Quadruped robot locomotion in unknown terrain using deep reinforcement learning

Muleilan Pei; Dongping Wu; Changhong Wang

doi:10.1109/ICUS50048.2020.9274920

Quadruped robot locomotion in unknown terrain using deep reinforcement learning

Muleilan Pei
, Dongping Wu
, Changhong Wang^*

^*Corresponding author for this work

School of Astronautics

Harbin Institute of Technology

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

6 Scopus citations

Abstract

This paper is concerned with locomotion problems for the quadruped robot in unknown and unstructured terrains, utilizing the emerging deep reinforcement learning technique. The state-of-the-art deep deterministic policy gradient (DDPG) algorithm is leveraged to acquire the gait policy by learning from interactions with the environment. The framework of learning system is presented based on the actor-critic architecture, and the additional domain-specific knowledge is exploited for shaping the reward function to enhance the learning efficiency during training. Moreover, a simulation study is implemented to validate the performance of the proposed DDPG-based controller.

Original language	English
Title of host publication	Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	517-522
Number of pages	6
ISBN (Electronic)	9781728180250
DOIs	https://doi.org/10.1109/ICUS50048.2020.9274920
State	Published - 27 Nov 2020
Event	3rd International Conference on Unmanned Systems, ICUS 2020 - Harbin, China Duration: 27 Nov 2020 → 28 Nov 2020

Publication series

Name	Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020

Conference

Conference	3rd International Conference on Unmanned Systems, ICUS 2020
Country/Territory	China
City	Harbin
Period	27/11/20 → 28/11/20

Keywords

Deep deterministic policy gradient
Deep reinforcement learning
Quadruped robot locomotion
Unknown terrain

Access to Document

10.1109/ICUS50048.2020.9274920

Cite this

Pei, M., Wu, D., & Wang, C. (2020). Quadruped robot locomotion in unknown terrain using deep reinforcement learning. In Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020 (pp. 517-522). Article 9274920 (Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICUS50048.2020.9274920

@inproceedings{ed17899814ab4c1083704f0044fa1a47,

title = "Quadruped robot locomotion in unknown terrain using deep reinforcement learning",

abstract = "This paper is concerned with locomotion problems for the quadruped robot in unknown and unstructured terrains, utilizing the emerging deep reinforcement learning technique. The state-of-the-art deep deterministic policy gradient (DDPG) algorithm is leveraged to acquire the gait policy by learning from interactions with the environment. The framework of learning system is presented based on the actor-critic architecture, and the additional domain-specific knowledge is exploited for shaping the reward function to enhance the learning efficiency during training. Moreover, a simulation study is implemented to validate the performance of the proposed DDPG-based controller.",

keywords = "Deep deterministic policy gradient, Deep reinforcement learning, Quadruped robot locomotion, Unknown terrain",

author = "Muleilan Pei and Dongping Wu and Changhong Wang",

note = "Publisher Copyright: {\textcopyright} 2020 IEEE.; 3rd International Conference on Unmanned Systems, ICUS 2020 ; Conference date: 27-11-2020 Through 28-11-2020",

year = "2020",

month = nov,

day = "27",

doi = "10.1109/ICUS50048.2020.9274920",

language = "英语",

series = "Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "517--522",

booktitle = "Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020",

address = "美国",

}

Pei, M, Wu, D & Wang, C 2020, Quadruped robot locomotion in unknown terrain using deep reinforcement learning. in Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020., 9274920, Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020, Institute of Electrical and Electronics Engineers Inc., pp. 517-522, 3rd International Conference on Unmanned Systems, ICUS 2020, Harbin, China, 27/11/20. https://doi.org/10.1109/ICUS50048.2020.9274920

Quadruped robot locomotion in unknown terrain using deep reinforcement learning. / Pei, Muleilan; Wu, Dongping; Wang, Changhong.
Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020. Institute of Electrical and Electronics Engineers Inc., 2020. p. 517-522 9274920 (Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Quadruped robot locomotion in unknown terrain using deep reinforcement learning

AU - Pei, Muleilan

AU - Wu, Dongping

AU - Wang, Changhong

PY - 2020/11/27

Y1 - 2020/11/27

N2 - This paper is concerned with locomotion problems for the quadruped robot in unknown and unstructured terrains, utilizing the emerging deep reinforcement learning technique. The state-of-the-art deep deterministic policy gradient (DDPG) algorithm is leveraged to acquire the gait policy by learning from interactions with the environment. The framework of learning system is presented based on the actor-critic architecture, and the additional domain-specific knowledge is exploited for shaping the reward function to enhance the learning efficiency during training. Moreover, a simulation study is implemented to validate the performance of the proposed DDPG-based controller.

AB - This paper is concerned with locomotion problems for the quadruped robot in unknown and unstructured terrains, utilizing the emerging deep reinforcement learning technique. The state-of-the-art deep deterministic policy gradient (DDPG) algorithm is leveraged to acquire the gait policy by learning from interactions with the environment. The framework of learning system is presented based on the actor-critic architecture, and the additional domain-specific knowledge is exploited for shaping the reward function to enhance the learning efficiency during training. Moreover, a simulation study is implemented to validate the performance of the proposed DDPG-based controller.

KW - Deep deterministic policy gradient

KW - Deep reinforcement learning

KW - Quadruped robot locomotion

KW - Unknown terrain

UR - https://www.scopus.com/pages/publications/85098934583

U2 - 10.1109/ICUS50048.2020.9274920

DO - 10.1109/ICUS50048.2020.9274920

M3 - 会议稿件

AN - SCOPUS:85098934583

T3 - Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020

SP - 517

EP - 522

BT - Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 3rd International Conference on Unmanned Systems, ICUS 2020

Y2 - 27 November 2020 through 28 November 2020

ER -

Pei M, Wu D, Wang C. Quadruped robot locomotion in unknown terrain using deep reinforcement learning. In Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020. Institute of Electrical and Electronics Engineers Inc. 2020. p. 517-522. 9274920. (Proceedings of 2020 3rd International Conference on Unmanned Systems, ICUS 2020). doi: 10.1109/ICUS50048.2020.9274920

Quadruped robot locomotion in unknown terrain using deep reinforcement learning

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this