InclusiveRender A metaverse Engine prototype to support Accessible Environments for people with ASD

Sammanfattning: The metaverse has seen increased usages in its capabilities as an educational tool by immersing users in virtual scenarios. This technology is inaccessible for user groups that require higher degrees of accessibility and personalization, as most metaverse implementations do not adjust to the user’s needs. One such group are individuals with Autism Spectrum Disorder (ASD), rendering them unable to use immersive learning to foster skills that enable independent living. To solve this issue, this thesis aims to employ design science to produce an artefact that answers the research question: ”How can machine learning-based adaptations to 3D-environments be integrated into existing virtual reality platforms in order to increase accessibility for users with ASD?”. By the use of literature studies along with personas supplied from a research project, four different user profiles were established to represent users with ASD. Requirements for the artefact to be produced were then established by exploring possible stakeholders affected by the artefact. Employing data generative techniques, a neural network was trained to predict how the virtual environment should be augmented given the user’s specific characteristics. This neural network was then integrated into a virtual environment set up via Unity, by use of participatory modelling. The resulting artefact was then evaluated against the established requirements, via an experiment that mirrors a typical morning routine meant to increase the person’s skills in independent living. The results of this evaluation found that the artefact could handle known user profiles with a high accuracy (87.7%). The artefact also proved effective in approximating what kind of aide should be presented for the unknown user profiles (8/10 cases). The use of modern development tooling also proved satisfactory in aiding developers to use the artefact to create accessible environments with ease. The artefact’s use of neural networks proved an effective way to model complex user groups, though further ethnographic studies and non-synthetic data is needed to validate this capability.