Robotic grasping and manipulation using a soft hand with two degrees of actuation : Mapping from parallel plates grasps to soft hand grasps

Robotics is taking a more important role every day in society. According to the Robot Institute of America [1], a robot is defined as “a reprogrammable, multifunctional manipulator designed to move material, parts, tools, or specialized devices, through variable programmed motions for the performanc...

Descripción completa

Detalles Bibliográficos
Autor: Amo Grau, Oscar
Tipo de recurso: tesis de maestría
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/398415
Acceso en línea:https://hdl.handle.net/2117/398415
Access Level:acceso abierto
Palabra clave:Manipulators (Mechanism) -- Design and construction
Manipulators (Mechanism) -- Control systems -- Software
Pattern recognition systems -- Software -- Mathematical models -- Disseny i construcció
Manipuladors (Mecanismes) -- Disseny i construcció
Manipuladors (Mecanismes) -- Sistemes de control -- Programari
Reconeixement de formes (Informàtica) -- Models matemàtics -- Disseny i construcció
Àrees temàtiques de la UPC::Informàtica::Robòtica
Descripción
Sumario:Robotics is taking a more important role every day in society. According to the Robot Institute of America [1], a robot is defined as “a reprogrammable, multifunctional manipulator designed to move material, parts, tools, or specialized devices, through variable programmed motions for the performance of a variety of tasks.”. To perform complex tasks, a robot typically makes use of lower level skills, the combination of which make the task execution possible. This project will focus on using two skills, robot manipulation, and object detection, to perform a higher level task, which is robotic grasping. Robotic grasping is defined by [2] as "a fundamental component of robotic manipulation that focuses on obtaining complete control of an object’s motion". To achieve this task, a combina- tion of object detection and object manipulation is used. Object detection is used to detect the object in the scene and to estimate a valid position for the end-effector (EEF) to grasp it. The second step is robot manipulation, which makes use of robot motio n to reach the desired grasping position and to interact with the target grasped object. Research in the field of robotic grasping, is abundant and several techniques have been used over the years. Nowadays, most research efforts [3, 4] use deep learning (DL) techniques to obtain valid grasps. These techniques are usually designed to perform grasps using parallel plate EEFs. This project intends to use DL algorithms, and adapt them, to use an anthropo- morphic soft hand gripper with two Degrees of Actuation (DoA) instead of a parallel plate EFF, and make a comparison on the grasping results between both EEF types. A soft hand provides several advantages compared to other EEF types: 1. Enables the robot to perform object manipulation tasks which would be impossible to perform with common parallel plate grippers, such as manipulating the object while mantaining a stable grip [5]. 2. Enables the robot to execute power as well as precision grasps. 1 3. Thanks to the adaptability of the hand, it allows to execute grasps in which the hand is in contact with other objects [6], which enables a larger number of valid grasps. 4. The flexibility and adaptability to objects of the soft hand compared to other types of EFFs, will increase the contact surface with the grasped object