pokaxpoka / rad_openaigym

RAD: Reinforcement Learning with Augmented Data (code for state augmentation)

Reinforcement Learning with Augmented Data (RAD): State augmentation on Mujoco Envs

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

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

1. Install and use the included Ananconda environment

$ conda env create -f environment/linux-gpu-env.yml
$ source activate rlkit

You'll need to get your own MuJoCo key if you want to use MuJoCo.

2. Add this repo directory to your PYTHONPATH environment variable or simply run:

pip install -e .

3. Install "benchmarking MBRL",

pip uninstall gym
pip install gym==0.9.4 mujoco-py==0.5.7 termcolor
cd mbbl_envs
pip install --user -e .

Try it out

SAC

./scripts/run_sac.sh [env_name]

SAC + BN

./scripts/run_sac_bn.sh [env_name]

SAC + Gaussian

./scripts/run_sac_gaussian.sh [env_name] [prob] [std]

SAC + random amplitude scaling

./scripts/run_rand_ampl.sh [env_name] [single_flag: 0 or 1] [equal_flag: 0 or 1] [lower] [upper]