About DoKS      NL  |  EN Search: Advanced Search
  Part of a word (e.g. tele*)    Exact wordgroup (e.g. "wireless communication")
 
Home
folder Authors
folder Departments
folder Help
folder Years
 
Most popular theses: 2014 2015 2016 2017 2018 2019


1,242 theses on-line.



Doks PHL
Doks XIOS



Open Archives Initiative
Home

Active Learning strategies on a Pepper robot

2019
Gabriels, Michiel
Professionele bachelor in de toegepaste informatica

Abstract :
The Austrian Research Institute for Artificial Intelligence (OFAI) does ongoing research on an incremental grounded language learning system that is integrated with a Pepper robot. This system has an active learning extension to help the learning process of the robot by having it ask for feedback in the form of non-verbal communication, like pointing. It will decide for itself to ask for feedback when it thinks it needs more information on certain actions or objects. This is inspired by research on child language acquisition, whereby a child will also use gestures like pointing to ask for new information whenever it is unsure about something. The basic model for this learning system is already established. However, there is still room for improvement on this existing model. By tweaking the parameters, more information could be gathered about whether or not the feedback given by a user should receive more weight than it is given now in the learning process of the robot. Other parameters include the time that is required to learn all the objects present in the scenario and whether or not to use vocalization like "uhm" to make it more clear to the user that the robot needs more feedback.

The thesis describes the methodology used for acquiring the information from these experiments. A hypothesis is formulated, based on the research done on the effects of a specific parameter, after which this hypothesis is tested on the existing system. This system uses the Robot Operating System (ROS) to facilitate the layer between the hardware of the Pepper robot and the software developed for multi-modal word learning. The Pepper robot has its own voice recognition software, but because this is not sufficient, Google’s cloud speech recognition software is used instead. The model uses the Action Verb Corpus made within OFAI, which has been developed using multi-modal data of twelve humans conducting in total 390 simple TAKE, PUT and PUSH actions.

The thesis also briefly touches on the subject of how a student that has no prior research experience integrates in an existing (international) research group working on a subject that is new to the student. This includes the student’s personal findings, how his social and soft skills have improved in an international setting, how this can be used during projects later in life and how prior experience in practical programming can be useful when integrating into a research group. It also includes the general workings of this particular research group and the cooperation with an adjacent research group at a technical university.

Full text:
File Size Type Checksum  
Eindwerk_Gabriels_Michiel.pdf 991 KB PDF MD5 Open file

Dit eindwerk werd 249 keer bekeken.
Translate to English (Google translate)
 

Show record details

Show ETD - Dublin Core

If you want to cite this thesis in your own thesis, paper, or report, use this format (APA):

Gabriels, M. (2019). Active Learning strategies on a Pepper robot. Unpublished thesis, Hogeschool PXL, PXL-Digital.
Retrieved from http://doks.pxl.be/doks/do/record/Get?dispatch=view&recordId=SEtd8ab2a8216cd3194f016cd3290d5d02a5.




©2004-2008 - Hogeschool PXL - webmaster - Contact - Disclaimer