01
VR System Development
Developed and refined two Unity-based VR training conditions: one using scripted dialogue and one integrating an AI-driven conversational agent.
XR ProjectsStandByMe
MSc. AIRE 1 Research Thesis · 2025
Laboratoire Interdisciplinaire des Sciences du Numérique (LISN), Centre national de la recherche scientifique (CNRS) / Université Paris-Saclay
Orsay, France
StandByMeEvaluating AI-Driven Virtual Training for Workplace Soft Skills
Virtual Reality · Generative AI · Soft Skills Training · Human–Computer Interaction
Supervision
Assoc. Prof. Huyen Nguyen
Associate Professor at Université Paris-Saclay · Member of the ARAI Team at LISN, CNRS UMR 9015
Project Overview
StandByMe is a mixed-method research project comparing scripted and AI-driven virtual agents for workplace soft-skills training. Ten participants completed both VR conditions while behavioural, physiological, and self-reported data were collected to evaluate engagement, presence, usability, and perceived realism.
Role
XR Researcher & Developer
Context
MSc. AIRE 1 Research Thesis · 2025
Areas
Virtual Reality · Generative AI · Soft Skills Training · Human–Computer Interaction
Tools & Technologies
Unity 2022.3 · C# · Python · Meta Quest 2 · GPT-4o · OpenAI API · ElevenLabs · Azure TTS · EmpaticaPlus
My Contribution
02
AI & Interaction Integration
Integrated speech-to-text, GPT-4o dialogue generation, and text-to-speech to create dynamic and context-sensitive interactions with virtual agents.
03
Experiment & Data Analysis
Supported participant testing and analysed questionnaire, interview, behavioural, and physiological data across both experimental conditions.
Experimental Design
Participants experienced two counterbalanced VR conditions in immersive workplace situations where they could observe, reflect, and practise responding to challenging social interactions.
Participants
Scripted VR
AI-Driven VR
Mixed-Method Evaluation
Questionnaires · Interviews · Behaviour · Physiology
Meta Quest 2 · EmpaticaPlus
Study at a Glance
0
Participants
Ten participants completed both virtual reality conditions in a within-subject experimental study.
00
VR Conditions
One scripted interaction and one GPT-4o-driven conversational interaction were compared.
Research Approach
Mixed-Method Evaluation
Questionnaires, semi-structured interviews, behavioural observations, and physiological signals were combined to evaluate the two experiences.
Data Collection
Multimodal Measures
Presence, engagement, agent perception, electrodermal activity, blood volume pulse, pulse rate, and skin temperature were collected.
Project OUTPUTS
01
More Natural Interaction
Participants generally described the AI-driven agent as more natural, responsive, and engaging than the scripted interaction.
02
Different Engagement Strengths
The scripted condition received higher usability ratings, while the AI condition showed slightly higher focused attention and comparable perceived reward.
03
Similar Physiological Responses
Pulse rate, electrodermal activity, skin temperature, and blood volume pulse showed no significant differences between the two VR conditions.
VR Conditions
Two virtual training experiences were developed in Unity to compare structured scripted interaction with dynamic AI-driven conversation in the same workplace-training context.
Scripted Condition
A structured VR scenario using predefined dialogue and fixed response options for workplace bystander-intervention training.
AI-Driven Condition
A conversational VR scenario using speech recognition, GPT-4o, and text-to-speech to support dynamic interaction with a virtual agent.
Project Snapshots
