Accelerating Reinforcement Learning with Value-Conditional State Entropy Exploration (VCSE)

This repository contains the implementation of various algorithms used in our paper, which can be found at ArXiv.

Algorithms

The repository is organized into several folders, each containing the implementation of a specific algorithm:

• VCSE_A2C/: Contains the implementation of the A2C algorithm.
• VCSE_DrQv2/: Contains the implementation of the DrQv2 algorithm.
• VCSE_MWM/: Contains the implementation of the MWM algorithm.
• VCSE_SAC/: Contains the implementation of the SAC algorithm.

If you want to refer this research, please cite:

@article{kim2023accelerating,
  title={Accelerating Reinforcement Learning with Value-Conditional State Entropy Exploration},
  author={Kim, Dongyoung and Shin, Jinwoo and Abbeel, Pieter and Seo, Younggyo},
  journal={arXiv preprint arXiv:2305.19476},
  year={2023}
}

VCSE + A2C

Our code is built on top of the RE3 + A2C implementation RE3. That trainning code can be found in rl-starter-files directory. Which is fork from rl-starter-files and A2C implementation is fork from torch-ac

Refer to the individual README files in VCSE_A2C for details of installation and instructions.

VCSE + DrQv2

Our code is built on top of the drqv2 repository.

Refer to the individual README files in VCSE_DrQv2 for details of installation and instructions.

VCSE + MWM

Our code is built on top of the MWM repository.

Refer to the individual README files in VCSE_MWM for details of installation and instructions.

VCSE + SAC

Our code is built on top of the pytorch_sac repository.

Refer to the individual README files in VCSE_SAC for details of installation and instructions.