Development of perception module for bobotic manipulation tasks

Robots performing manipulation tasks require the accurate location and orientation of an object in space. Previously, at the Robotics Laboratory of IOC-UPC this data has been generated artificially. In order to automate the process, a perception module has been developed for providing task and motio...

Full description

Bibliographic Details
Author: Gazikalović, Nikola
Format: master thesis
Publication Date:2022
Country:España
Institution:Universitat Politècnica de Catalunya (UPC)
Repository:UPCommons. Portal del coneixement obert de la UPC
Language:English
OAI Identifier:oai:upcommons.upc.edu:2117/372657
Online Access:https://hdl.handle.net/2117/372657
Access Level:Open access
Keyword:Robot vision -- Mathematical models -- Software
Computer vision -- Calibration -- Testing
Visió artificial (Robòtica) -- Models matemàtics -- Programari
Visió per ordinador -- Calibratge -- Proves
Àrees temàtiques de la UPC::Informàtica::Robòtica
Description
Summary:Robots performing manipulation tasks require the accurate location and orientation of an object in space. Previously, at the Robotics Laboratory of IOC-UPC this data has been generated artificially. In order to automate the process, a perception module has been developed for providing task and motion planners with the localization and pose estimation of objects used in robot manipulation tasks. The Robot Operating System provided a great framework for incorporating vision provided by Microsoft Kinect V2 sensors and the presentation of obtained data to be used in the generation of Planning Domain Definition Language files, which define a robots environment. Localization and pose estimation was done using fiducial markers along with studying possible enhancements using deep learning methods. Perfectly calibrating hardware and setting up a system play a big role in enhancing perception accuracy and while fiducial markers provide a simple and robust solution in laboratory conditions, real world applications with varying lighting, viewing angles and partial occlusions should rely on AI vision