Date of Award

December 2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Hassan Reza

Abstract

This dissertation presents a real-time adaptive framework for mitigating cybersickness and enhancing user experience in virtual reality (VR) environments by dynamically adjusting key visual rendering parameters in response to users’ head movements. Conventional VR systems typically rely on fixed or manually configured settings that do not account for individual motion patterns or varying susceptibility to discomfort, leading to suboptimal usability and degraded immersion. In response, this work integrates software architecture tactics with data-driven machine learning models to proactively manage cybersickness while preserving rendering performance.

The proposed system employs a client–server architecture based on the Model-View-Controller (MVC) pattern. We conducted a within-subjects experiment using two conditions: a baseline VR application and an adaptive version with a dynamic module. Each experimental session used an Oculus Quest 2 headset. During these sessions, we recorded six-degree-of-freedom head-tracking data and processed it into kinematic features, including linear and angular velocity, acceleration, and jerk. We measured cybersickness with the Virtual Reality Sickness Questionnaire (VRSQ) and assessed cognitive workload with the SIM-TLX. We trained multiple regression models to predict VRSQ scores based on kinematic features and selected a Gradient Boosting Regressor for its strong predictive performance. This model, deployed on the server, generated cybersickness predictions every ten seconds. On the client, a rule-based adaptation module utilized these predictions to adjust the field of view (FoV) and the strength of fixed foveated rendering (FFR), aiming to reduce discomfort and manage GPU load.

The experimental results reinforce the system's effectiveness. The adaptive framework resulted in lower cybersickness scores and a reduced subjective workload compared to the baseline condition. It incurs only a moderate decrease in average frame rate. Together, these findings empirically validate that real-time, prediction-driven FoV and FFR adjustments can balance usability and performance as first-class quality attributes in VR system design. The dissertation thus contributes: (i) a validated methodology for predicting cybersickness from consumer-grade head-tracking data; (ii) an interpretable, architecture-aligned adaptive module for real-time mitigation; and (iii) an evidence-based assessment of the trade-offs between usability and performance in immersive VR experiences.

Download

Share

COinS