Practicing Stress Relief for the Everyday: Designing Social Simulation Using VR, AR, and LLMs

arXiv Preprint PDF

Abstract
Stress is an inevitable part of daily life, yet many struggle to manage it without readily available support. This paper explores the potential of social simulation using VR, AR, and LLMs to provide safe, virtual environments for stress relief, highlighting key considerations such as accessibility, risks of hyper-realism, and distrust in LLM recommendations.

My Contributions

• I developed AR and VR simulations of different stressful situations (e.g., public speaking, interpersonal conflict) in Unity for the Meta Quest 3.
• I integrated the OpenAI GPT-4o API into Unity to enable our avatars in the scene to simulate realistic human dialogue.
• I implemented structured output prompting, ensuring GPT responses adhered to a JSON scheme for easy parsing and integration of chats between the user and different personas.
• I also implemented Open-AI’s Whisper Speech-to-Text API that allowed users to use their voice to directly converse with virtual avatars in the social simulations.
• Moreover, I helped conduct semi-structured interviews with participants to gauge the different needs, risks, and design considerations of using simulations in everyday stressful scenarios.