Adversarial discriminative sim-to-real transfer of visuo-motor policies

Fangyi Zhang, Jürgen Leitner, Zongyuan Ge, Michael Milford, Peter Corke

Research output: Contribution to journalArticleResearchpeer-review

2 Citations (Scopus)

Abstract

Various approaches have been proposed to learn visuo-motor policies for real-world robotic applications. One solution is first learning in simulation then transferring to the real world. In the transfer, most existing approaches need real-world images with labels. However, the labeling process is often expensive or even impractical in many robotic applications. In this article, we introduce an adversarial discriminative sim-to-real transfer approach to reduce the amount of labeled real data required. The effectiveness of the approach is demonstrated with modular networks in a table-top object-reaching task where a seven-degree-of-freedom arm is controlled in velocity mode to reach a blue cuboid in clutter through visual observations from a monocular RGB camera. The adversarial transfer approach reduced the labeled real data requirement by 50%. Policies can be transferred to real environments with only 93 labeled and 186 unlabeled real images. The transferred visuo-motor policies are robust to novel (not seen in training) objects in clutter and even a moving target, achieving a 97.8% success rate and 1.8 cm control accuracy. Datasets and code are openly available.

Original languageEnglish
Pages (from-to)1229-1245
Number of pages17
JournalInternational Journal of Robotics Research
Volume38
Issue number10-11
DOIs
Publication statusPublished - Sep 2019

Keywords

  • adversarial transfer learning
  • domain adaptation
  • robotic reaching
  • Sim-to-real transfer
  • visuo-motor policy learning

Cite this

Zhang, Fangyi ; Leitner, Jürgen ; Ge, Zongyuan ; Milford, Michael ; Corke, Peter. / Adversarial discriminative sim-to-real transfer of visuo-motor policies. In: International Journal of Robotics Research. 2019 ; Vol. 38, No. 10-11. pp. 1229-1245.
@article{a907e0c5a53d4a09b3a021e7ecd767a7,
title = "Adversarial discriminative sim-to-real transfer of visuo-motor policies",
abstract = "Various approaches have been proposed to learn visuo-motor policies for real-world robotic applications. One solution is first learning in simulation then transferring to the real world. In the transfer, most existing approaches need real-world images with labels. However, the labeling process is often expensive or even impractical in many robotic applications. In this article, we introduce an adversarial discriminative sim-to-real transfer approach to reduce the amount of labeled real data required. The effectiveness of the approach is demonstrated with modular networks in a table-top object-reaching task where a seven-degree-of-freedom arm is controlled in velocity mode to reach a blue cuboid in clutter through visual observations from a monocular RGB camera. The adversarial transfer approach reduced the labeled real data requirement by 50{\%}. Policies can be transferred to real environments with only 93 labeled and 186 unlabeled real images. The transferred visuo-motor policies are robust to novel (not seen in training) objects in clutter and even a moving target, achieving a 97.8{\%} success rate and 1.8 cm control accuracy. Datasets and code are openly available.",
keywords = "adversarial transfer learning, domain adaptation, robotic reaching, Sim-to-real transfer, visuo-motor policy learning",
author = "Fangyi Zhang and J{\"u}rgen Leitner and Zongyuan Ge and Michael Milford and Peter Corke",
year = "2019",
month = "9",
doi = "10.1177/0278364919870227",
language = "English",
volume = "38",
pages = "1229--1245",
journal = "International Journal of Robotics Research",
issn = "0278-3649",
publisher = "SAGE Publications Ltd",
number = "10-11",

}

Adversarial discriminative sim-to-real transfer of visuo-motor policies. / Zhang, Fangyi; Leitner, Jürgen; Ge, Zongyuan; Milford, Michael; Corke, Peter.

In: International Journal of Robotics Research, Vol. 38, No. 10-11, 09.2019, p. 1229-1245.

Research output: Contribution to journalArticleResearchpeer-review

TY - JOUR

T1 - Adversarial discriminative sim-to-real transfer of visuo-motor policies

AU - Zhang, Fangyi

AU - Leitner, Jürgen

AU - Ge, Zongyuan

AU - Milford, Michael

AU - Corke, Peter

PY - 2019/9

Y1 - 2019/9

N2 - Various approaches have been proposed to learn visuo-motor policies for real-world robotic applications. One solution is first learning in simulation then transferring to the real world. In the transfer, most existing approaches need real-world images with labels. However, the labeling process is often expensive or even impractical in many robotic applications. In this article, we introduce an adversarial discriminative sim-to-real transfer approach to reduce the amount of labeled real data required. The effectiveness of the approach is demonstrated with modular networks in a table-top object-reaching task where a seven-degree-of-freedom arm is controlled in velocity mode to reach a blue cuboid in clutter through visual observations from a monocular RGB camera. The adversarial transfer approach reduced the labeled real data requirement by 50%. Policies can be transferred to real environments with only 93 labeled and 186 unlabeled real images. The transferred visuo-motor policies are robust to novel (not seen in training) objects in clutter and even a moving target, achieving a 97.8% success rate and 1.8 cm control accuracy. Datasets and code are openly available.

AB - Various approaches have been proposed to learn visuo-motor policies for real-world robotic applications. One solution is first learning in simulation then transferring to the real world. In the transfer, most existing approaches need real-world images with labels. However, the labeling process is often expensive or even impractical in many robotic applications. In this article, we introduce an adversarial discriminative sim-to-real transfer approach to reduce the amount of labeled real data required. The effectiveness of the approach is demonstrated with modular networks in a table-top object-reaching task where a seven-degree-of-freedom arm is controlled in velocity mode to reach a blue cuboid in clutter through visual observations from a monocular RGB camera. The adversarial transfer approach reduced the labeled real data requirement by 50%. Policies can be transferred to real environments with only 93 labeled and 186 unlabeled real images. The transferred visuo-motor policies are robust to novel (not seen in training) objects in clutter and even a moving target, achieving a 97.8% success rate and 1.8 cm control accuracy. Datasets and code are openly available.

KW - adversarial transfer learning

KW - domain adaptation

KW - robotic reaching

KW - Sim-to-real transfer

KW - visuo-motor policy learning

UR - http://www.scopus.com/inward/record.url?scp=85071511672&partnerID=8YFLogxK

U2 - 10.1177/0278364919870227

DO - 10.1177/0278364919870227

M3 - Article

AN - SCOPUS:85071511672

VL - 38

SP - 1229

EP - 1245

JO - International Journal of Robotics Research

JF - International Journal of Robotics Research

SN - 0278-3649

IS - 10-11

ER -