tba

Labyrinthos is a Serious Game set in the myth of the Minotaur, in which the player navigates a dynamically generated maze through Ancient Greek vocabulary. This thesis compares a desktop and a virtual reality (VR) version of the game to examine what changes when the game is played in a headset. Secondary-school pupils learning Ancient Greek played one version during a lesson (56 sessions: 35 desktop, 21 VR). Both versions logged play behaviour, and the VR version additionally logged movement metrics. Workload was measured with the Raw NASA-TLX and vocabulary with a pre- and post-test. Play behaviour differed only slightly between the versions. Workload leaned towards VR being more demanding, though not significantly, and VR pupils found the game less repetitive. As the game revises known vocabulary, the test left little room for a measurable gain. Overall, the headset changed how the game felt more than how it was played.

Robody is a teleoperated support robot whose acceptance depends not only on its technical
capabilities but also on attitudinal factors such as perceived usefulness, enjoyment, and
trust. This thesis investigates whether personally customizing Robody’s clothing affects
short-term acceptance compared with a preset appearance, and whether presenting the
same customized Robody in augmented reality (AR) rather than a conventional virtual
scene affects perceived social presence and acceptance. To study these questions, a clothing
customization system and two presentation modes were implemented in Unity, together
with a voice-driven conversational pipeline based on speech recognition, a large language
model, and speech synthesis. A within-subjects study with nine participants comprised three
measured interactions: preset non-AR, customized non-AR, and the same customized Robody
in AR, each followed by an Almere-based acceptance questionnaire. Perceived enjoyment
was significantly higher after customization, while perceived adaptivity, usefulness, and trust
showed no significant change. Social presence was significantly higher in AR than in the
virtual condition, while trust and the remaining secondary measures showed no significant
differences. These findings are interpreted cautiously because of the fixed condition order,
differing conversation topics, and the small, non-representative sample.

This thesis explores how a real-time language agent can interact with objects in a physical room and support human memory. A Unity application combines ARCore tracking with instance segmentation to detect objects, preserve them across camera views, and represent them as stable instances. These objects can be reviewed by a human and safely exposed to the language agent.

The system also controls which objects the agent can access during each stage of a study.
The system was evaluated with 60 participants in three conditions: arrows, a voice agent, and an embodied agent. Participants later recalled the objects, their order, and their positions. The agent conditions showed better order recall than the arrow condition, although the difference was not statistically significant after correction. Object recall followed the same general pattern, while spatial accuracy did not differ reliably. Participants often used the narrated story as a memory aid. The agent conditions were also perceived as more novel, but sometimes caused frustration through speech, waiting, or visual distraction.

The results suggest that language agents can help people remember the order of objects in a physical space. At the same time, effective support depends on clear narration and smooth interaction. The developed system provides a foundation for future studies of spatially aware agents and memory.