A Learning from demonstration approach for robot trajectories through motion-sensing human demonstrations

View/Open
Cita com:
hdl:2117/192255
Document typeMaster thesis
Date2020-04-20
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial
property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public
communication or transformation of this work are prohibited without permission of the copyright holder
Abstract
The objective of this thesis is to teach a Baxter robot to learn
certain arm trajectories. The robot must be capable of generalizing
the primitive movement of the trajectory to new unseen
poses. The thesis is framed within a robotized kitchen project
with aims to help people with mobility problems. To solve this
problem end, a human will record demonstrations, which will
be translated to the robots’ morphology using an Inverse Kinematics
(IK) module. For the learning part Dynamic Movement
Primitives (DMP) will be used, due to their capability to take
profit of human experience. The proposed system works in the
majority of the scenarios, but, it would be expected to behave
better when generalizing to new orientations of the arm. However
a proposal has been suggested to correct this issue.
SubjectsKinect (Programmable controller), Robotics, Computer vision, Kinect (Controlador programable), Robòtica, Visió per ordinador
DegreeMÀSTER UNIVERSITARI EN INTEL·LIGÈNCIA ARTIFICIAL (Pla 2017)
Collections
Files | Description | Size | Format | View |
---|---|---|---|---|
147834.pdf | 4,007Mb | View/Open |