433 0

Reinforcement Learning for Door Opening Task of Robotic Manipulator based on Images captured from External Camera

Title
Reinforcement Learning for Door Opening Task of Robotic Manipulator based on Images captured from External Camera
Other Titles
외부 카메라 이미지 기반 로봇 문 열기 작업의 강화 학습
Author
조서남
Alternative Author(s)
조서남
Advisor(s)
최영진
Issue Date
2021. 8
Publisher
한양대학교
Degree
Master
Abstract
Current image-based deep reinforcement learning methods have recently become possible to learn to manipulate complex tasks in life for a robot. In this work, the door opening task of robotic manipulator was studied via SAC(Soft Actor Critic), CURL(Contrastive Unsupervised Representations for Reinforcement Learning), and Imitation learning. With an in-front fixed RGB_D camera and human demonstration by one time, the agent of reinforcement learning in an experiment of MuJoCo simulation platform can show the capability to learn action planning for door opening task of robotic manipulator after training only 30 episodes in one hour.|최근 우리의 삶에서 로봇을 이용한 복잡한 조작이 이미지기반 강화학습 방법으로 가능해져 왔다.본 논문에서는 SAC, CURL, 모방 학습 방식으로 로봇 매니퓰레이터의 문 여는 작업을 수행했다. MuJoCo 시뮬레이션 환경 상에서 RGB-D카메라를 로봇의 앞쪽에 고정해두고 사람이 한 번의 행동을 보여주고 이 행동을 학습하였다. 한시간 동안 30 에피소드로 구성하여 학습되었으며 로봇 매니퓰레이터가 문을 여는 작업을 수행할 수 있음을 확인했다
URI
http://hanyang.dcollection.net/common/orgView/200000497330https://repository.hanyang.ac.kr/handle/20.500.11754/163643
Appears in Collections:
GRADUATE SCHOOL[S](대학원) > DEPARTMENT OF ELECTRICAL AND ELECTRONIC ENGINEERING(전자공학과) > Theses (Master)
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE