User modelling for adaptive training in high performance driving

File Description SizeFormat 
Georgiou-T-2016-PhD-Thesis.pdfThesis36.08 MBAdobe PDFView/Open
Title: User modelling for adaptive training in high performance driving
Authors: Georgiou, Theodosis
Item Type: Thesis or dissertation
Abstract: User model creation is a fundamental component for the development of intelligent personalised systems. This thesis proposes an adaptive user modelling framework that uses a combination of unobtrusive task-related and physiological data with the aim of identifying strengths and weaknesses in user performance in the defined task. The research is focused on utilising the framework to provide personalised content adaptation in car racing games. Our system adopts concepts from the Trace Theory (TT) framework, and uses machine learning techniques to extract specific features from the user and the game. These metrics are then transformed and evaluated into higher level abstractions such as experience, exploration and physiological attention by utilising the educational theoretical frameworks of Flow and Zone Theory. The end result is to provide new game paths utilising the user’s model. We demonstrate that this procedural generation of user-tailored content drives the self-motivating behaviour of players to immerse and engage themselves in the game’s virtual world. Collection of data and feedback from multiple users (52) allowed us to associate the model’s outcomes to the user responses, as well as device multiple trial scenarios to verify their training and engagement. We have also evaluated the algorithms for the generation of new tracks for their suitability on the skill’s profile of 41 of our subjects and race track diversity among the evolved paths. We have also designed a method for predicting the states of the user-controlled system by combining information from both sources – vehicle and user – via Gaussian Processes (GPs). In the context of high speed car racing we showed that the forthcoming position and speed of the car can be predicted with high accuracy by our trained user models. This opens up future possibilities of generating better personalised tracks for individuals or even real-time share-control of the car to optimally assist the users in dangerous situations.
Content Version: Open Access
Issue Date: Sep-2016
Date Awarded: Dec-2016
Supervisor: Demiris, Yiannis
Sponsor/Funder: Engineering and Physical Sciences Research Council
Department: Electrical and Electronic Engineering
Publisher: Imperial College London
Qualification Level: Doctoral
Qualification Name: Doctor of Philosophy (PhD)
Appears in Collections:Electrical and Electronic Engineering PhD theses

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

Creative Commonsx