/Learning-Independent-SKills

Task dependent skill transformation is challenging due to the ignorance of the relationships between primitive skills. In this project, we propose a skill decomposition algorithm to learn independent skills, which are more suitable than primitive skills for task dependent skill transformation.

Primary LanguagePython

Learning Independent SKills

Task dependent skill transformation is challenging due to the ignorance of the relationships between primitive skills. In this project, we propose a skill decomposition algorithm to learn independent skills, which are more suitable than primitive skills for task dependent skill transformation.

Required Dependencies:

Pytorch 1.20

mujoco_py 2.0.2.8

Visualization of the practical skills for complex tasks:

image

image

image

image

Cite this Work

@inproceedings{10.1145/3366424.3383530,
author = {Qiangxing Tian, GuanchuWang, Jinxin Liu, Donglin Wang and Yachen Kang},
title = {Independent Skill Transfer for Deep Reinforcement Learning},
year = {2020},
series = {IJCAI '20}
}

Acknowledgements

Thanks Guangdong Bozhilin Inc. for their financial support, and thanks all of the contributors to this project!