Skip to content

Latest commit

 

History

History
88 lines (68 loc) · 2.36 KB

README.md

File metadata and controls

88 lines (68 loc) · 2.36 KB
Raw

Reinforcement Learning with Augmented Data (RAD): ProcGen

Official codebase for Reinforcement Learning with Augmented Data on Procgen Benchmark. This codebase was originally forked from Procgen. Official codebases for DM control and OpenAI Gym are available at RAD: DM control RAD: OpenAI Gym.

BibTex

@article{laskin2020reinforcement,
  title={Reinforcement learning with augmented data},
  author={Laskin, Michael and Lee, Kimin and Stooke, Adam and Pinto, Lerrel and Abbeel, Pieter and Srinivas, Aravind},
  journal={arXiv preprint arXiv:2004.14990},
  year={2020}
}

Install

You can get miniconda from https://docs.conda.io/en/latest/miniconda.html if you don't have it, or install the dependencies from environment.yml manually.

git clone https://github.com/openai/train-procgen.git
conda env update --name train-procgen --file train-procgen/environment.yml
conda activate train-procgen

got to Procgen_Envs

pip install -e .

comback to Procgen

pip install https://github.com/openai/baselines/archive/9ee399f5b20cd70ac0a871927a6cf043b478193f.zip
pip install -e train-procgen
pip uninstall tensorflow
conda install -n train-procgen tensorflow-gpu=1.15 cudatoolkit=10.0
pip install torch matplotlib scikit-image

change NeurIPS2020_Procgen_Envs/procgen/src/game.h for random crop

Try it out

Pixel PPO on the environment StarPilot:

./scripts/train_normal.sh starpilot

PPO + RAD (crop) on the environment StarPilot:

./scripts/train.sh starpilot crop

PPO + RAD (flip) on the environment StarPilot:

./scripts/train.sh starpilot flip

PPO + RAD (color_jitter) on the environment StarPilot:

./scripts/train.sh starpilot color_jitter

PPO + RAD (rotate) on the environment StarPilot:

./scripts/train.sh starpilot rotate

PPO + RAD (cutout_color) on the environment StarPilot:

./scripts/train.sh starpilot cutout_color

PPO + RAD (cutout) on the environment StarPilot:

./scripts/train.sh starpilot cutout

PPO + RAD (gray) on the environment StarPilot:

./scripts/train.sh starpilot gray

PPO + RAD (random conv) on the environment StarPilot:

./scripts/train_random.sh starpilot