O2A: One-Shot Observational Learning with Action Vectors
We present O2A, a novel method for learning to perform robotic manipulation tasks from a single (one-shot) third-person demonstration video. To our knowledge, it is the first time this has been done for a single demonstration. The key novelty lies in pre-training a feature extractor for creating a p...
Main Authors: | , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
Frontiers Media S.A.
2021-08-01
|
Series: | Frontiers in Robotics and AI |
Subjects: | |
Online Access: | https://www.frontiersin.org/articles/10.3389/frobt.2021.686368/full |
id |
doaj-665c71eac570445f8a03a3e3f6cfdd1e |
---|---|
record_format |
Article |
spelling |
doaj-665c71eac570445f8a03a3e3f6cfdd1e2021-08-02T08:11:14ZengFrontiers Media S.A.Frontiers in Robotics and AI2296-91442021-08-01810.3389/frobt.2021.686368686368O2A: One-Shot Observational Learning with Action VectorsLeo Pauly0Wisdom C. Agboh 1David C. Hogg 2Raul Fuentes 3University of Leeds, Leeds, United KingdomUniversity of Leeds, Leeds, United KingdomUniversity of Leeds, Leeds, United KingdomRWTH Aachen University, Aachen, GermanyWe present O2A, a novel method for learning to perform robotic manipulation tasks from a single (one-shot) third-person demonstration video. To our knowledge, it is the first time this has been done for a single demonstration. The key novelty lies in pre-training a feature extractor for creating a perceptual representation for actions that we call “action vectors”. The action vectors are extracted using a 3D-CNN model pre-trained as an action classifier on a generic action dataset. The distance between the action vectors from the observed third-person demonstration and trial robot executions is used as a reward for reinforcement learning of the demonstrated task. We report on experiments in simulation and on a real robot, with changes in viewpoint of observation, properties of the objects involved, scene background and morphology of the manipulator between the demonstration and the learning domains. O2A outperforms baseline approaches under different domain shifts and has comparable performance with an Oracle (that uses an ideal reward function). Videos of the results, including demonstrations, can be found in our: project-website.https://www.frontiersin.org/articles/10.3389/frobt.2021.686368/fullobservational learningvisual perceptionreinforcement learningtransfer learningrobotic manipulation |
collection |
DOAJ |
language |
English |
format |
Article |
sources |
DOAJ |
author |
Leo Pauly Wisdom C. Agboh David C. Hogg Raul Fuentes |
spellingShingle |
Leo Pauly Wisdom C. Agboh David C. Hogg Raul Fuentes O2A: One-Shot Observational Learning with Action Vectors Frontiers in Robotics and AI observational learning visual perception reinforcement learning transfer learning robotic manipulation |
author_facet |
Leo Pauly Wisdom C. Agboh David C. Hogg Raul Fuentes |
author_sort |
Leo Pauly |
title |
O2A: One-Shot Observational Learning with Action Vectors |
title_short |
O2A: One-Shot Observational Learning with Action Vectors |
title_full |
O2A: One-Shot Observational Learning with Action Vectors |
title_fullStr |
O2A: One-Shot Observational Learning with Action Vectors |
title_full_unstemmed |
O2A: One-Shot Observational Learning with Action Vectors |
title_sort |
o2a: one-shot observational learning with action vectors |
publisher |
Frontiers Media S.A. |
series |
Frontiers in Robotics and AI |
issn |
2296-9144 |
publishDate |
2021-08-01 |
description |
We present O2A, a novel method for learning to perform robotic manipulation tasks from a single (one-shot) third-person demonstration video. To our knowledge, it is the first time this has been done for a single demonstration. The key novelty lies in pre-training a feature extractor for creating a perceptual representation for actions that we call “action vectors”. The action vectors are extracted using a 3D-CNN model pre-trained as an action classifier on a generic action dataset. The distance between the action vectors from the observed third-person demonstration and trial robot executions is used as a reward for reinforcement learning of the demonstrated task. We report on experiments in simulation and on a real robot, with changes in viewpoint of observation, properties of the objects involved, scene background and morphology of the manipulator between the demonstration and the learning domains. O2A outperforms baseline approaches under different domain shifts and has comparable performance with an Oracle (that uses an ideal reward function). Videos of the results, including demonstrations, can be found in our: project-website. |
topic |
observational learning visual perception reinforcement learning transfer learning robotic manipulation |
url |
https://www.frontiersin.org/articles/10.3389/frobt.2021.686368/full |
work_keys_str_mv |
AT leopauly o2aoneshotobservationallearningwithactionvectors AT wisdomcagboh o2aoneshotobservationallearningwithactionvectors AT davidchogg o2aoneshotobservationallearningwithactionvectors AT raulfuentes o2aoneshotobservationallearningwithactionvectors |
_version_ |
1721238711124885504 |