• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.55.2021.tde-30082021-100712
Document
Author
Full name
Felipe Padula Sanches
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Carlos, 2021
Supervisor
Committee
Romero, Roseli Aparecida Francelin (President)
Grassi Junior, Valdir
Ramos, Josue Junior Guimarães
Wolf, Denis Fernando
Title in English
End-to-End Visual Obstacle Avoidance for a Robotic Manipulator using Deep Reinforcement Learning
Keywords in English
Deep reinforcement learning
Motion control
Obstacle avoidance
Robot manipulators
Robot vision
Abstract in English
Recent changes in industrial paradigms enforce that robots must be intelligent and capable of decision-making. Robotic manipulators need to satisfy many requirements for operating properly. Perhaps the most fundamental one is the capability of operating in its environment without collisions. In this work, we perform visual obstacle avoidance on goal-reaching tasks of a robotic manipulator using an end-to-end Deep Reinforcement Learning model. The motion control policy is responsible for reaching a target position while at the same time avoiding an obstacle positioned randomly in the scene. This policy uses vision and proprioceptive sensor data to operate. We train the reinforcement learning agent using Twin-Delayed DDPG (TD3) algorithm in a simulated environment, utilizing the Unity game engine and the ML-Agents toolkit. Experiments demonstrate that the agent can successfully learn a meaningful policy to avoid obstacles using images.
Title in Portuguese
Desvio de Obstáculo para um Manipulador Robótico utilizando Visão e Aprendizado por Reforço Profundo Ponta-a-Ponta
Keywords in Portuguese
Aprendizado por reforço profundo
Controle de movimento
Desvio de obstáculos
Manipuladores robóticos
Visão robótica
Abstract in Portuguese
Mudanças recentes nos paradigmas industriais esperam que os robôs sejam inteligentes e capazes de tomar decisões. Os manipuladores robóticos precisam satisfazer muitos requisitos para operar adequadamente. Talvez o mais fundamental seja a capacidade de operar em seu ambiente sem colisões. Neste trabalho, evitamos obstáculos visuais em tarefas de alcance de meta de um manipulador robótico usando um modelo de Aprendizado por Reforço Profundo de ponta-aponta. A política de controle de movimento é responsável por atingir uma posição alvo e, ao mesmo tempo, evitar um obstáculo posicionado aleatoriamente na cena. Esta política usa dados de sensores proprioceptivos e de visão para operar. O agente de aprendizagem por reforço foi treinado através do algoritmo Twin-Delayed DDPG (TD3) em um ambiente simulado, utilizando a game engine Unity e o framework ML-Agents. Experimentos demonstram que o agente pode aprender com sucesso uma política significativa para evitar obstáculos usando imagens.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2021-08-30
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.