noisyenv: Simple Noisy Environment Augmentation for Reinforcement Learning

This package contains a set of generic wrappers designed to augment RL environments with noise and encourage agent exploration and improve training data diversity which are applicable to a broad spectrum of RL algorithms and environments. For more details, please refer to our paper: https://arxiv.org/abs/2305.02882.

Note that this package has been developed for the new step and reset API introduced in OpenAI Gym v26 and Gymnasium v26. Use the gymnasium.wrappers.EnvCompatibility wrapper to update old environments for compatibility.

Installation

pip install noisyenv

Usage

import gymnasium as gym
from noisyenv.wrappers import RandomUniformScaleReward

base_env = gym.make("HalfCheetah-v2")
env = RandomUniformScaleReward(env=base_env, noise_rate=0.01, low=0.9, high=1.1)

# And just use as you would normally
observation, info = env.reset(seed=333)

for _ in range(100):
    action = env.action_space.sample()
    observation, reward, terminated, truncated, info = env.step(action)

    if terminated or truncated:
        observation, info = env.reset()
env.close()

Citing noisyenv

If you use noisyenv in your work, please cite our paper:

@misc{khraishi2023simple,
      title={Simple Noisy Environment Augmentation for Reinforcement Learning}, 
      author={Raad Khraishi and Ramin Okhrati},
      year={2023},
      eprint={2305.02882},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}