MSc Project defence-Guðrún Inga Baldursdóttir

Non-Verbal Behaviour of a VR Agent Playing a Board Game

  • 22.5.2018, 13:00 - 14:00

Title: Non-Verbal Behaviour of a VR Agent Playing a Board Game
Author: Guðrún Inga Baldurdóttir
Date and Time: May 22nd , 2018 at 13:00
Location: M121

Supervisors: Dr. Hannes Högni Vilhjálmsson  and Dr. Stephan Schiffel, Reykjavík University

Abstract:

Imagine a scenario where you are playing a board-game, which can be considered a social action, within a virtual environment. In this environment, you are playing against another player. This player can converse with you, show emotions and give you the feeling of playing against a real person. This project builds on a final project done by three Bachelor students at Reykjavik University  called the Virtual General Game Playing Agent. The focus of this project is to increase the believability of the agent, by improving its non-verbal expression capabilities during a board-game. The approach taken in this thesis was to collect natural human data and use that to build a mechanism that generates similar behavior for the virtual agent. The existing agent 'decides' on moves when playing a board-game based on among other things his personality and mood as well as showing facial expressions such as smiling when doing well and frowning when doing poorly. The behavior of the existing agent was hard coded which means that he always showed the same behavior in same order. This project changes that in a way that when the agent is making a move, his move is not always of the same length and where he looks and how long varies between moves. The existing agent also had few options for complex behavior and to fix that a new 3D model was added to the project. A model that is more complex but gives us more possibilities of showing complex behavior.The model was added to the project without breaking the existing agent. To collect data, volunteers were recorded while they were playing a board-game. The recordings were then annotated in regards to behavior type such as eye gaze and actions such as hand movement in relation to the board-game. The data was then analysed, both how often and how long it occurred as well as analysing co-occurrence of annotation belonging to separate types (co-occurrence of behavior with action). Finally, a data-driven module was created and added to the new agent. The data module uses the analysed data when generating and selecting behavior. At the moment it only uses the module to generate and select gaze behavior when making a move. The results showed that out of 12 distinct behaviors evaluated the mean duration of each behavior of the new agent was closer to the real data in 7 of them and the existing agent in 5. With this work, we are closer to our scenario even though much work is still to do.



Vinsamlegast athugið að á viðburðum Háskólans í Reykjavík (HR) eru teknar ljósmyndir og myndbönd sem notuð eru í markaðsstarfi HR. Hægt er að nálgast frekari upplýsingar á ru.is eða með því að senda tölvupóst á netfangið: personuvernd@ru.is
//
Please note that at events hosted at Reykjavik University (RU), photographs and videos are taken which might be used for RU marketing purposes. Read more about this on our ru.is or send an e-mail: personuvernd@ru.is