Emergence of Human-comparable Balancing Behaviors by Deep Reinforcement Learning

Chuanyu Yang; Taku Komura; Zhibin Li

doi:10.1109/HUMANOIDS.2017.8246900

Emergence of Human-comparable Balancing Behaviors by Deep Reinforcement Learning

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

Abstract / Description of output

This paper presents a hierarchical framework based on deep reinforcement learning that learns a diversity of policies for humanoid balance control. conventional zero moment point based controllers perform limited actions during under-actuation, whereas the proposed framework can perform human-like balancing behaviors such as active push-off of ankles. The learning is done through the design of an explainable reward based on physical constraints. The simulated results are presented and analyzed. The successful emergence of humanlike behaviors through deep reinforcement learning proves the feasibility of using an AI-based approach for learning humanoid balancing control in a unified framework.

Original language	English
Title of host publication	2017 IEEE-RAS International Conference on Humanoid Robots
Publisher	Institute of Electrical and Electronics Engineers
Number of pages	8
ISBN (Electronic)	978-1-5386-4678-6
ISBN (Print)	978-1-5386-4679-3
DOIs	https://doi.org/10.1109/HUMANOIDS.2017.8246900
Publication status	Published - 8 Jan 2018
Event	IEEE-RAS International Conference on Humanoid Robots - The Birmingham Repertory Theatre, Centenary Square, Broad Street, Birmingham, United Kingdom Duration: 15 Nov 2017 → 17 Nov 2017 http://humanoids2017.loria.fr/

Publication series

Name
Publisher	IEEE
ISSN (Electronic)	2164-0580

Conference

Conference	IEEE-RAS International Conference on Humanoid Robots
Country/Territory	United Kingdom
City	Birmingham
Period	15/11/17 → 17/11/17
Internet address	http://humanoids2017.loria.fr/

Access to Document

10.1109/HUMANOIDS.2017.8246900

UniversityOpenAccessAccepted author manuscript, 1.33 MB

http://ieeexplore.ieee.org/document/8246900/

Zhibin (Alex) Li
- Institute of Perception, Action and Behaviour
Person: Academic: Research Active

Cite this

@inproceedings{f644416425b44481b270a30407351bd6,

title = "Emergence of Human-comparable Balancing Behaviors by Deep Reinforcement Learning",

abstract = "This paper presents a hierarchical framework based on deep reinforcement learning that learns a diversity of policies for humanoid balance control. conventional zero moment point based controllers perform limited actions during under-actuation, whereas the proposed framework can perform human-like balancing behaviors such as active push-off of ankles. The learning is done through the design of an explainable reward based on physical constraints. The simulated results are presented and analyzed. The successful emergence of humanlike behaviors through deep reinforcement learning proves the feasibility of using an AI-based approach for learning humanoid balancing control in a unified framework.",

author = "Chuanyu Yang and Taku Komura and Zhibin Li",

year = "2018",

month = jan,

day = "8",

doi = "10.1109/HUMANOIDS.2017.8246900",

language = "English",

isbn = "978-1-5386-4679-3",

publisher = "Institute of Electrical and Electronics Engineers",

booktitle = "2017 IEEE-RAS International Conference on Humanoid Robots",

address = "United States",

note = "IEEE-RAS International Conference on Humanoid Robots ; Conference date: 15-11-2017 Through 17-11-2017",

url = "http://humanoids2017.loria.fr/",

}

TY - GEN

T1 - Emergence of Human-comparable Balancing Behaviors by Deep Reinforcement Learning

AU - Yang, Chuanyu

AU - Komura, Taku

AU - Li, Zhibin

PY - 2018/1/8

Y1 - 2018/1/8

N2 - This paper presents a hierarchical framework based on deep reinforcement learning that learns a diversity of policies for humanoid balance control. conventional zero moment point based controllers perform limited actions during under-actuation, whereas the proposed framework can perform human-like balancing behaviors such as active push-off of ankles. The learning is done through the design of an explainable reward based on physical constraints. The simulated results are presented and analyzed. The successful emergence of humanlike behaviors through deep reinforcement learning proves the feasibility of using an AI-based approach for learning humanoid balancing control in a unified framework.

AB - This paper presents a hierarchical framework based on deep reinforcement learning that learns a diversity of policies for humanoid balance control. conventional zero moment point based controllers perform limited actions during under-actuation, whereas the proposed framework can perform human-like balancing behaviors such as active push-off of ankles. The learning is done through the design of an explainable reward based on physical constraints. The simulated results are presented and analyzed. The successful emergence of humanlike behaviors through deep reinforcement learning proves the feasibility of using an AI-based approach for learning humanoid balancing control in a unified framework.

U2 - 10.1109/HUMANOIDS.2017.8246900

DO - 10.1109/HUMANOIDS.2017.8246900

M3 - Conference contribution

SN - 978-1-5386-4679-3

BT - 2017 IEEE-RAS International Conference on Humanoid Robots

PB - Institute of Electrical and Electronics Engineers

T2 - IEEE-RAS International Conference on Humanoid Robots

Y2 - 15 November 2017 through 17 November 2017

ER -

Emergence of Human-comparable Balancing Behaviors by Deep Reinforcement Learning

Abstract / Description of output

Publication series

Conference

Access to Document

Fingerprint

Profiles

Zhibin (Alex) Li

Cite this