Bleeding Edge0
Responsive Maintainers0
Poor Documentation0
Hard to Use0
Slow0
Buggy0
Abandoned0
Unwelcoming Community0
100